CTLF Corpus de textes linguistiques fondamentaux • IMPRIMER • RETOUR ÉCRAN
CTLF - Menu général - Textes

Bloomfield, Leonard. Language – T01


Chapter 1
The Study of Language

1. 1. Language plays a great part in our life. Perhaps because of
its familiarity, we rarely observe it, taking it rather for granted, as
we do breathing or walking. The effects of language are remarkable,
and include much of what distinguishes man from the animals, but
language has no place in our educational program or in the speculations
of our philosophers.

There are some circumstances, however, in which the conventionally
educated person discusses linguistic matters. Occasionally
he debates questions of “correctness” — whether it is “better,”
for instance, to say it's I or it's me. His discussion of such things
follows a fairly rigid pattern. If possible, he looks to the conventions
of writing for an answer — as, say, for the question whether
a t is to be pronounced in words like often or soften. Otherwise he
appeals to authority: one way of speaking, he believes, is inherently
right, the other inherently wrong, and certain learned
men, especially the authors of grammars and dictionaries, can tell
us which is which. Mostly, however, he neglects to consult these
authorities, and tries, instead, to settle the matter by a kind of
philosophical reasoning, which operates with terms such as “subject,”
“object,” “predicate,” and so on. This is the common-sense
way of dealing with linguistic matters. Like much else that masquerades
as common sense, it is in fact highly sophisticated, and
derives, at no great distance, from the speculations of ancient and
medieval philosophers.

It is only within the last century or so that language has been
studied in a scientific way, by careful and comprehensive observation;
the few exceptions will occupy us in a moment. Linguistics,
the study of language, is only in its beginnings. The knowledge it
has gained has not yet become part of our traditional education;
the “grammar” and other linguistic instruction in our schools
confines itself to handing on the traditional notions. Many people
have difficulty at the beginning of language study, not in grasping
the methods or results (which are simple enough), but in stripping
3off the preconceptions which are forced on us by our popular-scholastic

1. 2. The ancient Greeks had the gift of wondering at things
that other people take for granted. They speculated boldly and
persistently about the origin, history, and structure of language.
Our traditional lore about language is due largely to them.

Herodotus, writing in the fifth century b.c., tells us that King
Psammetichus of Egypt, in order to find out which was the oldest
nation of mankind (whatever this may mean), isolated two newborn
infants in a park; when they began to speak, they uttered
the word bekos, which turned out to be Phrygian for ‘bread.’

In his dialogue Cratylus, Plato (427-347 B.C.) discusses the
origin of words, and particularly the question whether the relation
between things and the words which name them is a natural and
necessary relation or merely the result of a human convention.
This dialogue gives us a first glimpse into a century-long controversy
between the Analogists, who believed that language was
natural and therefore at bottom regular and logical, and the
Anomalists, who denied these things and pointed out the irregularities
of linguistic structure.

The Analogists believed that the origin and the true meaning of
words could be traced in their shape; the investigation of this they
called etymology. We may illustrate their theory by English examples.
The word blackbird obviously consists of black and bird:
the species was named for its color, and, indeed, blackbirds are
birds and are black. In the same way, the Greeks would have concluded
that there was some deep-seated connection between a
gooseberry and a goose: it was the etymologist's task to find this
connection. The word mushroom would have presented a more
difficult problem. The components are often altered; thus, breakfast,
in spite of the difference in sound, is evidently the meal by
which we break our fast, and manly a shorter form of man-like.

In Greek, as in English, however, most words resist this kind of
analysis. Thus, early ends like manly, but the rest of the word is
obscure; woman resembles man, but what is the first syllable?
Then there is a residue of short, simple words that do not resemble
others — words such as man, boy, good, bad, eat, run. In such cases
the Greeks and their pupils, the Romans, resorted to guesswork.
For instance, they explained the Greek word lithos ‘stone’ as
derived from the phrase lian theein ‘to run too much,’ because this
4is what a stone does not do. A Latin example of this sort has become
proverbial: lucus a non lucendo ‘a grove (lucus) is so named
on account of its not being light (lucendo).’

These etymologies show us, at any rate, that the Greeks realized
that speech-forms change in the course of time. In the systematic
study of this change modern students have found the key to most
linguistic problems. The ancients never settled down to any
careful study of linguistic change.

The ancient Greeks studied no language but their own; they
took it for granted that the structure of their language embodied
the universal forms of human thought or, perhaps, of the cosmic
order. Accordingly, they made grammatical observations, but
confined these to one language and stated them in philosophical
form. They discovered the parts of speech of their language, its
syntactic constructions, such as, especially, that of subject and
predicate, and its chief inflectional categories: genders, numbers,
cases, persons, tenses, and modes. They defined these not in
terms of recognizable linguistic forms, but in abstract terms which
were to tell the meaning of the linguistic class. These teachings
appear most fully in the grammars of Dionysius Thrax (second
century B.C.) and of Apollonius Dyscolus (second century A.D.).

The Greeks made also some observations of detail, but this
phase of their work, unfortunately, had less effect upon posterity.
Their great epic poems, the Iliad and the Odyssey, which they
viewed somewhat as sacred scriptures, were composed in an
ancient and otherwise unknown kind of Greek. In order to understand
these texts and to make correct copies, one had to study
their language. Most famous in this work was Aristarchus (about
216-144 B.C.). Other works of Greek literature were composed in
conventionalized forms of various regional dialects: the Greeks
had the opportunity of comparing several divergent forms of
their language. When the language of the great Athenian writers
of the fourth century had become antiquated, it was made a
special subject of study, since it represented the ideal form of
written discourse. All this work demanded careful observation
of details. Some of the later grammarians, notably Herodian, the
son of Apollonius Dyscolus, assembled valuable information on
such topics as the inflection and accent of ancient Greek.

1. 3. The Greek generalizations about language were not improved
upon until the eighteenth century, when scholars ceased
5to view language as a direct gift of God, and put forth various
theories as to its origin. Language was an invention of ancient
heroes, or else the product of a mystical Spirit of the Folk. It
began in man's attempts to imitate noises (the “bow-wow”
theory), or in his natural sound-producing responses (the “ding-dong”
theory), or in violent outcries and exclamations (the “pooh-pooh”

In the etymological explanation of speech-forms there was no
improvement. Voltaire is reported to have said that etymology
is a science in which the vowels count for nothing and the consonants
for very little.

The Romans constructed Latin grammars on the Greek model;
the most famous of these, the work of Donatus (fourth century
A.D.) and of Priscian (sixth century A.D.), remained in use as
text-books through the Middle Ages. In the Middle Ages, when
Latin was changing from its ancient shape into the forms which
we know today as the Romance languages (French, Italian,
Spanish, and so on), the convention remained of writing, as well
as one could, in the ancient classical form of Latin. The medieval
scholar, accordingly, in both the Latin countries and others,
studied only classical Latin. The scholastic philosophers discovered
some features of Latin grammar, such as the distinction
between nouns and adjectives and the differences between
concord, government, and apposition. They contributed much
less than the ancients, who had, at any rate, a first-hand knowledge
of the languages they studied. The medieval scholar saw in
classical Latin the logically normal form of human speech. In
more modern times this doctrine led to the writing of general grammars,
which were to demonstrate that the structure of various
languages, and especially of Latin, embodies universally valid
canons of logic. The most famous of these treatises is the Grammaire
générale et raisonnée
of the Convent of Port-Royal, which
appeared in 1660. This doctrine persisted into the nineteenth
century; it appears, for instance, in the classical scholar, Gottfried
Hermann's work De emendanda ratione Graecae grammaticae
(1801). It is still embodied in our school tradition, which seeks
to apply logical standards to language. Philosophers, to this
day, sometimes look for truths about the universe in what are
really nothing but formal features of one or another language.

An unfortunate outgrowth of the general-grammar idea was
6the belief that the grammarian or lexicographer, fortified by his
powers of reasoning, can ascertain the logical basis of language
and prescribe how people ought to speak. In the eighteenth
century, the spread of education led many dialect-speakers to
learn the upper-class forms of speech. This gave the authoritarians
their chance: they wrote normative grammars, in which they
often ignored actual usage in favor of speculative notions. Both
the belief in “authority” and some of the fanciful rules (as, for
instance, about the use of shall and will) still prevail in our

For the medieval scholar, language meant classical Latin, as
it appears in books; we find few traces of interest in any other form
of speech. The horizon widened at the time of the Renaissance.
At the end of the Middle Ages, the study of Greek came back
into fashion; soon afterward, Hebrew and Arabic were added.
What was more important, some scholars in various countries
began to take an interest in the language of their own time.

The era of exploration brought a superficial knowledge of many
languages. Travelers brought back vocabularies, and missionaries
translated religious books into the tongues of newly-discovered
countries. Some even compiled grammars and dictionaries of exotic
languages. Spanish priests began this work as early as in the
sixteenth century; to them we owe a number of treatises on American
and Philippine languages. These works can be used only
with caution, for the authors, untrained in the recognition of
foreign speech-sounds, could make no accurate record, and, knowing
only the terminology of Latin grammar, distorted their exposition
by fitting it into this frame. Down to our own time, persons
without linguistic training have produced work of this sort;
aside from the waste of labor, much information has in this way
been lost.

The increase of commerce and travel led also to the compilation
of grammars and dictionaries for languages closer at hand.
The linguistic horizon at the end of the eighteenth century can be
surveyed in the glossary of 285 words in two hundred languages
of Europe and Asia which P. S. Pallas (1741-1811) edited at the
behest of Empress Catharine of Russia in 1786. A second edition
of this, in 1791, added eighty more languages, including some
African and American. In the years 1806 to 1817 there appeared
a four-volume treatise under the title Mithridates, by J. C. Adelung
7and J. S. Vater, which contained the Lord's Prayer in nearly five
hundred languages.

The Renaissance turned the interest of a few scholars to the
older records of their own languages. Franciscus Junius (15891677)
accomplished an enormous amount of work in the study of
the ancient documents of English and of the closely related languages,
Frisian, Dutch, German, Scandinavian, and Gothic. This
last — a language no longer spoken today — Junius knew from
the famous Silver Codex, then recently discovered, a manuscript
of the sixth century A.D. containing fragments of a Gospel translation;
Junius published its text, together with that of the Anglo-Saxon
Gospels. George Hickes (1642-1715) continued this work,
publishing a Gothic and Anglo-Saxon grammar and a Thesaurus
of miscellaneous information about the older stages of English
and the sister tongues.

1. 4. The development so far outlined shows us what eighteenth-century
scholars knew about language. They stated the grammatical
features of language in philosophical terms and took no
account of the structural difference between languages, but obscured
it by forcing their descriptions into the scheme of Latin
grammar. They had not observed the sounds of speech, and confused
them with the written symbols of the alphabet. This failure
to distinguish between actual speech and the use of writing distorted
also their notions about the history of language. They saw
that in medieval and modern times highly cultivated persons
wrote (and even spoke) good Latin, while less educated or careless
scribes made many mistakes: failing to see that this Latin-writing
was an artificial and academic exercise, they concluded that languages
are preserved by the usage of educated and careful people
and changed by the corruptions of the vulgar. In the case of
modern languages like English, they believed, accordingly, that
the speech-forms of books and of upper-class conversation represented
an older and purer level, from which the “vulgarisms” of
the common people had branched off as “corruptions” by a process
of “linguistic decay.” The grammarians felt free, therefore, to
prescribe fanciful rules which they derived from considerations of

These misconceptions prevented scholars from making use of
the data that were at hand: the modern languages and dialects,
the records of ancient languages, the reports about exotic languages,
8and, above all, the documents which show us successive
stages of one and the same language, as for instance of Anglo-Saxon
(Old English) and modern English, or of Latin and the modern
Romance languages. One knew that some languages resembled
each other, but the doctrine of linguistic decay discouraged systematic
study of this relation, since the changes which led, say,
from Latin to modern French, were viewed as haphazard corruptions.

The illusion that Latin had lived on, unchanged, beside the
Romance languages, led scholars to derive contemporary languages
one from the other. Mostly they took Hebrew to be the language
from which all others had sprung, but some thought otherwise,
as, for example, Goropius Becanus of Antwerp, who patriotically
derived all languages from Dutch.

It was plain that the more familiar languages of Europe fell
into three groups by virtue of close resemblances within each
group, resemblances such as appear in the following words:

tableau germanic group | English | Dutch | German | Danish | Swedish | romance group | French | Italian | Spanish | slavic group | Russian | Polish | Bohemian | Serbian | ‘hand’ | ‘foot’ | ‘winter’9

tableau germanic group | English | Dutch | German | Danish | Swedish | romance group | French | Italian | Spanish | slavic group | Russian | Polish | Bohemian | Serbian | ‘drink’

There was apparent also a less striking resemblance between
these groups; this wider resemblance extended to some other languages,
such as, notably, Greek:

‘mother’: Greek mētēr, Latin māter (with its modern forms in
the Romance languages), Russian mat' (genitive case materi
with similar forms in the other Slavic languages), English mother
(with similar forms in the other Germanic languages);

‘two’: Greek duo, Latin duo, Russian dva, English two;

‘three’: Greek treis, Latin trēs, Russian tri, English three;

‘is’: Greek esti, Latin est, Russian jest', English is (German ist).

1. 5. Outside the tradition of Europe, several nations had developed
linguistic doctrines, chiefly on an antiquarian basis. The
Arabs had worked out a grammar of the classical form of their
language, as it appears in the Koran; on the model of this, the
Jews in Mohammedan countries constructed a Hebrew grammar.
At the Renaissance, European scholars became acquainted with
this tradition; the term root, for instance, as a designation for the
central part of a word, comes from Hebrew grammar. In the Far
East, the Chinese had gained a great deal of antiquarian linguistic
knowledge, especially in the way of lexicography. A Japanese
grammar seems to have grown up independently.

It was in India, however, that there arose a body of knowledge
which was destined to revolutionize European ideas about language.
The Brahmin religion guarded, as sacred texts, some very
ancient collections of hymns; the oldest of these collections, the
Rig-Veda, dates in part, at a conservative estimate, from about
1200 B.C. As the language of these texts grew antiquated, the
proper way of pronouncing them, and their correct interpretation,
became the task of a special class of learned men. The antiquarian
interest in language which arose in this way, was carried over
into a more practical sphere. Among the Hindus, as among us,
different classes of society differed in speech. Apparently there
10were forces at work which led upper-class speakers to adopt lower-class
forms of speech. We find the Hindu grammarians extending
their interest from the Scriptures to the upper-caste language,
and making rules and lists of forms descriptive of the correct type
of speech, which they called Sanskrit. In time they worked out
a systematic arrangement of grammar and lexicon. Generations
of such labor must have preceded the writing of the oldest treatise
that has come down to us, the grammar of Pāṇini. This grammar,
which dates from somewhere round 350 to 250 B.C., is one of the
greatest monuments of human intelligence. It describes, with the
minutest detail, every inflection, derivation, and composition, and
every syntactic usage of its author's speech. No other language,
to this day, has been so perfectly described. It may have been due,
in part, to this excellent codification that Sanskrit became, in
time, the official and literary language of all of Brahmin India.
Long after it had ceased to be spoken as anyone's native language,
it remained (as classical Latin remained in Europe) the artificial
medium for all writing on learned or religious topics.

Some knowledge of Sanskrit and of the Hindu grammar had
reached Europe, through missionaries, in the sixteenth and seventeenth
centuries. In the eighteenth century, Englishmen in India
transmitted more exact reports; round the beginning of the nineteenth
century, the knowledge of Sanskrit became part of the
equipment of European scholars.

1. 6. The Indian grammar presented to European eyes, for the
first time, a complete and accurate description of a language,
based not upon theory but upon observation. Moreover, the discovery
of Sanskrit disclosed the possibility of a comparative study
of languages.

To begin with, the concept of related languages was strikingly
confirmed by the existence, in far-off India, of a sister of the
familiar languages of Europe; witness, for example, the Sanskrit
equivalents of the words above cited:

mātā ‘mother,’ accusative case mātaram;
dvāu ‘two;’
trayah ‘three;’
asti ‘he is.’

Even more important was the insight into linguistic structure
which one got from the accurate and systematic Hindu grammar.
Until now, one had been able to see only vague and fluid similarities,
11for the current grammars, built on the Greek model, did not
clearly set off the features of each language. The Hindu grammar
taught Europeans to analyze speech-forms; when one compared
the constituent parts, the resemblances, which hitherto had been
vaguely recognized, could be set forth with certainty and precision.

The old confused notions of linguistic relationship lived on for
a brief time in the opinion that the European languages were
derived from Sanskrit, but this opinion soon gave way to the obviously
correct explanation, namely, that Sanskrit, Latin, Greek,
and so on, were divergent later forms of some one prehistoric language.
This explanation seems to have been first stated by Sir
William Jones (1746-1794), the first great European Sanskrit
scholar, in an address delivered in 1786: Sanskrit bears a resemblance
to Greek and Latin which is too close to be due to chance,
but shows, rather, that all three “have sprung from some common
source which, perhaps, no longer exists,” and Gothic (that is,
Germanic) and Celtic probably had the same origin.

In order to work out the comparison of these languages, one
needed, of course, descriptive data for each one of them. The
prospect of comparison, however, with all that it revealed about
ancient speech-forms and tribal migrations and the origin of
peoples and customs, proved so alluring that no one undertook
the humdrum task of analyzing the other languages on the model
of Sanskrit. European scholars had a sound knowledge of Latin
and Greek; most of them spoke some Germanic language as their
mother-tongue. Confronting a precise statement of Sanskrit
grammar or a carefully analyzed lexical form, they could usually
recall a similar feature from some of the more familiar languages.
In reality, of course, this was a makeshift; often enough the comparer
had to make a preliminary investigation to establish the
facts, and sometimes he went astray for lack of methodically
arranged data. If European scholars had possessed descriptions
of the sister languages comparable to the Hindus' description of
Sanskrit, the comparative study of the Indo-European languages
(as they are now called) would have progressed far more speedily
and accurately. Yet, in spite of poor equipment, and thanks to
the energy of its workers, the historical and comparative study of
the Indo-European languages became one of the principal enterprises,
and one of the most successful, of European science in the
nineteenth century.12

The languages of Persia (the so-called Iranian languages) so
closely resembled Sanskrit that their kinship was certain from the
start. A similar relation, though less close, was found to exist
between the Baltic languages (Lithuanian, Lettish, and Old
Prussian) and the Slavic. Jones' surmise that the Germanic languages
were related to Latin, Greek, and Sanskrit, at once proved
true, as did later his surmise about Celtic (Irish, Welsh, Cornish,
Breton, and the ancient language of Gaul). Later, Armenian and
Albanese, and a few ancient languages known to us only from
scant written records, proved also to belong to the Indo-European

Although there was some dispute as to details, the general presuppositions
of historical and comparative language-study soon
became clear. Languages change in the course of time. Apparent
exceptions, such as the medieval and modern use of Latin (or, in
India, of Sanskrit), amount only to this, that by long schooling
people can be trained to imitate the language of ancient writings.
This antiquarian feat is utterly different from the normal transmission
of speech from parents to children. All writing, in fact,
is a relatively recent invention, and has remained, almost to our
day, the property of only a chosen few: the effect of writing upon
the forms and the development of actual speech is very slight.

If a language is spoken over a large area, or thanks to migration,
in several separate areas, then it will change differently in different
places, and the result will be a set of related languages, like Italian,
French, Spanish, Portuguese, Roumanian, and the other Romance
dialects. We infer that other groups of related languages, such
as the Germanic (or the Slavic or the Celtic), which show a similar
resemblance, have arisen in the same way; it is only an accident
of history that for these groups we have no written records of the
earlier state of the language, as it was spoken before the differentiation
set in. To these unrecorded parent languages we give
names like Primitive Germanic (Primitive Slavic, Primitive Celtic,
and so on). 11 In the same way, finding that all these languages
and groups (Sanskrit, Iranian, Armenian, Greek, Albanese, Latin,
Celtic, Germanic, Baltic, Slavic) resemble each other beyond the
possibility of mere chance, we call them the Indo-European family
13of languages
, and conclude, with Jones, that they are divergent
forms of a single prehistoric language, to which we give the name
Primitive Indo-European.

The method of comparison, too, was clear from the start. In
general, any feature that is common to all or to several of the
related languages, must have been present in their common antecedent
stage, in the “parent language.” Thus, from the above
cited forms of the word for ‘mother,’ it is clear that in Primitive
Indo-European this word must have begun with the sound which
we indicate in writing by means of the letter m. Where the related
languages do not agree, some or all of them must have made some
change. Thus, it is clear that the second consonant in the word
for ‘mother’ was in Primitive Indo-European a t-sound, and that
the th-sound in English (as well as the earlier d-sound in the Old
English form, mōdor) must be due to change.

1. 7. The beginning of a systematic comparison of the Indo-European
languages was a treatise on the inflectional endings of
verbs in Sanskrit, Greek, Latin, Persian, and Germanic, published
in 1816 by Franz Bopp (1791-1867). In 1818 Rasmus Kristian
Rask (1787-1832) showed that the words of the Germanic languages
bear a regular formal relation in matters of sound, to the
words of the other Indo-European languages. For instance,
where the others have p, the Germanic languages have f, as in
father: Latin pater, foot: Latin pēs, five: Greek pente, few: Latin
paucī. In 1819 Jakob Grimm (1787-1863) published the first
volume of his Deutsche Grammatik, which was not, as the title nowadays
would indicate, a German grammar, but a comparative
grammar of the Germanic languages (Gothic, Scandinavian,
English, Frisian, Dutch, and German). In the second edition,
in 1822, of this volume, Grimm presented a systematic exposition
of the correspondences of consonants between Germanic and
the other Indo-European languages; since then, these correspondences
have been known to English-speaking scholars as Grimm's
. These correspondences are a matter of historical detail,
but their significance was overwhelming, since they showed
that human action, in the mass, is not altogether haphazard, but
may proceed with regularity even in so unimportant a matter as
the manner of pronouncing the individual sounds within the flow
of speech. Grimm's comparison of the Germanic languages remains
to this day unrivaled; three more volumes appeared in 1826,
141831, and 1837; a fifth volume, which was to complete the syntax,
never appeared.

In 1833 Bopp began the publication of a comprehensive treatise,
a comparative grammar of the Indo-European languages. In
the years 1833 to 1836 there appeared the first edition of the
Etymological Investigations of August Friedrich Pott (1802-1887).
The term etymology, here as in all modern discussions, has taken
on a precise meaning: the etymology of a speech-form is simply its
history, and is obtained by finding the older forms in the same
language and the forms in related languages which are divergent
variants of the same parent form. Thus, to state the etymology
of the English word mother is to say that this form is the modern
version of the ninth-century Old English mōdor; that this is related
to Old Norse mōðer, Old Frisian mōder, Old Saxon mōdar,
Old High German muoter (these are the forms in our oldest records
of the respective languages), in the sense that all these are
divergent variants of a single Primitive Germanic word, which
we symbolize as *mōder; and that these Germanic forms are in
turn related to (“cognate with”) Sanskrit mātā, Avestan (Old
Iranian) mātā, Old Armenian mair, ancient Greek mētēr, Albanese
motrɛ (which, however, means ‘sister’), Latin māter, Old Irish
māthir, Lithuanian motē (which means ‘wife’), Old Bulgarian
(Slavic) mati, and with the other corresponding forms in each of
the groups of languages here illustrated, in the sense that all
these are divergent later forms of a single Primitive Indo-European
word, which we symbolize as *māter. As this example shows, etymologies,
in the modern sense, do not necessarily show us an older,
more transparent meaning of words. Our modern etymologies in
the Indo-European languages are due largely to the researches
of Pott.

During the following decades progress was so rapid that both
smaller treatises and the great handbooks rapidly became antiquated.
Of the latter, Bopp's, in spite of new editions, was superseded
in 1861 by the Compendium of the Comparative Grammar of
the Indo-European Languages
of August Schleicher (1823-1868).
In 1886 Karl Brugmann (1849-1919) and Berthold Delbruck
(1842-1922) began the publication of their Outline of the Comparative
Grammar of the Indo-European Languages
; the standard work
of reference today is the second edition of this, which appeared
from 1897 to 1916.15

As the work went on, other, more detailed treatises were devoted
to the separate branches of the Indo-European family, in
the manner of Grimm's great treatise on Germanic. Friedrich
Diez (1794-1876) began the serious study of the Romance languages
in his Grammar of the Romance Languages (1836-1844);
Johann Kaspar Zeuss (1806-1856) opened the field of the Celtic
languages in his Grammatica Celtica (1853); Franz von Miklosich
(1813-1891) wrote a Comparative Grammar of the Slavic Languages

1. 8. These studies could not fail to throw light upon many an
aspect of history and archaeology, but their immediate interest
lay in what they told about human speech. Although the various
Indo-European languages had a common origin, their later careers
were independent: the student had now a vast collection of details
concerning the changes in human speech, which enabled him
to generalize on the manner of this change.

To draw the conclusions as to the way in which languages change,
was to replace the speculation of earlier times by the results of
scientific induction. William Dwight Whitney (1827-1894), an
American scholar, wrote Language and the Study of Language
(1867) and The Life and Growth of Language (1874). These books
were translated into several European languages; today they
seem incomplete, but scarcely antiquated, and still serve as an
excellent introduction to language study. In 1880 there appeared
the Principles of Linguistic History by Hermann Paul (1846-1921),
which, in its successive editions (the fifth appeared in 1920), became
the standard work on the methods of historical linguistics.

Paul's book of Principles illustrates, with a wealth of examples,
the process of linguistic change which had been revealed by Indo-European
studies. Not so well written as Whitney's, but more
detailed and methodical, this book exercised a great influence on
linguistic studies; students of a more recent generation are neglecting
it, to their disadvantage. Aside from its very dry style,
Paul's Principles suffers from faults that seem obvious today,
because they are significant of the limitations of nineteenth-century

One of these faults is Paul's neglect of descriptive language
study. He admitted that descriptions of languages were necessary,
but confined his actual discussion to matters of linguistic
change. This shortcoming he shares with his epoch. We can study
16linguistic change only by comparing related languages or different
historical stages of the same language. For instance, by
noting the similarities and differences of English, Frisian, Dutch,
German, Scandinavian, and Gothic, we can get a notion of the
older language (“Primitive Germanic”) from which they have
differentiated in the course of time, and we can then study the
changes which have occurred in each of these later languages.
Or else, by comparing our records of Old English (say, in the writings
of King Alfred) with modern English, we can see how English
has changed in the last thousand years. Evidently our power
of making this comparison depends upon our knowledge of the
things to be compared. For example, our knowledge about the
compounding of words (as in blackbird or footsore) in the several
Germanic languages is decidedly incomplete; therefore we cannot
go very far with a comparative study of this matter, which
would tell us how words were compounded in Primitive Germanic,
and how these habits have changed in the subsequent history of
each Germanic language. The historical language students of
the nineteenth century suffered under these limitations, but they
seem not to have grasped the nature of the difficulty.

The other great weakness of Paul's Principles is his insistence
upon “psychological” interpretation. He accompanies his statements
about language with a paraphrase in terms of mental
processes which the speakers are supposed to have undergone. The
only evidence for these mental processes is the linguistic process;
they add nothing to the discussion, but only obscure it. In Paul's
book and largely to the present day, linguistics betrays its descent
from the philosophical speculations of the ancient Greeks. Paul
and most of his contemporaries dealt only with Indo-European
languages and, what with their neglect of descriptive problems,
refused to work with languages whose history was unknown. This
limitation cut them off from a knowledge of foreign types of
grammatical structure, which would have opened their eyes to the
fact that even the fundamental features of Indo-European grammar,
such as, especially, the part-of-speech system, are by no
means universal in human speech. Believing these features to be
universal, they resorted, whenever they dealt with fundamentals,
to philosophical and psychological pseudo-explanations.

1. 9. Alongside the great stream of historical research, there ran,
however, a small but accelerating current of general linguistic
17study. The Hindu grammar of Sanskrit was never quite forgotten;
while many pupils used its results without knowing of its
existence, the masters, who knew the antecedents of their science,
appreciated its value. For the less-known Indo-European languages
descriptive studies could not be avoided. It is surely no
accident that the best of these, in the field of the Slavic and
Baltic languages, were furnished by August Leskien (1840-1916),
a scholar who took a leading part in laying the foundations of
historical methods of research.

For the most part, however, descriptive studies did not merge
with the main stream of historical work. Some students were
attracted by the structural peculiarities of languages outside the
Indo-European group, even though the history of these languages
was unknown. Other students examined a variety of languages in
order to get a philosophical survey of human speech; in fact, much
of the older descriptive work is almost unintelligible today because
it is pervaded by philosophical notions that are no longer familiar
to us.

The first great book on general linguistics was a treatise on the
varieties of human speech by Wilhelm von Humboldt (1767-1835),
which appeared in 1836. H. Steinthal (1823-1899), beside
more general writings on the fundamentals of language, published
in 1861 a treatise on the principal types of language structure.
G. von der Gabelentz' (1840-1893) work on the science of language
(1891) is much less philosophical. This direction of study culminated
in a great work on language by the philosopher and
psychologist, Wilhelm Wundt (1832-1920), which appeared in
1900 as the first part of a treatise on social psychology. Wundt
based his psychology of speech upon any and all accessible descriptions
of languages. It is interesting today to read the Indo-Europeanist
Delbrück's critique and Wundt's rejoinder, both of
which appeared in the following year. Delbrück objects to Wundt's
use of languages whose history is unknown; for him the only aspect
of language worth studying is its change in the course of time.
Wundt, on the other hand, insists upon the importance of psychological
interpretation in terms of his system, while Delbrück says
that it does not matter what particular system of psychology a
linguist may choose.

Meanwhile some students saw more and more clearly the natural
relation between descriptive and historical studies. Otto Böhtlingk
18(1815-1904), who made the modern European edition of Panini,
applied the descriptive technique to a language of totally different
structure, the Yakut of Asiatic Russia (1851). Friedrich Müller
(1834-1898) published an outline of linguistic science (1876-1888)
which contained brief sketches of the languages of the world,
regardless of whether a historical treatment was possible. Franz
Nikolaus Finck (1867-1910), both in a theoretical essay (1905) and
in a little volume (1910) in which he analyzed descriptively eight
unrelated languages, insisted upon descriptive study as a basis for
both historical research and philosophical generalization. Ferdinand
de Saussure (1857-1913) had for years expounded this matter
in his university lectures; after his death, they were published in
book form (1915).

Most convincing in this respect was the historical treatment of
language families other than the Indo-European. On the one hand,
the need of descriptive data as a prerequisite for comparative work
was here self-evident; on the other hand, the results showed that
the processes of linguistic change were the same in all languages,
regardless of their grammatical structure. The comparative study
of the Finno-Ugrian languages (Finnish, Lappish, Hungarian, and
their kin) began as early as 1799, and has been greatly elaborated.
The second volume of Humboldt's great treatise founded the
comparative grammar of the Malayo-Polynesian language family.
Today we have comparative studies of other families, such as the
Semitic family and the Bantu family in Africa. Students of
American languages could indulge in no self-deception as to the
need of descriptive data: north of Mexico alone there are dozens
of totally unrelated groups of languages, presenting the most varied
types of structure. In the stress of recording utterly strange forms
of speech one soon learned that philosophical prepossessions were
only a hindrance.

The merging of these two streams of study, the historical-comparative
and the philosophical-descriptive, has made clear
some principles that were not apparent to the great Indo-Europeanists
of the nineteenth century, as represented, say, by Hermann
Paul. All historical study of language is based upon the comparison
of two or more sets of descriptive data. It can be only as accurate
and only as complete as these data permit it to be. In order to
describe a language one needs no historical knowledge whatever;
in fact, the observer who allows such knowledge to affect his
19description, is bound to distort his data. Our descriptions must
be unprejudiced, if they are to give a sound basis for comparative

The only useful generalizations about language are inductive
generalizations. Features which we think ought to be universal
may be absent from the very next language that becomes accessible.
Some features, such as, for instance, the distinction of verb-like and
noun-like words as separate parts of speech, are common to many
languages, but lacking in others. The fact that some features are,
at any rate, widespread, is worthy of notice and calls for an explanation;
when we have adequate data about many languages, we
shall have to return to the problem of general grammar and to
explain these similarities and divergences, but this study, when it
comes, will be not speculative but inductive.

As to change in language, we have enough data to show that the
general processes of change are the same in all languages and tend
in the same direction. Even very specific types of change occur in
much the same way, but independently, in the most diverse
languages. These things, too, will some day, when our knowledge is
wider, lend themselves to a systematic survey and to fruitful

Chapter 2
The Use of Language

2. 1. The most difficult step in the study of language is the
first step. Again and again, scholarship has approached the study
of language without actually entering upon it. Linguistic science
arose from relatively practical preoccupations, such as the use of
writing, the study of literature and especially of older records, and
the prescription of elegant speech, but people can spend any
amount of time on these things without actually entering upon
linguistic study. As the individual student is likely to repeat the
delays of history, we may do well to speak of these matters, so as to
distinguish them from the subject of our study.

Writing is not language, but merely a way of recording language
by means of visible marks. In some countries, such as China,
Egypt, and Mesopotamia, writing was practised thousands of
years ago, but to most of the languages that are spoken today it has
been applied either in relatively recent times or not at all. Moreover,
until the days of printing, literacy was confined to a very
few people. All languages were spoken through nearly all of their
history by people who did not read or write; the languages of such
peoples are just as stable, regular, and rich as the languages of
literate nations. A language is the same no matter what system of
writing may be used to record it, just as a person is the same no
matter how you take his picture. The Japanese have three systems
of writing and are developing a fourth. When the Turks, in 1928,
adopted the Latin alphabet in place of the Arabic, they went on
talking in just the same way as before. In order to study writing,
we must know something about language, but the reverse is not
true. To be sure, we get our information about the speech of past
times largely from written records — and for this reason we shall,
in another connection, study the history of writing — but we find
this to be a handicap. We have to use great care in interpreting the
written symbols into terms of actual speech; often we fail in this,
and always we should prefer to have the audible word.

Literature, whether presented in spoken form or, as is now our
21custom, in writing, consists of beautiful or otherwise notable utterances.
The student of literature observes the utterances of certain
persons (say, of a Shakspere) and concerns himself with the
content and with the unusual features of form. The interest of the
philologist is even broader, for he is concerned with the cultural significance
and background of what he reads. The linguist, on the
other hand, studies the language of all persons alike; the individual
features in which the language of a great writer differs from the
ordinary speech of his time and place, interest the linguist no more
than do the individual features of any other person's speech, and
much less than do the features that are common to all speakers.

The discrimination of elegant or “correct” speech is a by-product
of certain social conditions. The linguist has to observe it as he
observes other linguistic phenomena. The fact that speakers
label a speech-form as “good” or “correct,” or else as “bad” or
“incorrect,” is merely a part of the linguist's data concerning this
speech-form. Needless to say, it does not permit him to ignore
part of his material or to falsify his records: he observes all speech-forms
impartially. It is part of his task to find out under what
circumstances the speakers label a form in one way or the other,
and, in the case of each particular form, why they label it as they
do: why, for example, many people say that ain't is “bad” and
am not is “good.” This is only one of the problems of linguistics,
and since it is not a fundamental one, it can be attacked only
after many other things are known. Strangely enough, people
without linguistic training devote a great deal of effort to futile
discussions of this topic without progressing to the study of language,
which alone could give them the key.

A student of writing, of literature or philology, or of correct
speech, if he were persistent and methodical enough, might realize,
after some waste of effort, that he had better first study language
and then return to these problems. We can save ourselves this
detour by turning at once to the observation of normal speech.
We begin by observing an act of speech-utterance under very
simple circumstances.

2. 2. Suppose that Jack and Jill are walking down a lane. Jill
is hungry. She sees an apple in a tree. She makes a noise with her
larynx, tongue, and lips. Jack vaults the fence, climbs the tree,
takes the apple, brings it to Jill, and places it in her hand. Jill
eats the apple.22

This succession of events could be studied in many ways, but
we, who are studying language, will naturally distinguish between
the act of speech and the other occurrences, which we shall call
practical events. Viewed in this way, the incident consists of three
parts, in order of time:

A. Practical events preceding the act of speech.

B. Speech.

C. Practical events following the act of speech.

We shall examine first the practical events. A and C. The events
in A concern mainly the speaker, Jill. She was hungry; that is,
some of her muscles were contracting, and some fluids were being
secreted, especially in her stomach. Perhaps she was also thirsty:
her tongue and throat were dry. The light-waves reflected from
the red apple struck her eyes. She saw Jack by her side. Her
past dealings with Jack should now enter into the picture; let
us suppose that they consisted in some ordinary relation, like
that of brother and sister or that of husband and wife. All these
events, which precede Jill's speech and concern her, we call the
speaker's stimulus.

We turn now to C, the practical events which came after Jill's
speech. These concern mainly the hearer, Jack, and consist of
his fetching the apple and giving it to Jill. The practical events
which follow the speech and concern the hearer, we call the hearer's
. The events which follow the speech concern also
Jill, and this in a very important way: she gets the apple into her
grasp and eats it

It is evident at once that our whole story depends upon some
of the more remote conditions connected with A and C. Not every
Jack and Jill would behave like these. If Jill were bashful or if
she had had bad experiences of Jack, she might be hungry and
see the apple and still say nothing; if Jack were ill disposed toward
her, he might not fetch her the apple, even though she asked
for it. The occurrence of a speech (and, as we shall see, the wording
of it) and the whole course of practical events before and after
it, depend upon the entire life-history of the speaker and of the
hearer. We shall assume in the present case, that all these predisposing
were such as to produce the story as we have
told it. Supposing this, we want to know what part the speech-utterance
(B) played in this story.

If Jill had been alone, she might have been just as hungry and
23thirsty and might have seen the same apple. If she had sufficient
strength and skill to get over the fence and climb the tree, she
could get hold of the apple and eat it; if not, she would have to
stay hungry. The lone Jill is in much the same position as the
speechless animal. If the animal is hungry and sees or smells
food, it moves toward the food; whether the animal succeeds
in getting the food, depends upon its strength and skill. The state
of hunger and the sight or smell of the food are the stimulus (which
we symbolize by S) and the movements toward the food are the
reaction (which we symbolize by R). The lone Jill and the speechless
animal act in only one way, namely

image S → R

If this works, they get the food; if it does not work — if they are
not strong or skilful enough to get the food by the actions R — they
must stay hungry.

Of course, it is important for Jill's welfare that she get the apple.
In most instances it is not a matter of life and death, though sometimes
it is; in the long run, however, the Jill (or the animal) that
gets the food has far better chances of surviving and populating
the earth. Therefore, any arrangement which adds to Jill's chances
of getting the apple, is enormously valuable for her. The speaking
Jill in our story availed herself of just such an arrangement.
She had, to begin with, the same chance of getting the apple as
had the lone Jill or the speechless animal. In addition to this,
however, the speaking Jill had a further chance which the others
did not share. Instead of struggling with the fence and the tree,
she made a few small movements in her throat and mouth, which
produced a little noise. At once, Jack began to make the reactions
for her; he performed actions that were beyond Jill's strength,
and in the end Jill got the apple. Language enables one person to
make a reaction (R) when another person has the stimulus (S)

In the ideal case, within a group of people who speak to each
other, each person has at his disposal the strength and skill of
every person in the group. The more these persons differ as to
special skills, the wider a range of power does each one person
control. Only one person needs to be a good climber, since he
can get fruit for all the rest; only one needs to be a good fisherman,
since he can supply the others with fish. The division of labor,
and, with it, the whole working of human society, is due to language

2. 3. We have yet to examine B, the speech-event in our story.
24This, of course, is the part of the story with which we, as students
of language, are chiefly concerned. In all of our work we are observing
B; A and C concern us only because of their connection
with B. Thanks to the sciences of physiology and physics, we know
enough about the speech-event to see that it consists of three parts:

(B1) The speaker, Jill, moved her vocal chords (two little
muscles inside the adam's-apple), her lower jaw, her tongue, and
so on, in a way which forced the air into the form of sound-waves.
These movements of the speaker are a reaction to the stimulus S.
Instead of performing the practical (or handling) reaction R —
namely, starting realistically off to get hold of the apple — she
performed these vocal movements, a speech (or substitute) reaction,
which we shall symbolize by a small letter r. In sum, then,
Jill, as a speaking person, has not one but two ways of reacting
to a stimulus:

image S → R (practical reaction)
image S → r (linguistic substitute reaction).

In the present case she performed the latter.

(B2) The sound-waves in the air in Jill's mouth set the surrounding
air into a similar wave-motion.

(B3) These sound-waves in the air struck Jack's ear-drums
and set them vibrating, with an effect on Jack's nerves: Jack
heard the speech. This hearing acted as a stimulus on Jack: we
saw him running and fetching the apple and placing it in Jill's
grasp, much as if Jill's hunger-and-apple stimulus had been acting
on him. An observer from another planet, who did not know
that there was such a thing as human speech, would have to conclude
that somewhere in Jack's body there was a sense-organ
which told him, “Jill is hungry and sees an apple up there.” In
short, Jack, as a speaking person, reacts to two kinds of stimuliː
practical stimuli of the type S (such as hunger and the sight of
food) and speech (or substitute) stimuli, certain vibrations in his
ear-drums, which we shall symbolize by a small letter s. When
we seek Jack doing anything (fetching an apple, say), his action
may be due not only, as are an animal's actions, to a practical
stimulus (such as hunger in his stomach, or the sight of an apple),
but, just as often, to a speech-stimulus. His actions, R, may be
prompted not by one, but by two kinds of proddings:

image (practical stimulus) S → R
image (linguistic substitute stimulus) s → R.25

It is evident that the connection between Jill's vocal movements
(B1) and Jack's hearing (B3) is subject to very little uncertainty
or variation, since it is merely a matter of sound-waves
passing through the air (B2). If we represent this connection by
a dotted line, then we can symbolize the two human ways of
responding to a stimulus by these two diagrams:

image speechless reaction: S→R
image reaction mediated by speech: S → r … s → R.

The difference between the two types is evident. The speechless
reaction occurs always in the same person as does the stimulus;
the person who gets the stimulus is the only one who can make
the response. The response, accordingly, is limited to whatever
actions the receiver of the stimulus can make. In contrast with
this, the reaction mediated by speech may occur in a person who
did not get the practical stimulus; the person who gets a stimulus
can prompt another person to make a response, and this other
person may be able to do things which the speaker cannot. The
arrows in our diagrams represent the sequence of events within
one person's body — a sequence of events which we think is due
to some property of the nervous system. Therefore the speechless
reaction can take place only in the body which received the stimulus.
In the reaction mediated by speech, on the other hand, there
is the link, represented by a dotted line, which consists of sound-waves
in the air: the reaction mediated by speech can take place
in the body of any person who hears the speech; the possibilities
of reaction are enormously increased, since different hearers may
be capable of a tremendous variety of acts. The gap between the
bodies of the speaker and the hearer — the discontinuity of the two
nervous systems — is bridged by the sound-waves

The important things, biologically, are the same in both the
speechless and the speaking occurrence, namely S (the hunger
and sight of the food) and R (movements which get the food or
fail to get it). These are the practical phase of the affair. The
speech-occurrence, s…r, is merely a means by which S and
R may occur in different individuals. The normal human being is
interested only in S and R; though he uses speech, and thrives by
it, he pays no attention to it. Saying the word apple or hearing
it said, appeases no one's hunger. It, along with the rest of speech,
is only a way of getting one's fellow-men to help. As students of
language, however, we are concerned precisely with the speech
26event (s…r), worthless in itself, but a means to great ends.
We distinguish between language, the subject of our study, and
real or practical events, stimuli and reactions. When anything
apparently unimportant turns out to be closely connected with
more important things, we say that it has, after all, a “meaning;”
namely, it “means” these more important things. Accordingly,
we say that speech-utterance, trivial and unimportant in itself,
is important because it has a meaning: the meaning consists of
the important things with which the speech-utterance (B) is
connected, namely the practical events (A and C).

2. 4. Up to a certain point, some animals respond to each
others' stimuli. Evidently the marvelous co-ordination in a group
of ants or bees must be due to some form of interaction. Sounds
as a means for this are common enough: crickets, for instance,
call other crickets by stridulation, noisily rubbing the leg against
the body. Some animals, like man, use vocal noises. Birds produce
sound-waves by means of the syrinx, a pair of reed-like organs at
the head of the lungs. The higher mammals have a larynx, a box
of cartilage (in man called the adam's-apple) at the top of the
wind-pipe. Inside the larynx, at the right and left, two shelf-like
muscles run along the walls; when these muscles, the vocal chords,
are stretched taut, the outgoing breath sets them into a regular
vibration which produces sound. This sound we call the voice.

Human speech differs from the signal-like actions of animals,
even of those which use the voice, by its great differentiation.
Dogs, for instance, make only two or three kinds of noise — say,
barking, growling, and whining: a dog can set another dog acting
by means of only these few different signals. Parrots can make
a great many kinds of noise, but apparently do not make different
responses to different sounds. Man utters many kinds of vocal
noise and makes use of the variety: under certain types of stimuli
he produces certain vocal sounds, and his fellows, hearing these
same sounds, make the appropriate response. To put it briefly,
in human speech, different sounds have different meanings. To
study this co-ordination of certain sounds with certain meanings
is to study language.

This co-ordination makes it possible for man to interact with
great precision. When we tell someone, for instance, the address
of a house he has never seen, we are doing something which no
animal can do. Not only has each person at his service the abilities
27of many other persons, but this co-operation is very precise. The
extent and accuracy of this working-together is the measure of
success of our social organization. The term society or social organism
is not a metaphor. A human social group is really a unit
of a higher order than a single animal, just as a many-celled animal
is a unit of a higher order than a single cell. The single cells in the
many-celled animal co-operate by means of such arrangements as
the nervous system; the individuals in a human society co-operate
by means of sound-waves.

The different ways in which we profit by language are so obvious
that we need mention only a few. We can relay communication.
When some farmers or traders say We want a bridge over this stream,
this news may pass through a town meeting, a state legislature, a
bureau of roads, an engineering staff, and a contractor's office,
running through many speakers and many relays of speech, until
at last, in response to the farmers' original stimulus, a corps of
workmen make the actual (practical) response movements of
putting up a bridge. Closely connected with the relay character
of speech is its abstraction. The relays of speech, between the
practical stimulus and the practical response, have no immediate
practical effect. Therefore they can be put into all kinds of forms,
provided only one changes them back correctly before proceeding
to the final, practical response. The engineer who plans the bridge
does not have to handle the actual beams and girders; he works
merely with speech-forms (such as numbers in calculation); if he
makes a mistake, he does not destroy any materials; he need only
replace the ill-chosen speech-form (say, a wrong figure) by a suitable
one before he begins the actual building. In this lies the value
of talking to oneself or thinking. As children, we talk to ourselves
aloud, but, under the correction of our elders, we soon learn to
suppress the sound-producing movements and replace them by
very slight inaudible ones: we “think in words.” The usefulness
of thinking can be illustrated by the process of counting. Our
ability to estimate numbers without using speech, is extremely
limited, as anyone may see by glancing, say, at a row of books on
a shelf. To say that two sets of objects “have the same number”
means that if we take one object from the first set and place it
next to one object of the second set, and keep on doing this without
using any object more than once, we shall have no unpaired objects
left over. Now, we cannot always do this. The objects may
28be too heavy to move, or they may be in different parts of the world,
or they may exist at different times (as, say, a flock of sheep before
and after a storm). Here language steps in. The numerals one,
two, three, four, and so on, are simply a series of words which we
have learned to say in a fixed order, as substitutes for the above-described
process. Using them, we can “count” any set of objects
by placing them into one-to-one correspondence (as mathematicians
call it) with the number-words, saying one for one of the
objects, two for another, three for the next, and so on, taking care
to use each object only once, until the objects of the set are exhausted.
Suppose that when we had said nineteen, there were no
more objects left. Thereafter, at any time or place, we can decide
whether any set of objects has the same number as this first set,
by merely repeating the counting process with the new set. Mathematics,
the ideal use of language, consists merely of elaborations
of this process. The use of numbers is the simplest and clearest
case of the usefulness of talking to oneself, but there are many
others. We think before we act.

2. 5. The particular speech-sounds which people utter under
particular stimuli, differ among different groups of men; mankind
speaks many languages. A group of people who use the same
system of speech-signals is a speech-community. Obviously, the
value of language depends upon people's using it in the same way.
Every member of the social group must upon suitable occasion
utter the proper speech-sounds and, when he hears another utter
these speech-sounds, must make the proper response. He must
speak intelligibly and must understand what others say. This
holds good for even the least civilized communities; wherever we
find man, he speaks.

Every child that is born into a group acquires these habits
of speech and response in the first years of his life. This is doubtless
the greatest intellectual feat any one of us is ever required to
perform. Exactly how children learn to speak is not known; the
process seems to be something like this:

(1) Under various stimuli the child utters and repeats vocal
sounds. This seems to be an inherited trait. Suppose he makes a
noise which we may represent as da, although, of course, the
actual movements and the resultant sounds differ from any that
are used in conventional English speech. The sound-vibrations
strike the child's ear-drums while he keeps repeating the movements.
29This results in a habit: whenever a similar sound strikes
his ear, he is likely to make these same mouth-movements, repeating
the sound da. This babbling trains him to reproduce vocal
sounds which strike his ear.

(2) Some person, say the mother, utters in the child's presence a
sound which resembles one of the child's babbling syllables. For
instance, she says doll. When these sounds strike the child's ear,
his habit (1) comes into play and he utters his nearest babbling
syllable, da. We say that he is beginning to “imitate.” Grown-ups
seem to have observed this everywhere, for every language seems
to contain certain nursery-words which resemble a child's babbling
— words like mama, dada: doubtless these got their vogue because
children easily learn to repeat them.

(3) The mother, of course, uses her words when the appropriate
stimulus is present. She says doll when she is actually showing
or giving the infant his doll. The sight and handling of the doll
and the hearing and saying of the word doll (that is, da) occur
repeatedly together, until the child forms a new habit: the sight
and feel of the doll suffice to make him say da. He has now the
use of a word. To the adults it may not sound like any of their
words, but this is due merely to its imperfection. It is not likely
that children ever invent a word.

(4) The habit of saying da at sight of the doll gives rise to further
habits. Suppose, for instance, that day after day the child is given
his doll (and says da, da, da) immediately after his bath. He has
now a habit of saying da, da after his bath; that is, if one day the
mother forgets to give him the doll, he may nevertheless cry da, da
after his bath. “He is asking for his doll,” says the mother, and
she is right, since doubtless an adult's “asking for” or “wanting”
things is only a more complicated type of the same situation.
The child has now embarked upon abstract or displaced speech:
he names a thing even when that thing is not present.

(5) The child's speech is perfected by its results. If he says da,
da well enough, his elders understand him; that is, they give him
his doll. When this happens, the sight and feel of the doll act as
an additional stimulus, and the child repeats and practises his
successful version of the word. On the other hand, if he says his
da, da imperfectly, — that is, at great variance from the adults'
conventional form doll, — then his elders are not stimulated to
give him the doll. Instead of getting the added stimulus of seeing
30and handling the doll, the child is now subject to other distracting
stimuli, or perhaps, in the unaccustomed situation of having no
doll after his bath, he goes into a tantrum which disorders his
recent impressions. In short, his more perfect attempts at speech
are likely to be fortified by repetition, and his failures to be wiped
out in confusion. This process never stops. At a much later stage,
if he says Daddy bringed it, he merely gets a disappointing answer
such as No! You must say “Daddy brought it”; but if he says Daddy
brought it
, he is likely to hear the form over again: Yes, Daddy
brought it
, and to get a favorable practical response.

At the same time and by the same process, the child learns
also to act the part of a hearer. While he is handling the doll he
hears himself say da, da and his mother say doll. After a time,
hearing the sound may suffice to make him handle the doll. The
mother will say Wave your hand to Daddy, when the child is doing
this of his own accord or while she is holding up the child's arm
and waving it for him. The child forms habits of acting in conventional
ways when he hears speech.

This twofold character of the speech-habits becomes more and
more unified, since the two phases always occur together. In
each case where the child learns the connection image S → r
(for instance, to say doll when he sees his doll), he learns also the
connection image s → R (for instance, to reach for his doll
or handle it when he hears the word doll). After he has learned
a number of such twofold sets, he develops a habit by which one
type always involves the other: as soon as he learns to speak
a new word, he is also able to respond to it when he hears others
speak it, and, vice versa, as soon as he learns how to respond to
some new word, he is usually able, also, to speak it on proper
occasion. The latter transference seems to be the more difficult of
the two; in later life, we find that a speaker understands many
speech-forms which he seldom or never employs in his own speech.

2. 6. The happenings which in our diagram are represented by
a dotted line, are fairly well understood. The speaker's vocal
chords, tongue, lips, and so on, interfere with the stream of his
outgoing breath, in such a way as to produce sound-waves; these
waves are propagated through the air and strike the hearer's
ear-drums, which then vibrate in unison. The happenings, however,
which we have represented by arrows, are very obscure.
We do not understand the mechanism which makes people say
31certain things in certain situations, or the mechanism which makes
them respond appropriately when these speech-sounds strike their
ear-drums. Evidently these mechanisms are a phase of our general
equipment for responding to stimuli, be they speech-sounds
or others. These mechanisms are studied in physiology and,
especially, in psychology. To study them in their special bearing
on language, is to study the psychology of speech, linguistic
. In the division of scientific labor, the linguist deals
only with the speech-signal (r…s); he is not competent to
deal with problems of physiology or psychology. The findings of
the linguist, who studies the speech-signal, will be all the more
valuable for the psychologist if they are not distorted by any prepossessions
about psychology. We have seen that many of the
older linguists ignored this; they vitiated or skimped their reports
by trying to state everything in terms of some psychological theory.
We shall all the more surely avoid this fault, however, if we
survey a few of the more obvious phases of the psychology of

The mechanism which governs speech must be very complex
and delicate. Even if we know a great deal about a speaker and
about the immediate stimuli which are acting upon him, we usually
cannot predict whether he will speak or what he will say.
We took our story of Jack and Jill as something known to us,
after the fact. Had we been present, we could not have foretold
whether Jill would say anything when she saw the apple, or, in
case she did speak, what words she would utter. Even supposing
she asked for the apple, we could not foretell whether she would
preface her request by saying I'm hungry or whether she would
say please or whether she would say I want that apple or Get me
that apple
or I was just wishing I had an apple, and so on: the
possibilities are almost infinite. This enormous variability has led
to two theories about human conduct, including speech.

The mentalistic theory, which is by far the older, and still prevails
both in the popular view and among men of science, supposes
that the variability of human conduct is due to the interference of
some non-physical factor, a spirit or will or mind (Greek psyche,
hence the term psychology) that is present in every human being.
This spirit, according to the mentalistic view, is entirely different
from material things and accordingly follows some other kind of
causation or perhaps none at all. Whether Jill will speak or what
32words she will use, depends, then, upon some act of her mind or
will, and, as this mind or will does not follow the patterns of succession
(cause-and-effect sequences) of the material world, we
cannot foretell her actions.

The materialistic (or, better, mechanistic) theory supposes that
the variability of human conduct, including speech, is due only
to the fact that the human body is a very complex system. Human
actions, according to the materialistic view, are part of cause-and-effect
sequences exactly like those which we observe, say in the
study of physics or chemistry. However^ the human body is so
complex a structure that even a relatively simple change, such
as, say, the impingement on the retina of light-waves from a red
apple, may set off some very complicated chain of consequences,
and a very slight difference in the state of the body may result in
a great difference in its response to the light-waves. We could
foretell a person's actions (for instance, whether a certain stimulus
will lead him to speak, and, if so, the exact words he will utter),
only if we knew the exact structure of his body at the moment, or,
what comes to the same thing, if we knew the exact make-up of
his organism at some early stage — say at birth or before — and
then had a record of every change in that organism, including
every stimulus that had ever affected the organism.

The part of the human body responsible for this delicate and
variable adjustment, is the nervous system. The nervous system
is a very complex conducting mechanism, which makes it possible
for a change in one part of the body, (a stimulus, say, in the eye)
to result in a change in some other part (a response, say, of reaching
with the arm, or of moving the vocal chords and tongue).
Further, it is clear that the nervous system is changed, for a time
or even permanently, by this very process of conduction: our
responses depend very largely upon our earlier dealings with the
same or similar stimuli. Whether Jill will speak depends largely
on her liking for apples and on her past experience of Jack. We
remember and acquire habits and learn. The nervous system is
evidently a trigger-mechanism: a very slight change may set the
match to a large store of explosive material. To take the case that
interests us, only so can we explain the fact that large-scale movements
like Jack's fetching the apple, are set off by very slight
changes, such as the minute thrumming of air-waves on his eardrum.33

The working of the nervous system is not accessible to observation
from without, and the person himself has no sense-organs
(such as he has, for instance, for the working of the muscles in his
hand) with which he himself could observe what goes on in his
nerves. Therefore the psychologist must resort to indirect methods
of approach.

2. 7. One such method is experiment. The psychologist submits
numbers of people to carefully prearranged stimuli under the
simplest conditions, and records their responses. Usually he also
asks these persons to “introspect,” — that is, to describe as much
as possible of what goes on inside them when they get the stimulus.
At this point psychologists often go astray for want of linguistic
knowledge. It is a mistake, for instance, to suppose that language
enables a person to observe things for which he has no sense-organs,
such as the workings of his own nervous system. An observer's
only advantage in reporting what goes on inside him is that
he can report stimulations which an outsider cannot detect — say,
a pain in his eye or a tickling in his throat. Even here, we must
not forget that language is a matter of training and habit; a
person may be unable to report some stimulations, simply because
his stock of speech-habits provides no formula; this is the case
with many of our less useful adventures, such as smaller goings-on
in our internal organs. Often the very structure of our body leads
to a false report; we show the physician exactly the spot where we
feel a pain, and he finds the injury some distance away, at a
point which his experience may teach him to locate at once from
our false description. In this respect many psychologists go astray
by actually training their observers to use a set of technical terms
for obscure stimuli and then attaching significance to the observer's
use of these terms.

Abnormal conditions in which speech is disturbed, seem to
reflect general maladjustments or lesions and to throw no light on
the particular mechanism of language. Stuttering is probably due to
imperfect specialization of the two cerebral hemispheres: in the
normal speaker the left hemisphere (or, if he is left-handed, the
right hemisphere) dominates more delicate actions, such as those of
speech; in the stutterer this one-sided specialization is incomplete.
Imperfect production of specific sounds (stammering), where it is
not due to anatomical defects in the organs of speech, seems to
result from similar maladjustments. Head-wounds and diseases
34which injure the brain often result in aphasia, disturbances in the
manner of making speech-responses and in responding to speech.
Dr. Henry Head, who had unusually good opportunities for the
study of aphasia in wounded soldiers, recognizes four types.

Type 1 reacts well to other people's speech, and in milder cases,
uses words for the proper objects, but mispronounces or confuses
his words; in extreme cases, the sufferer can say little more than
yes and no. A patient reports, with some difficulty: “I know it's
not … the correct … pronunciation … I don't always…
corret it … because I shouldn't get it right…
in five or six times … unless someone says it for
me.” In a more serious case, the patient, when asked his name,
answers Honus instead of ‘Thomas,’ and says erst for ‘first’ and
hend for ‘second.’

Type 2 reacts fairly well to simple speech, and pronounces
appropriate words and short phrases, but not in the conventional
constructions; he may talk an unintelligible jargon, although each
word is correct enough. To the question “Have you played any
games?” a patient answers: “Played games, yes, played one, daytime,
garden.” He says, “Get out, lay down, go to sleep, sometimes
goes away. If sit in kitchen, moving about working, makes
me getting worse on it.” He comments, “Funny thing, this worse,
that sort of thing,” and by way of explanation, writes down the
words as and at. We shall see later that the structure of normal
language forces us to distinguish between lexical and grammatical
habits of speech; the latter are disturbed in these patients.

Type 3 reacts with difficulty to the names of objects, and has
trouble in finding the right words, especially names of things. His
pronunciation and arrangement are good, but he has to use ingenious
circumlocutions for the words he cannot find. For ‘scissors’
a patient says “what you cut with;” for ‘black’ he says: “people
who are dead, — the other people who are not dead, have this
color.” He may use the wrong word, as button for ‘scissors.’ The
words lost are chiefly the names of concrete objects. This state
seems like an exaggeration of many normal persons' difficulty in
recalling people's names and the designations of objects, especially
under preoccupation, excitement, or fatigue.

Type 4 often does not respond correctly to the speech of others;
he has no trouble in uttering single words, but he cannot finish a
connected speech. It is significant that these patients suffer from
35apraxia; they cannot find their way about and are confused by
being set, say, on the opposite side of the street. One patient
reports: “I don't seem to understand all you say, and then I forget
what I've got to do.” Another patient says: “When at table, I
am very slow in picking out the object, say the milk-jug, which I
want. I don't spot it at once … I see them all, but I don't
spot them. When I want the salt or the pepper or a spoon, I
suddenly tumble to its presence.” The disturbance of speech
appears in this answer of a patient: “Oh, yes! I know the difference
between the Nurse and the Sister by the dress: Sister blue;
Nurse — oh! I get muddled, just ordinary nurse's clothes, white,

Ever since 1861, when Broca showed that damage to the third
frontal convolution in the left hemisphere of the brain was accompanied
by aphasia, there has been dispute as to whether “Broca's
center” and other regions of the cortex act as specific centers for
the activity of speech. Head finds some correlation between
different points of lesion and each of his four types of aphasia. The
demonstrable functional identifications of cortical areas always
concern some specific organ: an injury in one area of the brain is
accompanied by paralysis of the right foot, an injury in another
area by failure to respond to stimulation in the left-hand side of the
retina, and so on. Now, speech is a very complex activity, in which
stimulation of every kind leads to highly specific movements of
the throat and mouth; these last, moreover, are not, in a physiologic
sense, “organs of speech,” for they serve biologically earlier
uses in man and in speechless animals. Many injuries to the nervous
system, accordingly, will interfere with speech, and different
injuries will result in different kinds of difficulty, but the points of
the cortex are surely not correlated with specific socially significant
features of speech, such as words or syntax; this appears plainly
from the fluctuating and contradictory results of the search for
various kinds of “speech centers.” We may expect the physiologist
to get better results when he looks for correlations between points
of the cortex and specific physiologic activities concerned in
speech, such as the movement of special muscles or the transmission
of kinesthetic stimuli from the larynx and tongue. The error of
seeking correlations between anatomically defined parts of the
nervous system and socially defined activities appears clearly when
we see some physiologists looking for a “visual word-center” which
36is to control reading and writing: one might as well look for a
specific brain-center for telegraphy or automobile-driving or the
use of any modern invention. Physiologically, language is not a
unit of function, but consists of a great many activities, whose
union into a single far-reaching complex of habits results from
repeated stimulations during the individual's early life.

2. 8. Another way of studying human responses is to observe
them in the mass. Some actions are highly variable in each person,
but fairly constant in large groups of persons. We cannot
predict whether any particular unmarried adult will marry during
the next twelve months, or which particular persons will
commit suicide, or which ones will get into prison, but, given a
large enough community, and the figures for past years (and perhaps
certain other data, such as those which concern economic
conditions), statisticians can foretell the number of marriages,
suicides, convictions for crime, and so on, which will take place.
If we found it possible and worth while to register every speech-utterance
in a large community, we should doubtless be able to
foretell how many times any given utterance such as Good-morning
or I love you or How much are oranges today? would be spoken
within a fixed number of days. A detailed study of this kind
would tell us a great deal, especially about the changes that are
constantly going on in every language.

However, there is another and simpler way of studying human
action in the mass: the study of conventional actions. When we
go to a strange country, we soon learn many established modes of
action, such as the system of currency and of weights and measures,
the rules of the road (does one keep to the right, as in America
and Germany, or to the left, as in England and Sweden?),
good manners, hours for meals, and so on. The traveler does not
gather statistics: a very few observations put him on the track,
and these are confirmed or corrected by further experience. Here
the linguist is in a fortunate position: in no other respect are the
activities of a group as rigidly standardized as in the forms of
language. Large groups of people make up all their utterances
out of the same stock of lexical forms and grammatical constructions.
A linguistic observer therefore can describe the speech-habits
of a community without resorting to statistics. Needless
to say, he must work conscientiously and, in particular, he must
record every form he can find and not try to excuse himself from
37this task by appealing to the reader's common sense or to the
structure of some other language or to some psychological theory,
and, above all, he must not select or distort the facts according
to his views of what the speakers ought to be saying. Aside from
its intrinsic value for the study of language, a relevant and unprejudiced
description of this kind, serves as a document of major
importance for psychology. The danger here lies in mentalistic
views of psychology, which may tempt the observer to appeal to
purely spiritual standards instead of reporting the facts. To say,
for instance, that combinations of words which are “felt to be”
compounds have only a single high stress (e.g. blackbird as opposed
to black bird), is to tell exactly nothing, since we have no way of
determining what the speakers may “feel”: the observer's task
was to tell us, by some tangible criterion, or, if he found none, by
a list, which combinations of words are pronounced with a single
high stress. A worker who accepts the materialistic hypothesis in
psychology is under no such temptation; it may be stated as a
principle that in all sciences like linguistics, which observe some
specific type of human activity, the worker must proceed exactly
as if he held the materialistic view. This practical effectiveness is
one of the strongest considerations in favor of scientific materialism.

The observer who, by this mass-observation, gives us a statement
of the speech-habits of a community, can tell us nothing
about the changes which are going on in the language of this as
of every community. These changes could be observed only by
means of genuinely statistical observation through a considerable
length of time; for want of this, we are ignorant of many matters
concerning linguistic change. In this respect, too, the science of
language is fortunate, however, because comparative and geographical
methods of study, again through mass-observation, supply a
good deal of what we should hope to get from statistics. The
fortunate position of our science in these matters is due to the
fact that language is the simplest and most fundamental of our
social (that is, peculiarly human) activities. In another direction,
however, the study of linguistic change profits by a mere accident,
namely by the existence of written records of speech of the past,

2. 9. The stimulus which calls forth speech, leads also to some
other reactions. Some of these are not visible from the outside;
these are muscular and glandular actions which are of no immediate
importance to the speaker's fellow-men. Others are important
38handling responses, such as locomotion or the displacement
of objects. Still other responses are visible, but not directly important;
they do not change the lay-out of things, but they do,
along with speech, serve as stimuli to the hearer. These actions
are facial expression, mimicry, tone of voice (in so far as it is not
prescribed by the conventions of the language), insignificant
handling of objects (such as fiddling with a rubber band), and,
above all, gesture.

Gesture accompanies all speech; in kind and in amount, it
differs with the individual speaker, but to a large extent it is
governed by social convention. Italians use more gesture than
English-speaking people; in our civilization people of the privileged
class gesticulate least. To some extent, individual gestures are conventional
and differ for different communities. In saying good-by
we wave the hand with palm outward; Neapolitans wave it with
the back outward.

Most gestures scarcely go beyond an obvious pointing and picturing.
American Indians of plains or woodland tribes will accompany
a story by unobtrusive gestures, foreign to us, but quite
intelligible: the hand, palm in, thumb up, is held just under the
eyes to represent spying; a fist is slapped into a palm for a shot;
two fingers imitate a man walking, and four the running of a horse.
Even where gestures are symbolic, they go little beyond the obvious,
as when one points back over one's shoulder to indicate
past time.

Some communities have a gesture language which upon occasion
they use instead of speech. Such gesture languages have been
observed among the lower-class Neapolitans, among Trappist
monks (who have made a vow of silence), among the Indians of
our western plains (where tribes of different language met in
commerce and war), and among groups of deaf-mutes.

It seems certain that these gesture languages are merely developments
of ordinary gestures and that any and all complicated
or not immediately intelligible gestures are based on the conventions
of ordinary speech. Even such an obvious transference as
pointing backward to indicate past time, is probably due to a linguistic
habit of using the same word for ‘in the rear’ and ‘in the
past.’ Whatever may be the origins of the two, gesture has so long
played a secondary role under the dominance of language that it
has lost all traces of independent character. Tales about peoples
39whose language is so defective that it has to be eked out by gesture,
are pure myths. Doubtless the production of vocal sound by animals,
out of which language has grown, originated as a response-movement
(say, contraction of the diaphragm and constriction
of the throat) which happened to produce noise. It seems certain,
however, that in the further development, language always ran
ahead of gesture.

If one gestures by moving some object so as to leave a trace on
another object, one has entered upon marking and drawing. This
kind of reaction has the value of leaving a permanent mark, which
may serve as a stimulus repeatedly and even after intervals of
time and can be transported to stimulate persons far away. For
this reason, doubtless, many peoples attribute magic power to
drawings, apart from their esthetic value, which is still with us.

In some parts of the world drawing has developed into writing.
The details of this process will concern us later; the point of interest
here is that the action of tracing an outline becomes subordinate
to language: drawing a particular set of lines becomes
attached, as an accompaniment or substitute, to the utterance of
a particular linguistic form.

The art of symbolizing particular forms of speech by means of
particular visible marks adds a great deal to the effective uses of
language. A speaker can be heard only a short ways and only for
an instant or two. A written record can be carried to any place
and preserved for any length of time. We can see more things at
one time than we can hear, and we can deal better with visible
things: charts, diagrams, written calculations, and similar devices,
enable us to deal with very complex matters. The speech-stimuli
of distant people, and especially of persons in the past, are available
to us through writing. This makes possible an accumulation of
knowledge. The man of science (but not always the amateur)
surveys the results of earlier students and applies his energies at
the point where they left off. Instead of always starting over
again from the beginning, science progresses cumulatively and
with acceleration. It has been said that, as we preserve more and
more records of more and more speech-reactions of highly gifted
and highly specialized individuals, we approach, as an ideal limit,
a condition where all the events in the universe, past, present, and
future, are reduced (in a symbolic form to which any reader may
react) to the dimensions of a large library. It is no wonder that
40the discovery of printing, which manifolds a written record to any
desired number of copies, brought about, in all our manner of
living, a revolution which has been under way for some centuries
and is still in full swing.

There is no need of dilating upon the significance of other means
for recording, transmitting, and multiplying speech, such as the
telegraph, telephone, phonograph, and radio. Their importance
for the simpler uses of language is obvious, as in the use of wireless
telegraphy in cases of shipwreck.

In the long run, anything which adds to the viability of language
has also an indirect but more pervasive effect. Even acts
of speech that do not prompt any particular immediate response,
may change the predisposition of the hearer for further responses:
a beautiful poem, for instance, may make the hearer more sensitive
to later stimuli. This general refinement and intensification
of human response requires a great deal of linguistic interaction.
Education or culture, or whatever name we choose to give it, depends
upon the repetition and publication of a vast amount of

Chapter 3

3. 1. A speech-community is a group of people who interact by
means of speech (§ 2.5). All the so-called higher activities of man
— our specifically human activities — spring from the close adjustment
among individuals which we call society, and this adjustment,
in turn, is based upon language; the speech-community,
therefore, is the most important kind of social group. Other
phases of social cohesion, such as economic, political, or cultural
groupings, bear some relation to the grouping by speech-communities,
but do not usually coincide with it; cultural features, especially,
are almost always more widespread than any one language.
Before the coming of the white man, an independent Indian tribe
which spoke a language of its own, formed both a speech-community
and a political and economic unit; as to religion and general
culture, however, it resembled neighboring tribes. Under more
complex conditions there is less correlation between language and
the other groupings. The speech-community which consists of
all English-speaking people is divided into two political communities:
the United States and the British Empire, and each of
these is in turn subdivided; economically, the United States and
Canada are more closely united than politically; culturally, we
are part of a great area which radiates from western Europe. On
the other hand, even the narrowest of these groups, the political
United States, includes persons who do not speak English: American
Indians, Spanish-speakers in the Southwest, and linguistically
unassimilated immigrants. Colonial occupation, as in the Philippines
or India, puts a speech-community into political and economic
dependence upon a foreign speech-community. In some
countries the population is divided into several speech-communities
that exist together without local division: a town in Poland consists
of Polish-speaking and German-speaking people; by religion,
the former are Catholics, the latter Jews, and, until quite recently,
very few persons in either group troubled themselves to understand
the other group's language.42

I have said nothing about biological grouping, because this
does not, like the other groupings, depend upon language for its
existence. Most matings, of course, take place between persons
of like speech, so that a speech-community is always something
of an inbred group; the exceptions, however, are very many, both
in the mating of persons of different speech, one of whom usually
acquires the other's language, and, what is more important, in
the assimilation into a speech-community of whole groups of foreigners,
such as immigrants, conquered people, or captives. These
deviations are so many that, if we had records, we should doubtless
find very few persons whose ancestors of a few generations ago all
spoke the same language. What concerns us most, however, is
the fact that the features of a language are not inherited in the
biologic sense. A child cries out at birth and would doubtless in
any case after a time take to gurgling and babbling, but the particular
language he learns is entirely a matter of environment.
An infant that gets into a group as a foundling or by adoption,
learns the language of the group exactly as does a child of native
parentage; as he learns to speak, his language shows no trace of
whatever language his parents may have spoken. Whatever
hereditary differences there may be in the structure of the larynx,
mouth, lips, and so on, of normal human beings, it is certain that
these differences are not such as to affect the actions which make
up language. The child learns to speak like the persons round him.
The first language a human being learns to speak is his native
; he is a native speaker of this language.

3. 2. Speech-communities differ greatly in size. More than one
American Indian tribe of only a few hundred persons spoke a
language of its own. On the other hand, even before the corning
of modern communication and travel, some speech-communities
were very large: in the first centuries of the Christian Era, Latin
and Greek were each spoken by millions of people over large areas
round the Mediterranean. Under modern conditions, some speech-communities
have grown to enormous size. Jespersen estimates
the number of speakers of the principal European languages, in
millions, for the years 1600 and 1912 as follows:

tableau English | German | Russian | French | Spanish | Italian | 1600 | 191243

Figures such as these have only a very indefinite value, because
one cannot always tell which local groups form a single speech-community.
Tesniere, estimating the numbers round the year
1920, names Chinese as the largest speech-community, with 400
million speakers, but the term Chinese denotes a family of mutually
unintelligible languages. Doubtless one of these, North Chinese,
has today more native speakers than any other language, but I
know no estimate of their number. Another language of this
group, Cantonese, probably ranks among the largest speech-communities.
In any case, English (to continue with Tesniere's
figures) ranks second, with 170 million native speakers. Russian
comes third; Tesniere divides the figures between Great Russian
(80 millions), Little Russian (Ukrainian, 34 millions), and White
Russian (6½ millions), but these are mutually intelligible varieties,
about as different as British and American English. Similarly,
Tesniere splits the fourth-greatest language, German, into German
(80 millions) and Judeo-German (7½ millions), although the
rest of his figures do not consider dialectal differences; Jespersen's
figure of 90 millions is probably nearer right. Tesniere's remaining
figures omit Javanese, which has at least 20 millions of native
speakers. With these modifications his figures are: Spanish 65,
Japanese 55, Bengali 12 50, French 45, Italian 41, Turco-Tartar
39, Western Hindi 13 38, Arabic 37, Bihari 14 36, Portuguese 36,
Eastern Hindi 15 25, Telugu 26 24, Polish 23, Javanese 20, Marathi 17
19, Tamil 28 19, Korean 17, Panjabi 19 16, Annamite 14, Roumanian
14, Rajasthani 110 13, Dutch 13, Bohemian-Slovak 12, Canarese 211 10,
Oriya 112 10, Hungarian 10.

Another element of uncertainty in figures like these arises from
the differences within speech-communities. Dutch and German
actually form only one speech-community, in the sense that there
is no break between local speech-forms, but the extreme types are
mutually unintelligible, and the political groups (on the one side
Flemish Belgium and the Netherlands, and on the other side,
Germany, Austria, and German Switzerland) have adopted two
mutually unintelligible speech-forms, Standard Dutch-Flemish and
Standard German, as their official languages. On the other hand,
Turco-Tartar and some of the languages of India in our list probably
44include equally great differences, although the extremes may
be connected by local gradations. A final and insurmountable difficulty
lies in people's acquisition of foreign languages. If we could
determine a degree of proficiency which makes a student a member
of a foreign speech-community, English, studied all over the world,
would receive a much larger figure. Tesniere estimates that Malay
is native to some three million people, but is spoken as a foreign
language, especially in commerce, by some thirty millions.

3. 3. The difficulty or impossibility of determining in each case
exactly what people belong to the same speech-community, is
not accidental, but arises from the very nature of speech-communities.
If we observed closely enough, we should find that
no two persons — or rather, perhaps, no one person at different
times — spoke exactly alike. To be sure, within a relatively homogeneous
set of speakers — say, the native speakers of English in
the Middle Western part of the United States — the habits of
speech are far more uniform than the needs of communication
would demand. We see the proof of this when an outsider — say,
a Southerner or an Englishman or a foreigner who has mastered
English — comes into our midst: his speech may be so much like
ours as to cause not the slightest difficulty in communication, and
yet strikingly noticeable on account of inessential differences,
such as “accent” and “idiom.” Nevertheless there are great
differences even among the native members of such a relatively
uniform group as Middle Western American, and, as we have
just seen, even greater differences within a speech-community
(e.g. English) as a whole. These differences play a very important
part in the history of languages; the linguist is forced to consider
them very carefully, even though in some of his work he is forced
provisionally to ignore them. When he does this, he is merely
employing the method of abstraction, a method essential to
scientific investigation, but the results so obtained have to be
corrected before they can be used in most kinds of further

The difference between speakers is partly a matter of bodily
make-up and perhaps of purely personal habit; we recognize our
friends by their voices from the next room and over the telephone.
Some people are more talented for speech than others: they remember
more words and turns of phrase, apply them better to the
situation, and combine them in more pleasing style; the extreme
45case is the literary genius. Sometimes convention assigns certain
speech-forms to certain speakers, as when the soldier, the well-trained
servant, and the child in certain schools, learn to say sir
or ma'm to certain persons, who do not reciprocate. Some exclamations,
such as Goodness gracious! or Dear me! are largely reserved
for the use of women. In some communities very different
speech-forms are conventional for the sexes. The classical
instance is that of the Carib Indians; a recently authenticated one
is the language of the Yana Indians in northern California. Examples
of Yana words are:

tableau men's language | women's language | fire | my fire | deer | grizzly-bear

The differences between the two sets of Yana forms can be stated
by means of a fairly complex set of rules.

3. 4. The most important differences of speech within a community
are due to differences in density of communication. The
infant learns to speak like the people round him, but we must not
picture this learning as coming to any particular end: there is
no hour or day when we can say that a person has finished learning
to speak, but, rather, to the end of his life, the speaker keeps
on doing the very things which make up infantile language-learning.
Our description of the latter (§ 2.5) might be taken, in many
respects, as a slow-motion picture of the ordinary processes of
speech. Every speaker's language, except for personal factors
which we must here ignore, is a composite result of what he has
heard other people say.

Imagine a huge chart with a dot for every speaker in the community,
and imagine that every time any speaker uttered a sentence,
an arrow were drawn into the chart pointing from his dot
to the dot representing each one of his hearers. At the end of a
given period of time, say seventy years, this chart would show us
the density of communication within the community. Some speakers
would turn out to have been in close communication: there
would be many arrows from one to the other, and there would be
many series of arrows connecting them by way of one, two, or
three intermediate speakers. At the other extreme there would be
46widely separated speakers who had never heard each other speak
and were connected only by long chains of arrows through many
intermediate speakers. If we wanted to explain the likeness and
unlikeness between various speakers in the community, or, what
comes to the same thing, to predict the degree of likeness for any
two given speakers, our first step would be to count and evaluate
the arrows and series of arrows connecting their dots. We shall
see in a moment that this would be only the first step; the reader
of this book, for instance, is more likely to repeat a speech-form
which he has heard, say, from a lecturer of great fame, than one
which he has heard from a street-sweeper.

The chart we have imagined is impossible of construction. An
insurmountable difficulty, and the most important one, would be
the factor of time: starting with persons now alive, we should be
compelled to put in a dot for every speaker whose voice had ever
reached anyone now living, and then a dot for every speaker whom
these speakers had ever heard, and so on, back beyond the days
of King Alfred the Great, and beyond earliest history, back indefinitely
into the primeval dawn of mankind: our speech depends
entirely upon the speech of the past.

Since we cannot construct our chart, we depend instead upon
the study of indirect results and are forced to resort to hypothesis.
We believe that the differences in density of communication
within a speech-community are not only personal and individual,
but that the community is divided into various systems
of sub-groups such that the persons within a sub-group speak
much more to each other than to persons outside their sub-group.
Viewing the system of arrows as a network, we may say that
these sub-groups are separated by lines of weakness in this net of
oral communication. The lines of weakness and, accordingly, the
differences of speech within a speech-community are local — due
to mere geographic separation — and non-local, or as we usually
say, social. In countries over which a speech-community has
recently spread and settled, the local differences are relatively
small, as, say, in the United States (especially the western part)
or Russia; in countries that have been long settled by the same
speech-community the local differences are much greater, as, say,
in England, where English has been spoken for some 1500 years,
or in France where Latin (now called French) has been spoken for
two-thousand years.47

3. 5. We shall examine first the simpler case, as it appears in
the United States. The most striking line of cleavage in our speech
is one of social class. Children who are born into homes of privilege,
in the way of wealth, tradition, or education, become native
speakers of what is popularly known as “good” English; the
linguist prefers to give it the non-committal name of standard
English. Less fortunate children become native speakers of “bad”
or “vulgar” or, as the linguist prefers to call it, non-standard
English. For instance, I have none, I haven't any, I haven't got any
are standard (“good”) English, but I ain't got none is non-standard
(“bad”) English.

These two main types of American English are by no means
treated alike. The standard forms are used in school, in church,
and in all discourse that officially concerns the whole community,
as in law-courts and legislative assemblies. All our writing (except
by way of jest) is based on the standard forms, and these forms
are registered in grammars and dictionaries and presented in
text-books to foreigners who want to learn our language. Both
groups of speakers, standard and non-standard, agree in calling
the standard forms “good” or “correct” and non-standard forms
“bad,” “incorrect,” “vulgar,” or even, “not English.” The
speaker of standard English does not trouble himself to learn the
non-standard forms, but very many speakers of non-standard English
try to use the standard forms. A native of the less favored group
who acquires prestige, say, in the way of wealth or political eminence,
is almost sure to learn, as well as may be, the standard forms
of speech; in fact, noticeable lapses in this respect — even a single
I seen it or I done it — may endanger his newly acquired position.

Within the standard language there, are minor differences. In
this case again, the divergent forms are estimated as higher and
lower. A Chicagoan, for instance, who uses the ah-vowel of father
instead of the more common a-vowel of man in words like laugh,
half, bath, dance, can't, is said to be speaking a “higher-class”
kind of English. In cases like these, however, people's attitudes
differ: many Chicagoans find these ah-forms silly and affected.
Speakers of standard English often dispute as to which of two
forms is “better”: it's I or it's me, forehead or “forrid”. Since the
disputants do not trouble themselves to agree on a definition of
“better,” these disputes never reach any conclusion. This is a
matter which will occupy us again.48

Within the standard language, further, there are differences
that obviously depend upon density of communication: different
economic classes, — say, the very rich and the so-called “middle
class” in its various gradations, — differ in speech. Then there
are differences of education, in the way both of family tradition
and of schooling. These differences are crossed by less important
divisions of technical occupation: different kinds of craftsmen,
merchants, engineers, lawyers, physicians, scientists, artists, and
so on, differ somewhat in speech. Sports and hobbies have at
least their own vocabulary. The factor of age-groups will concern
us later; it is a tremendous force, but works almost unseen, and
scarcely appears on the level that now concerns us, except perhaps
in young people's fondness for slang.

The most stable and striking differences, even in the United
States and even in our standard language, are geographic. In the
United States we have three great geographic types of standard
English: New England, Central-Western and Southern. Within
these types there are smaller local differences: speakers of standard
English from older-settled parts of the country can often tell a
fellow-speaker's home within fairly narrow limits. In matters of
pronunciation, especially, the range of standard English in America
is wide: greatly different pronunciations, such as those, say, of
North Carolina and Chicago, are accepted equally as standard.
Only from the stage do we demand a uniform pronunciation, and
here our actors use a British type rather than an American. In
England there are similar regional types, but they are not granted
equal value. The highest social recognition is given to the “public
school” English of the south. The innumerable gradations from
this toward the decidedly provincial types of standard, enjoy less
prestige as they depart from the most favored type. The social
recognition of a speaker of standard English from Scotland or
Yorkshire or Lancashire, depends in part upon how closely his
pronunciation approaches the upper-class southern type. In
England, but scarcely in the United States, provincial colorings of
standard English are tied up with differences of social level.

3. 6. Non-standard speech shows greater variety than standard.
The higher the social position of the non-standard speaker, the
more nearly does he approach the standard language. At the top
are the transitional speakers who use an almost standard form of
speech, with only a sprinkling of non-standard forms, and perhaps
49a pronunciation with too provincial a twang. At the bottom are
the unmistakably rustic or proletarian speakers who make no
pretense at using standard forms.

Apart from this continuous gradation, various groups of non-standard
speakers have their own speech-forms. Occupational
groups, such as fishermen, dairy workers, bakers, brewers, and so
on, have, at any rate, their own technical language. Especially,
minor groups who are in any way cut off from the great mass, use
clearly-marked varieties of speech. Thus, sea-faring men used to
speak their own type of non-standard English. Tramps and some
kinds of law-breakers have many speech-forms of their own; so
do circus people and other wandering entertainers. Among non-standard
speakers of German, Christians and Jews, and in some
places Catholics and Protestants, differ in many of their linguistic
forms. If the special group is at odds with the rest of the community,
it may use its peculiarities of speech as a secret dialect, as do
the English-speaking Gipsies. Criminals in various countries have
developed such secret dialects.

The greatest diversity in non-standard speech, however, is
geographic. The geographic differences, which we hear even in the
standard English of the United States, are more audible when we
listen to non-standard speakers. In remote districts within the
older-settled parts of the country these local characteristics are
very pronounced, to the point where we may describe them as
local dialects.

In older-settled speech-communities, the type exemplified by
France, or by the British part of the English-speaking group, local
dialects play a much greater part. In such communities the nonstandard
language can be divided, roughly, to be sure, and without
a sharp demarcation, into sub-standard speech, intelligible at least,
though not uniform, throughout the country, and local dialect,
which differs from place to place to such an extent that speakers
living some distance apart may fail to understand each other. Substandard
speech, in such countries, belongs to the “lower middle
class,” — to the more ambitious small tradesfolk, mechanics, or
city workmen, — and the local dialects are spoken by the peasants
and the poorest people of the towns.

The local dialects are of paramount importance to the linguist,
not merely because their great variety gives him work to do, but
because the origin and history of the standard and sub-standard
50types of speech can be understood only in the light of the local
dialects. Especially during the last decades, linguists have come
to see that dialect geography furnishes the key to many problems.

In a country like France, Italy, or Germany — better studied in
this respect than England — every village or, at most, every
group of two or three villages, has its own local dialect. The differences
between neighboring local dialects are usually small, but
recognizable. The villagers are ready to tell in what way their
neighbors' speech differs from theirs, and often tease their neighbors
about these peculiarities. The difference from place to place is
small, but, as one travels in any one direction, the differences
accumulate, until speakers, say from opposite ends of the country,
cannot understand each other, although there is no sharp line of
linguistic demarcation between the places where they live. Any
such geographic area of gradual transitions is called a dialect area.

Within a dialect area, we can draw lines between places which
differ as to any feature of language. Such lines are called isoglosses.
If a village has some unique peculiarity of speech, the isogloss based
on this peculiarity will be simply a line round this village. On the
other hand, if some peculiarity extends over a large part of the
dialect area, the isogloss of this feature will appear as a long line,
dividing the dialect area into two sections. In Germany, for instance,
the northern dialects pronounce the word bite with a t-sound,
as we do in English, but the southern dialects pronounce it with an
s-sound (as in standard German beiszen); the isogloss which separates
these two forms is a long and very irregular line, running east
and west across the whole German speech area. In the north and
northeast of England one can mark off an area where the past tense
of bring has the form brang. Dialect atlases, collections of maps of
a speech area with isoglosses drawn in, are an important tool for the

The speakers' attitude toward local dialects differs somewhat in
different countries. In England the local dialects have little prestige;
the upper-class speaker does not bother with them and the
native speaker of a local dialect who rises socially will try to cast
it off, even if only in exchange for some form of sub-standard
speech. The Germans, on the other hand, have developed, within
the last century, a kind of romantic fondness for local dialects.
While the middle-class speaker, who is not quite sure of his social
position, will shy away from them, some upper-class Germans make
51it a point to speak the local dialect of their home. In German
Switzerland this goes farthest: even the upper-class Swiss, who is
familiar with standard German, uses local dialect as the normal
medium of communication in his family and with his neighbors.

3. 7. The main types of speech in a complex speech-community
can be roughly classed as follows:

(1) literary standard, used in the most formal discourse and in
writing (example: I have none);

(2) colloquial standard, the speech of the privileged class (example:
I haven't any or I haven't got any — in England only if
spoken with the southern “public school” sounds and intonation);

(3) provincial standard, in the United States probably not to be
differentiated from (2), spoken by the “middle” class, very close
to (2), but differing slightly from province to province (example:
I haven't any or I haven't got any, spoken, in England, with sounds
or intonations that deviate from the “public school” standard);

(4) sub-standard, clearly different from (1), (2), and (3), spoken
in European countries by the “lower middle” class, in the United
States by almost all but the speakers of type (2-3), and differing
topographically, without intense local difference (example: I ain't
got none

(5) local dialect, spoken by the least privileged class; only slightly
developed in the United States; in Switzerland used also, as a
domestic language, by the other classes; differs almost from village
to village; the varieties so great as often to be incomprehensible to
each other and to speakers of (2-3-4) (Example: a hae nane).

3. 8. Our survey of differences within a speech-community has
shown us that the members of a speech-community may speak
so much alike that anyone can understand anyone else, or may
differ so much that persons who live some distance apart may
fail to understand each other. The former case is illustrated by
an Indian tribe of a few hundred persons, the latter by a far-flung
speech community like English, where an American and a
dialect-speaking Yorkshireman, for instance, do not understand
each other's speech. Actually, however, we can draw no line
between the two cases, because there are all kinds of gradations
between understanding and failing to understand. Whether the
American and the Yorkshireman understand each other, may
depend on the intelligence of the two individuals concerned, upon
their general experience with foreign dialects or languages, upon
52their disposition at the moment, upon the extent to which the
situation clarifies the value of the speech-utterance, and so on.
Again, there are endless gradations between local and standard
speech; either or both persons may make concessions which aid
understanding, and these concessions will usually run in the direction
of the standard language.

All this prevents our drawing a plain line round the borders
of many a speech community. The clear cases are those where two
mutually unintelligible languages abut on each other, as do, say,
English and Spanish in our Southwest. Here each person's native
language — if, for simplicity sake, we ignore the languages of
Indians and recent immigrants — is either English or Spanish,
and we can draw an imaginary line, a language boundary, which
will separate the English-speakers from the Spanish-speakers.
This language boundary will of course not appear as a simple
and fixed line between two topographically solid communities.
There will be English-speaking settlements thrown out, in the
shape of speech-islands, into totally Spanish surroundings, and,
vice versa, Spanish speech-islands surrounded by English-speaking
communities. Families and individuals of either group will be
found living among the other and will have to be enclosed in a
separate little circle of our language boundary. Our language
boundary, then, consists not only of a great irregular line, but
also of many little closed curves around speech-islands, some of
which contain only a single family or a single person. In spite of
its geometrical complexity and of its instability from day to day,
this language boundary at any rate represents a plain distinction.
It is true that linguistic scholars have found enough resemblance
between English and Spanish to prove beyond a doubt that these
languages are related, but the resemblance and relationship are
too distant to affect the question with which we are here concerned.

The same might be said, for instance, of German and Danish:
across the Jutland peninsula, just north of the city of Flensburg,
we could draw a boundary between the two languages, and this
boundary would show, on a smaller scale, the same features as
the English-Spanish boundary in our Southwest. In this case,
however, the resemblance between the two languages is sufficiently
close to warn us of further possibilities. The two languages
are mutually unintelligible, but resemble each other so closely
that it takes no linguistic research to see the relationship. If one
53can compare such things at all, the difference is no greater than
the difference between, say, a German local dialect spoken in
Sleswick and one spoken in Switzerland. German and Danish,
where they abut on each other, show a difference no greater than
the differences which may exist within a single locally differentiated
speech-community — only that in the latter case the intermediate
gradations intervene, while between German and Danish
we find no intermediate dialects.

The purely relative nature of this distinction appears more
plainly in other cases. We speak of French and Italian, of Swedish
and Norwegian, of Polish and Bohemian as separate languages,
because these communities are politically separate and use different
standard languages, but the differences of local speech-forms
at the border are in all these cases relatively slight and no greater
than the differences which we find within each of these speech-communities.
The question comes down to this: what degree of
difference between adjoining speech-forms justifies the name of a
language border? Evidently, we cannot weigh differences as
accurately as all this. In some cases, certainly, our habits of nomenclature
will not apply to linguistic conditions. The local dialects
justify no line between what we call German and what we call
Dutch-Flemish: the Dutch-German speech area is linguistically a
unit, and the cleavage is primarily political; it is linguistic only
in the sense that the political units use different standard languages.
In sum, the term speech-community has only a relative value. The
possibility of communication between groups, or even between
individuals, ranges all the way from zero up to the most delicate
adjustment. It is evident that the intermediate degrees contribute
very much to human welfare and progress.

3. 9. The possibilities of communication are enhanced and the
boundaries of the speech-community are further obscured by
another very important factor, namely, people's use of foreign
languages. This is by no means a modern accomplishment; among
peoples of simpler civilization, such as some tribes of American
Indians, well-bred persons often speak more than one of the
languages of neighboring tribes. The factor of foreign-language
speaking does not lend itself to measurement, since proficiency
ranges all the way down to a smattering so slight as to be of almost
no actual use. To the extent that the learner can communicate,
he may be ranked as a foreign speaker of a language. We have
54already seen that the usefulness of some languages, such as English
or Malay, is partly due to the adherence of foreign speakers.
Often enough, as among the educated classes in India, English
serves as the means of communication between foreign speakers
who do not understand each other's native languages.

Some people entirely give up the use of their native language
in favor of a foreign one. This happens frequently among immigrants
in the United States. If the immigrant does not stay in a
settlement of others from his own country, and especially if he
marries outside his original nationality, he may have no occasion
at all to use his native language. Especially, it would seem, in the
case of less educated persons, this may result, after a time, in
wholesale forgetting: people of this kind understand their native
language when they chance to hear it spoken, but can no longer
speak it freely or even intelligibly. They have made a shift of
; their only medium of communication is now English,
and it is for them not a native but an adopted language. Sometimes
these persons have nevertheless acquired English very imperfectly
and therefore are in the position of speaking no language

Another, more common case of shift of language occurs in the
children of immigrants. Very often the parents speak their native
language at home, and make it the native language of their children,
but the children, as soon as they begin to play out of doors
or to attend school, refuse to speak the home language, and in
time succeed in forgetting all but a smattering of it, and speak
only English. For them, English has become what we may call
their adult language. In general, they speak it perfectly — that is,
in a manner indistinguishable from that of the surrounding native
speakers — but in some cases they carry over foreign peculiarities
from their native language. This latter they speak very imperfectly
or not at all, but their passive understanding, when they hear it,
is somewhat better. A study of similar cases in Wales, where the
children of Welsh-speaking parents shift to English, seems to show
that this process retards the child's development.

3. 10. In the extreme case of foreign-language learning the
speaker becomes so proficient as to be indistinguishable from the
native speakers round him. This happens occasionally in adult
shifts of language and frequently in the childhood shift just
described. In the cases where this perfect foreign-language learning
55is not accompanied by loss of the native language, it results in
bilingualism, native-like control of two languages. After early
childhood few people have enough muscular and nervous freedom
or enough opportunity and leisure to reach perfection in a foreign
language; yet bilingualism of this kind is commoner than one might
suppose, both in cases like those of our immigrants and as a result
of travel, foreign study, or similar association. Of course, one
cannot define a degree of perfection at which a good foreign
speaker becomes a bilingual: the distinction is relative.

More commonly the bilingual acquires his second language in
early childhood. This happens frequently in communities near a
language border, or where a family lives as a speech-island, or
where the parents are of different speech. Many well-to-do European
families make their children bilingual by employing foreign
nurses or governesses. The educated Swiss-German is bilingual
in the sense that he speaks both the local dialect and the highly
divergent standard German. In the United States, better-educated
immigrants often succeed in making their children bilingual; this
development contrasts with the shifting of language among less
privileged groups. In all these cases, apparently, the two languages
play somewhat different parts in the life of the bilingual. Ordinarily
one language is the home language, while the other serves a
wider range, but other dispositions also occur. The apparent
frequency with which one meets bilinguals among artists and men
of science may indicate a favorable effect of bilingualism on the
general development of the child; on the other hand, it may mean
merely that bilingualism results from generally favorable childhood

Chapter 4
The Languages of the World

4. 1. Among the languages that are spoken today, only few
are even tolerably well known to science. Of many we have inadequate
information, of others none at all. The older stages of
some present-day languages, and some languages no longer spoken
are known to us from written records; these records, however,
acquaint us with only an infinitesimal part of the speech-forms
of the past. Some extinct languages are known from the scantiest
of records, such as a few proper names, many more only by the
name of the people who spoke them, and doubtless a vastly greater
number has disappeared without a trace. More than one language
now spoken, especially in Africa and in South America, will pass
out of existence without being recorded.

The inadequacy of our knowledge makes it impossible to determine
the relationships that may exist between many languages.
In general, students who deal with slightly-known languages, have
a weakness for setting up relationships on insufficient evidence.
By relationship of languages we mean, of course, resemblances
that can be explained only on the assumption that the languages
are divergent forms of a single older language. Such resemblances
show themselves in phonetic correspondences like those cited in
Chapter 1, correspondences which can be determined only on the
basis of extensive and accurate data. The less known the languages
and the less expert the student, the greater is the danger
of his making false assumptions of kinship. Even the most positive
announcements often turn out, upon examination, to be based
upon insufficient evidence.

4. 2. English is spoken by more native speakers than any other
language except, presumably, North Chinese; if we count the
important factor of foreign speakers, English is the most widespread
of languages. The number of native speakers of English
was estimated for 1920 at about 170 millions (§ 3.2). Almost all of
these speakers use standard or sub-standard English; local dialects
are of small extent and for the most part mutually intelligible.57

English is unmistakably related to the other Germanic languages,
but at the same time differs plainly from all of them.
History tells us that it came to Britain as the language of invaders,
the Angles, Saxons, and Jutes, who conquered the island in the
fifth century of our era. The marked difference of English from
the Germanic speech along the continental shore of the North Sea
is explained by the millennium and a half of separation. The oldest
written records of English, dating from the eighth and ninth
centuries, confirm this, for their language closely resembles that
of the oldest records of continental Germanic speech, which date
from about the same time. The splitting off of English is a classical
example of the way in which a dialect area is divided by migration.

The resemblance is closest between English and the dialects
of the Frisian area, spoken by some 350,000 persons on the coast
and coastal islands along the North Sea. This resemblance appears
strikingly in the oldest Frisian texts, which date from the
second half of the thirteenth century. We conclude that English
is an offshoot of an Anglo-Frisian (or Ingweonic) dialect area,
which must have been fairly extensive before the migration to

Outside of Frisian, the Germanic-speaking area of the European
mainland (excluding Scandinavia) shows no sharp cleavages. The
nearest thing to a break is a heavy bundle of isoglosses running
east and west across Germany: north of the bundle one speaks
p, t, k in words like hope, bite, make; south of it, sounds like f, s, kh,
as in standard German hoffen, beiszen, machen. The speech of the
northern type is known as Low German, that of the southern as
High German; since the various isoglosses do not coincide, the
distinction can be sharply drawn only if one resorts to an arbitrary
definition. This difference appears already in our oldest
records, which date from about the same time as those of English.
Various kinds of evidence show us that the divergence of the
southern type is due to changes which took place in the south
during the fifth and sixth centuries of our era. The Continental
West Germanic
dialects, as they are called in contrast with Anglo-Frisian,
made a vigorous eastward expansion during the Middle
Ages; to the east and southeast of the main area there are many
speech-islands, especially of the High German type, such as Yiddish
in Poland and Russia. Continental West Germanic is spoken
today by over 100 millions of persons. It has developed two great
58standard languages, Dutch-Flemish, which is used in Belgium and
the Netherlands and is based on western coastal dialects of the
Low-German type, and New High German, based on eastern central
dialects of the district that was gained by medieval expansion.

Anglo-Frisian and Continental West Germanic resemble each
other closely enough to be viewed as a West Germanic unit, in contrast
with the smaller Scandinavian (or North Germanic) group.
Within this group, Icelandic differs markedly from the rest, what
with the thousand years of separation since Iceland was colonized
from western Norway. Icelandic is spoken today by some 100,000
speakers. The language of the Faroese Islands, with about 23,000
speakers, is close to Icelandic. The rest of the area, comprising
Denmark, Norway, Sweden, Gotland, and part of the Finnish
coast, shows no marked cleavages; the speakers number some
15 millions. Our oldest records of North Germanic speech are inscriptions,
some of which may date as early as the fourth century
A.D.; the oldest manuscripts date from the twelfth century, but
the wording of the texts, especially in the case of some Icelandic
literature, may be several centuries older. The present-day standard
languages are Icelandic, Danish, Dano-Norwegian, Norwegian
Landsmaal, and Swedish.

We have some information about Germanic languages that are
no longer spoken, such as the languages of the Goths, Vandals,
Burgundians, and Lombards. Parts of a Bible translation in the
Gothic language of the Visigoths, made by Bishop Ulfila in the
fourth century, are preserved to us in sixth-century manuscripts,
notably the Silver Codex. While the language of the Lombards
seems to have been of the West Germanic type, the others, including
Gothic, were closer to Scandinavian and are usually set
apart as an East Germanic group. East Germanic settlers seem
to have kept their language in the Crimea and elsewhere on the
Black Sea until the eighteenth century.

All the languages so far named resemble each other closely in
contrast with all others, and accordingly constitute the Germanic
family of languages; they are divergent modern forms of a single
prehistoric language to which we give the name Primitive Germanic
(§ 1.6).

4. 3. The kinship of the Germanic family, as a whole, with
certain other languages and language families of Europe and
Asia, is not superficially apparent, but has been fully established
59by the researches of the last century; together, all these languages
make up the Indo-European family (§ 1.6).

To the west of the Germanic languages we find today the remnants
of the Celtic family. Irish is known to us from a manuscript
literature since the eighth century of our era; a few inscriptions
on stone are perhaps much earlier. Irish is spoken by some 400,000
people, and its offshoot, Scotch Gaelic, by some 150,000; Manx,
as a home language, alongside English, by a few hundred. Another
branch of the Celtic family consists of Welsh and Breton, each
with about a million speakers and known through written records
since the eighth century. The latter, spoken on the northwestern
coast of France, was brought there from Britain, perhaps as early
as the fourth century. Another language of this branch, Cornish,
whose earliest records date from the ninth century, died out round
the year 1800. History and the evidence of place-names show that
Celtic was in earlier times spoken over a large part of Europe,
including what is now Bohemia, Austria, southern Germany,
northern Italy, and France. It was superseded in these regions by
Latin, as a result of Roman conquests, and by Germanic languages,
as a result of the great migrations in the early centuries of our era.
We have a few scant inscriptions, dating from round 100 B.C. in
the ancient Celtic language of Gaul.

Northeast of the Germanic languages lies the Baltic family.
The two surviving languages of this family, Lithuanian, spoken by
some 2½ million people, and Lettish, spoken by some 1½ millions,
have written records dating from the sixteenth century; thanks
to the political independence of Lithuania and Latvia, both of
these dialect-groups are now developing vigorous standard languages.
A third language of this group, Old Prussian, is known
to us from a few written documents of the fifteenth and sixteenth
centuries; it ceased to be spoken in the seventeenth century.

South of the Baltic languages, and east and southeast of the
Germanic, we find the great Slavic family. The eastward expansion
of German in the Middle Ages overlaid various languages of the
West Slavic branch. One of these, Lusatian (Wendish, Sorbian),
survives as a speech-island of some 30,000 persons in Upper
Saxony; another, Polabian, survived into the eighteenth century
and has left a few written texts; the rest have died out, leaving a
trace only in Germanized place-names. As a result of the struggle,
the two great surviving West Slavic dialect areas show a peculiar
60geographic configuration: a narrow streak of speech-islands trails
off northward from the main Polish area along the Vistula toward
Danzig, and Bohemian juts out westward as a kind of peninsula
into the domain of German. Polish, recorded since the fourteenth
century, is spoken by more than 20 million people. The Bohemian
area, divided on the basis of standard languages, into Czech and
Slovak, comprises perhaps 12 millions of speakers; the oldest
records date from the thirteenth century. East Slavic consists of
but one enormous dialect area, Russian, with at least 110 million
speakers, and written records dating back to the twelfth century.
The South Slavic branch is separated from the others by the intervention
of Hungarian, an unrelated intruder. It consists of Bulgarian,
with some 5 million speakers, Serbo-Croatian, with some
10 millions, and Slovene, with about 1½ millions. Our oldest written
records of Slavic speech are Old Bulgarian records from the ninth
century, preserved in manuscripts written at least a century later,
and a scant tenth-century text in Old Slovene. Some students
find a relatively close resemblance between the Baltic and Slavic
groups, and include them together as a Balto-Slavic sub-group
within the Indo-European family.

To the south of the Germanic languages, Romance languages
are spoken: the Portuguese-Spanish-Catalan area (with three
standard languages indicated by these names) comprising in all
over 100 million speakers, the French area with 45 millions, the
Italian with over 40 millions, and Ladin (Rhaeto-Romanic) in
Switzerland, spoken by some 16,000 persons. A further group,
the Dalmatian, is extinct: one of the dialects, Ragusan, died out
in the fifteenth century; another, Veliote, survived into the nineteenth.
To the east, on the Black Sea, cut off from the western
areas by the intrusion of South Slavic, lies the Roumanian area,
estimated as having 14 millions of speakers. All the Romance
languages, of course, are modern forms of Latin, the ancient dialect
of the city of Rome. Our oldest records of Latin date from somewhere
round 300 B.C. In medieval and modern time, Latin has
been used as an artificial medium for writing and learned discourse.
Ancient inscriptions show us, in Italy, some sister languages of
Latin, notably Oscan and Umbrian; these and others, which in
the course of Roman expansion were superseded by Latin, belong,
together with Latin, into the Italic family. Some scholars believe
that Italic and Celtic are connected by special resemblances, so
61as to form an Italo-Celtic sub-group within the Indo-European

East of the Adriatic, south of Serbo-Croatian, is the Albanese
area. Albanese, known from records only since the seventeenth
century, is spoken by a population of 1½ millions. Although
Albanese is full of loan-words from the surrounding languages, the
native nucleus of its forms shows it to be a separate branch of the
Indo-European stock.

Greek is spoken today by some 7 millions of speakers, in many
local dialects and in a widespread standard language. The modern
dialects are almost entirely descended from the standard language
(the so-called Koiné) which prevailed in the first centuries of
the Christian Era, having superseded the local and provincial
dialects of ancient times. These Ancient Greek dialects are known
to us from many inscriptions, beginning in the seventh century
B.C., from fragments of writing on papyrus, beginning in the fourth
century b.c, and from a copious literature (transmitted, to be sure,
in much later manuscripts), whose oldest compositions, the Homeric
poems, are at least as old as 800 B.C.

In Asia Minor we find one branch of the Indo-European stock,
Armenian, spoken today by 3 or 4 million people; our oldest
written records of Armenian date from the fifth century A.D.

The great Asiatic offshoot of the Indo-European family is the
Indo-Iranian group. This consists of two sub-groups, Iranian and
Indic (or Indo-Aryan), very different today, but in the forms of
our earliest records so similar that we can with certainty view them
as descendants of a Primitive Indo-Iranian parent language.

The principal dialect areas of modern Iranian are Persian
(with a standard language of high prestige, spoken by perhaps
7 or 8 millions of people), the Caspian group, and Kurdish; then,
eastward, the Pamir dialects, Afghan (Pushto), with some 4 million
speakers, and Baluchi; an isolated offshoot, far to the west is
Ossete, in the Caucasus, spoken by some 225,000 persons. Our
oldest records of Iranian are the rock inscriptions, in Old Persian,
of King Darius the Great and his successors (from the sixth to the
fourth centuries B.C.), and the sacred texts, in Avestan, of the
Zoroastrian (Parsi) religion, whose oldest portions may have been
composed as early as 600 b.c, though our manuscripts are quite
modern and contain a text which has undergone serious orthographic
revision. Intermediate stages, except for Persian (Pehlevi),
62are less well known, but early in the present century discoveries of
manuscript fragments in Chinese Turkestan gave us knowledge of
other medieval Iranian languages, which have been identified as
Parthian, Sogdian, and Sakian.

The other sub-branch of Indo-Iranian, Indic, comprises a total
of more than 230 millions of speakers, distributed among a number
of dialect areas which cover the larger part of India and include
such great languages as Marathi (19 millions), Gujerali (10 millions),
Panjabi (16 millions), Rajasthani (13 millions), Western
(38 millions), Eastern Hindi (25 millions), Oriya (10 millions),
Bihari (36 millions), Bengali (50 millions). The language of
the Gipsies (Romani) is an emigrant offshoot of the Paigachi area in
northwestern India. Our oldest written records of Indic speech, the
inscriptions of King Acoka, dating from the third century B.C.,
show us a number of Indic dialects in what is called the Prakrit
(or Middle Indic) stage; Indic languages in the Prakrit stage are
known to us also from later inscriptions and from manuscript
texts; among these last is Pali, the language of the Buddhist
scriptures. An even older stage of Indic speech, the Sanskritic
(or Old Indic) stage, is known to us, strangely enough, from somewhat
later documents. Our oldest texts in this stage are the Vedic
collections of hymns; the original composition of the oldest parts
of the oldest collection, the Rig-Veda, is placed conservatively at
1200 B.C. These hymns form the basic part of the scriptures of the
Brahmin religion. A second, slightly divergent type of Old Indic
speech is known to us from the Brahmana's, the prose, texts of
the Brahmin religion, and from the grammar of Panini (§ 1.5) and
its ancillary works. This language, known as Sanskrit, was spoken
round the fourth century B.C. by the upper class somewhere in
northwestern India. As a standard dialect and later as a literary
and scholastic language, it gradually came into official use all over
Brahmin India; in the inscriptions it appears first round 150 B.C.
and a few centuries later entirely supersedes the dialects of the
Prakrit type; from that time to the present, written according to
the rules of Panini's grammar, it has served as the medium of an
enormous body of artistic and scholarly literature.

Beside the branches so far named, all of which are represented
by languages spoken today, there must have existed at different
times many other offshoots of Primitive Indo-European, come
closely related to surviving branches, others intermediate between
63them, and perhaps still others quite apart. Of some such languages
we have a slight knowledge. Round the Adriatic, the Illyrian
languages were spoken in ancient times: Illyrian, in which we have
only a few proper names, Venetic, known from inscriptions that
date from the fourth to the second centuries B.C., and Messapian
in southern Italy, with inscriptions dating from 450 to 150 B.C.
Of Thracian, in the western part of the Balkan peninsula, we have
only a few names and words and a single inscription (round 400
B.C.) ; it seems to have been closely related to Phrygian, in Asia
Minor, which is known to us from a set of inscriptions dating as
early as the eighth century B.C. and another set from the first
centuries of our era. Macedonian seems to have been closely related
to Greek. Ligurian (round the present Riviera) and Sicilian in
Sicily, may have been close to Italic. Tocharian, in Central Asia,
is known to us from manuscript fragments of the sixth century A.D.,
found in Chinese Turkestan.

Primitive Indo-European, in its turn, must have been related
to other languages; with one exception, however, these have either
died out or else changed so much as to obscure the kinship. The
one exception is Hittite, an ancient language of Asia Minor, known
to us from cuneiform inscriptions that begin round 1400 B.C. This
relationship, though distant, enables us to reconstruct some of the
pre-history of Primitive Indo-European and some features of a
presumable Primitive Indo-Hittite parent language.

4. 4. As the various languages of the Indo-European stock
spread over their present vast territory, they must have obliterated
many unrelated forms of speech. A remnant of such a language is
Basque, spoken today by some half-million people in the western
Pyrenees. Our oldest texts in Basque date from the sixteenth
century. It is the only surviving form of ancient Iberian, once
spoken over southern France and Spain, and known to us from
inscriptions and place-names.

Of other such languages, now extinct, we have only scant information.
In Italy, Etruscan, a totally unrelated neighbor that
exerted a powerful influence on the Latin people, has left us
copious inscriptions, which begin as early as the sixth century B.C.
They are in the Greek alphabet and can be read, but not understood.
The inscriptions in ancient Rhaetian show this language
to have been an offshoot of Etruscan. An inscription of about
600 B.C. on the island of Lemnos and a series of inscriptions of the
64fourth and third centuries B.C., mostly from Sardis in Asia Minor,
show that Etruscan was related to Lemnian and Lydian; the texts
of only the last-named have been interpreted.

From ancient Crete we have several inscriptions in the Greek
alphabet but in an unknown language, two from the fourth century
B.C. and one (from the town of Praisos) somewhat older. From a
much earlier period, round 1500 B.C. we have Cretan inscriptions
partly in picture-writing and partly in a simplified system derived
from this.

From Asia Minor we have copious inscriptions in Lycian, from
the fifth and fourth centuries B.C., and less extensive ones in
Carian, from the seventh century B.C. The former are in a Greek
alphabet and have been partly interpreted; the writing of the
latter may be of the same provenience, but is undeciphered. In
Syria and the adjacent part of Asia Minor copious inscriptions in
picture-writing from about 1000 B.C. to about 550 B.C. have been
attributed to the Hittites, but there is no reason for believing that
these undeciphered inscriptions were made by the same people as
our Hittite cuneiform records (§ 4.3).

Cuneiform inscriptions on rock and clay from the Near East
acquaint us with extinct languages of an older time: Sumerian in
Mesopotamia, from 4000 B.C., Elamitic, in Persia, from 2000 B.C.;
scant records of Cossean, east of Mesopotamia, from 1600 B.C.,
Mitanni, east of Mesopotamia, from round 1400 B.C.; the language
of Van (near Lake Van) from the ninth and eighth centuries B.C.;
and several uninterpreted languages within the Hittite empire in
Asia Minor. Of the other languages represented in records of this
type, we have already mentioned Old Persian and Hittite (§ 4.3),
and shall immediately speak of Babylonian-Assyrian, a Semitic

4. 5. Of the present-day families which border upon Indo-European,
one or more may be distantly akin; the Semitic-Hamitic
and the Finno-Ugrian families seem to show some resemblance
to Indo-European, but, in spite of much effort, no conclusive evidence
has been found.

The Semitic-Hamitic family consists of four branches which
resemble each other but distantly: Semitic, Egyptian, Berber, and

The Semitic branch appears in two offshoots. The eastern, now
extinct, consists of Babylonian-Assyrian, known to us from inscriptions
65on stone and clay in cuneiform writing, from about
2500 B.C. onward; this language was superseded by Aramaic before
the beginning of the Christian Era. The western branch of Semitic
is divided, again, into two main offshoots, a northern and a southern.
The former appears in the Canaanite glosses in cuneiform
tablets found at Tel-el-Amarna, dating round 1400 B.C., and in
the Moabite of the famous inscription of King Mesha, ninth
century B.C. Phoenician, known first from inscriptions of the
ninth century b.c, was spoken not only in Phoenicia, where it
died out before the Christian Era, but also in the Phoenician
colony of Carthage, where it lived some centuries longer. Hebrew
is known from inscriptions of equal age and from the manuscript
tradition of the Old Testament, whose earliest portion may have
been composed by 1000 B.C. It was superseded by Aramaic in the
second century b.c, but remained in written use through the
Middle Ages; of late, there have been attempts to restore it,
artificially, to the status of a spoken language. Aramaic, finally,
consists of a group of dialects, first known from inscriptions of the
eighth century b.c In a tremendous wave of expansion, Aramaic,
in the centuries just before the Christian Era, spread over Syria
and large tracts of Asia, vying with Greek, and replacing many
languages, among them Hebrew and Assyrian. For a millennium
(from round 300 B.C. to round 650 A.D.) it served as the leading
official and written language of the Near East; in the latter capacity
it exercised a great effect upon Asiatic systems of writing.
It was superseded, in its turn, by the spread of Arabic, and is
spoken today in isolated patches by some 200,000 people. The
southern branch of West Semitic is represented by several still
flourishing languages. South Arabic, known from inscriptions
ranging from about 800 B.C. to the sixth century A.D., is still
spoken, in several dialects, along the southern coast of Arabia and
on the island of Sokotra. Arabic, whose earliest record is an inscription
from 328 A.D., owes its expansion, since the seventh
century of our era, to the conquests of the Mohammedan Arabs.
It is spoken today by some 37 millions of people and, beyond this,
has served for centuries as the sacred, literary, and official language
of Islam. Ethiopian, on the east coast of Africa (Abyssinia), is
first known to us from inscriptions beginning with the fourth century
A.D.; the present-day languages of this group are Tigre,
Tigrina, and Amharic.66

The Egyptian, Berber, and Cushite branches of Semitic-Hamitic
are usually included under the name of Hamitic languages.

Egyptian is recorded for us in hieroglyphic inscriptions from
4000 B.C.; the later form of the language, known as Coptic, appears
in a manuscript literature of Christian times. Egyptian died out,
superseded by Arabic, in the seventeenth century.

The Berber branch of Semitic-Hamitic, is known from ancient
times through inscriptions in the Libyan language, from the fourth
century b.c; it is represented today by various languages, such
as Tuareg and Kabyle, which have maintained themselves against
Arabic in northern Africa and are said to total some 6 or 7 million

The fourth branch of Semitic-Hamitic is Cushite, south of
Egypt; it includes a number of languages, among them Somali
and Galla, the latter with some 8 million speakers.

4. 6. South of the Arab and Berber areas of northern Africa, a
broad belt of many languages stretches across the continent from
the Ethiopian and Cushite areas in the east to the Gulf of Guinea
in the west. The languages of this vast belt, spoken by a population
of presumably some 50 millions, are little known. Some
scholars, upon very scant evidence, believe them all to be related;
others connect some of these languages with Hamitic, or some
with Bantu. Among the languages of this region that are more
often named, we may mention Wolof and Ful in Senegal; Grebo,
Ewe, and Yoruba along the Guinea coast; Haussa in the central
region; and in the east, Nuba in a large territory round Khartoum,
south of this, Dinka, and still further south, Masai.

South of this Guinean and Soudanese belt we come upon the
vast Bantu family of languages, which before the European invasion
covered all the rest of Africa except only a southwestern
district. The languages of the Bantu family, totaling some 50 millions
of speakers, are very numerous; among the better known
are Luganda, Swaheli, Kaffir, Zulu, Tebele, Subiya, Herero.

The portion of southwestern Africa that was not Bantu-speaking,
belonged, before the coming of the European, to two unrelated
linguistic areas: the Bushman, with some 50,000 speakers, and
the Hottentot, with some 250,000.

4. 7. Returning to the continent of Eurasia, we find, to the
east of the Indo-European languages and in topographic alternation
with them, the great Finno-Ugrian family. This family
67consists of six major branches. The first is the Finnish-Lapponic.
In the northerly parts of Norway, Sweden, and Finland, some 30,000
people speak Lappish. The other languages of the Finnish-Lapponic
branch form a closer group, the Finnish (or Baltic-Finnish).
The largest language of this type is Finnish, recorded
in a fragmentary way as early as the thirteenth century and in
printed books since 1544; Finnish is native to some 3 million
speakers. Esthonian, with earliest records of about the same dates,
is spoken by about a million people. Both Finnish and Esthonian
have standard languages which are official in the republics of
Finland and Esthonia. The other languages of the Baltic branch,
Carelian, Olonetsian, Ludian, Vepsian, Livonian, Ingrian, and
Votian, are far smaller, and some of them are near extinction.
Four further branches of the Finno-Ugrian stock lie in patches
across the extent of European and Asiatic Russia; they are Mordvine
(a million speakers); Cheremiss (375,000); Permian, consisting
of Votyak (420,000) and Zyrian (258,000), the latter with
written records from the fourteenth century; Ob-Ugrian, consisting
of Ostyak (18,000) and Vogule (5000). The sixth branch of
Finno-Ugrian is Hungarian, brought by invaders at the end of
the ninth century into central Europe. Aside from scattered
words in Latin documents, the oldest written record of Hungarian
dates from the thirteenth century. In a flourishing standard
language and in a number of local dialects Hungarian is spoken
by some 10 million persons.

To the east of the Ostyak area, along the Yenisei River, some
180,000 persons speak languages of the Samoyede family. These
languages are dispersed over a wide area and show great local
diversity. Some investigators believe that Samoyede and Finno-Ugrian
are related.

4. 8. The Turkish (Turco-Tartar or Altaic) family of languages
covers a vast main area, from Asia Minor, conquered, at the end
of the Middle Ages, by the Ottoman Turks, all the way to the
upper reaches of the Yenisei. These languages, with little differentiation,
are spoken by some 39 millions of people; Turkish,
Tartar, Kirgiz, Uzbeg, Azerbaijani are the more familiar language-names.
Our oldest texts are some Siberian inscriptions, dating
from the eighth century A.D., a Turkish-Arabic vocabulary from
the eleventh century, and a Latin-Persian-Turkish vocabulary
from the fourteenth. Separated from the other languages of the
68group, but not very different from them, is Yakut, spoken by
over 200,000 people in northernmost Siberia. Some students
believe that Turco-Tartar is related to the Mongol and Manchu
families; others, on even slighter grounds, claim a relationship
of all these with Finno-Ugrian and Samoyede (in what they call
a Ural-Altaic family).

The Mongol languages lie for the most part east of the Turco-Tartar,
in Mongolia, but, in consequence of the former wandering
and predatory habits of these tribes, scattered communities
are found in various parts of Asia, and even in European Russia.
The total number of speakers is estimated at 3 millions. The oldest
known written record is an inscription from the time of Gengis
Khan, in the thirteenth century.

The Tunguse-Manchu family lies to the north of the Mongol,
dividing Yakut from the rest of the Turco-Tartar area. Tunguse
is spoken by some 70,000 persons dwelling over a relatively large
tract in Siberia. The number of actual speakers of Manchu is
uncertain, since most of the so-called Manchus in China speak only
Chinese; Deny estimates it at well under a million. As a literary
and official language, Manchu has been printed since 1647; the
manuscript tradition goes back to an even earlier date.

The great Indo-Chinese (or Sino-Tibetan) family consists of
three branches. One of these is Chinese, spoken by some 400 millions
of people; it forms really a vast dialect area containing many,
in part mutually unintelligible, dialects or languages. These
have been classified into four main groups: the Mandarin group
(North Chinese, including the language of Peking; Middle Chinese,
including Nanking; West Chinese, in Szechuen), the Central Coastal
group (Shanghai, Ningpo, Hangkow), the Kiangsi group, and the
South Chinese group (Foochow; Amoy-Swatow; Cantonese-Hakka).
Our oldest texts are inscriptions, some of which may date as far
back as 2000 B.C., but since Chinese writing uses a separate symbol
for each word, with little indication of sounds, even an intelligible
document may tell us little or nothing of the language:
our knowledge of Chinese speech, therefore, does not set in before
about 600 A.D. The second branch of Indo-Chinese is the
Tai family, which includes Siamese, spoken by some 7 millions
of people; the oldest record is an inscription from 1293 A.D.
The third branch is Tibeto-Burman, consisting of four groups:
in the Tibetan group, the language of the same name, with records
69reaching back to the ninth century A.D., is the most important;
in the Burmese group, Burmese, with some 8 million speakers,
holds a similar position; the other two groups, Bodo-Naga-Kachin
and Lo-lo, consist of lesser dialects.

The Hyperborean family, in the extreme northeastern corner
of Asia, consists of Chukchee, spoken by some 10,000 persons,
Koryak, with almost as many speakers, and Kamchadal, with

Along the Yenisei River, Yenisei-Ostyak, with some 1000
speakers, and Cottian, probably by this time extinct, form an
independent family.

No relationship has been found for several other languages of
eastern Asia. Gilyak is spoken in the northern part of Sakhalin
Island and round the mouth of the Amur River. Ainu is spoken
by some 20,000 persons in Japan. Japanese has 56 million speakers;
the written records begin in the eighth century. Korean has
17 millions of speakers.

4. 9. Turning southeastward from Europe, we find in the Caucasus
region a great variety of languages. Apart from Ossete,
an Iranian language (§ 4.3), these are generally classed into two
families, North Caucasian and South Caucasian, with between 1
and 2 million speakers in each. The best known of these languages,
Georgian, belongs to the latter group; the written records begin
as early as the tenth century A.D.

In India, south of the Indo-Aryan languages, lies the great
Dravidian family, including, beside many lesser languages, the
great speech-areas (and standard literary languages) of Tamil
(18 millions), Malayalam (6 millions), Canarese (10 millions; oldest
inscriptions from the fifth century A.D.), Telugu (24 millions).
A single Dravidian language, Brahui (with 174,000 speakers) is
spoken, far off from the rest, in the mountains of Baluchistan;
it seems to be a relic of a time when Dravidian occupied a much
wider territory, before the invasion of Indo-Aryan and Iranian

The languages of the Munda family are spoken by 3 millions
of persons in two separate parts of India, namely, on the southern
slope of the Himalayas and round the plateau of Chota Nagpur
in central India.

The Mon-Khmer family lies in patches over southeastern Asia,
including the Nicobar Islands and some districts in the Malay
70Peninsula. Our oldest records are inscriptions in Cambogian,
dating from the seventh century A.D. This family includes at
present one great cultural language, Annamite, spoken by 14
millions of people. Some scholars believe both the Munda and the
Mon-Khmer families to be related to the Malayo-Polynesian
family (forming the so-called Austric family of languages).

The Malayo-Polynesian (or Austronesian) family extends from
the Malay Peninsula across the Pacific to Easter Island. It
consists of four branches. The Malayan (or Indonesian) branch
includes Malay, with some 3 million native speakers and wide use
as a language of commerce and civilization; further, it embraces
the languages of the great islands of the East, such as Formosan,
Javanese (20 millions), Sundanese (6½ millions), Maduran (3 millions),
Balinese (1 million), and the many Philippine languages,
among them Bisaya (2¾ millions) and Tagalog (1½ millions); a distant
offshoot is Malagasy, the language of Madagascar, spoken
by some 3 million people. The second, Melanesian, branch of
Malayo-Polynesian includes many languages of smaller island
groups, such as the languages of the Solomon Islands and Fijian.
The Micronesian branch contains the languages of a smaller tract,
the Gilbert, Marshall, Caroline, and Marianne archipelagos and
the Island of Yap. The fourth, Polynesian branch includes Maori,
the native language of New Zealand, and the languages of the
more easterly Pacific islands, such as Samoan, Tahitian, Hawaiian,
and the language of Easter Island.

The other families of this part of the earth have been little
studied; the Papuan family, on New Guinea and adjacent islands,
and the Australian languages.

4. 10. There remains the American continent.
It is estimated that the territory north of Mexico was inhabited,
before the coming of the white man, by nearly 1.500,000 Indians;
in this same territory the number of speakers of American languages
today cannot be much over a quarter of a million, with
English making ever more rapid encroachment. As the languages
have been insufficiently studied, they can be but tentatively
grouped into families: estimates vary between twenty-five and
fifty entirely unrelated families of languages for the region north
of Mexico. Most of this region is covered by great linguistic
stocks, but some areas, notably the region round Puget Sound
and the coastal district of California, were closely packed with
71small unrelated speech-communities. At least half a dozen linguistic
stocks are known to have died out. Of those that still
exist, we may name a few of the largest. In the far north, the
Eskimo family, ranging from Greenland over Baffinland and Alaska
to the Aleutian Islands, forms a fairly close-knit dialect-group.
The Algonquian family covers the northeastern part of the continent
and includes the languages of eastern and central Canada
(Micmac, Montagnais, Cree), of New England (Penobscot, Massachusetts,
Natick, Narraganset, Mohican, and so on, with Delaware
to the south), and of the Great Lakes region (Ojibwa, Potawatomi,
Menomini, Sauk, Fox, Kickapoo, Peoria, Illinois, Miami, and so
on), as well as a few detached languages in the west: Blackfoot,
Cheyenne, and Arapaho. The Athabascan family covers all but the
coastal fringe of northwestern Canada (Chipewyan, Beaver, Dogrib,
Sarsi, etc.), a number of isolated groups in California (such as
Hupa and Matole), and a third, large area in the south, the Apache
and Navajo languages. The Iroquoian family was spoken in a district
surrounded by Algonquian; it includes, among others, the
Huron (or Wyandot) language, and the languages of the Iroquois
type (Mohawk, Oneida, Onondaga, Cayuga, Seneca, Tuscarora); in
a detached region to the south Cherokee was spoken. The Muskogean
family includes, among other languages, Choctaw, Chickasaw,
Creek, and Seminole. The Siouan family includes many languages,
such as Dakota, Teton, Oglala, Assiniboine, Kansa, Omaha, Osage,
Iowa, Missouri, Winnebago, Mandan, Crow. A Uto-Aztecan family
has been proposed, on the basis of a probable relationship, to
include, as three branches, the Piman family (east of the Gulf of
California), the Shoshonean family (in southern California and
eastward, including Ute, Paiute, Shoshone, Comanche, and Hopi),
and the great Nahuatlan family in Mexico, including Aztec, the
language of an ancient civilization.

The number of speakers of American languages in the rest of
America is uncertain: a recent estimate places the figure for
Mexico alone at 4½ millions and for Peru and Brazil at over 3
millions each, with a total of over 6 millions for Mexico and Central
America and of over 8½ millions for South America. The number
of languages and their relationships are quite unknown; some
twenty or so independent families have been set up for Mexico
and Central America, and round eighty for South America. In the
former region, beside Nahuatlan, we may mention the Mayan
72family in Yucatan as the bearer of an ancient civilization. In
South America, we note, in the northwest, the Arawak and Carib
families, which once prevailed in the West Indies; the Tupi-Guarani,
stretched along the coast of Brazil, the Araucanian in
Chile, and Kechuan, the language of the Inca civilization. Both
the Aztec and the Maya had developed systems of writing; as
both the systems were largely hieroglyphic and have been only in
part deciphered, these records do not give us information about
the older forms of speech.73

Chapter 5
The Phoneme

5. 1. In Chapter 2 we distinguished three successive events in
an act of speech: A, the speaker's situation; B, his utterance of
speech-sound and its impingement on the hearer's ear-drums; and
C, the hearer's response. Of these three types of events, A and C
include all the situations that may prompt a person to speak and
all the actions which a hearer may perform in response; in sum,
A and C make up the world in which we live. On the other hand,
B, the speech-sound, is merely a means which enables us to respond
to situations that would otherwise leave us unaffected, or
to respond more accurately to situations that otherwise might
prompt less useful responses. In principle, the student of language
is concerned only with the actual speech (B); the study of speakers'
situations and hearers' responses (A and C) is equivalent to the
sum total of human knowledge. If we had an accurate knowledge
of every speaker's situation and of every hearer's response — and
this would make us little short of omniscient — we could simply
register these two facts as the meaning (A-C) of any given speech-utterance
(B), and neatly separate our study from all other domains
of knowledge. The fact that speech-utterances themselves
often play a part in the situation of a speaker and in the response
of a hearer, might complicate things, but this difficulty would
not be serious. Linguistics, on this ideal plane, would consist of
two main investigations: phonetics, in which we studied the speech-event
without reference to its meaning, investigating only the
sound-producing movements of the speaker, the sound-waves, and
the action of the hearer's ear-drum, and semantics, in which we
studied the relation of these features to the features of meaning,
showing that a certain type of speech-sound was uttered in certain
types of situations and led the hearer to perform certain types of

Actually, however, our knowledge of the world in which we live
is so imperfect that we can rarely make accurate statements about
the meaning of a speech-form. The situations (A) which lead to
74an utterance, and the hearer's responses (C), include many things
that have not been mastered by science. Even if we knew much
more than we do about the external world, we should still have to
reckon with the predispositions of the speaker and the hearer.
We cannot foretell whether, in a given situation, a person will
speak, or if so, what words he will use, and we cannot foretell
how he will respond to a given speech.

It is true that we are concerned not so much with each individual
as with the whole community. We do not inquire into the minute
nervous processes of a person who utters, say, the word apple,
but content ourselves rather with determining that, by and large,
for all the members of the community, the word apple means a
certain kind of fruit. However, as soon as we try to deal accurately
with this matter, we find that the agreement of the community is
far from perfect, and that every person uses speech-forms in a
unique way.

5. 2. The study of language can be conducted without special
assumptions only so long as we pay no attention to the meaning
of what is spoken. This phase of language study is known as
phonetics (experimental phonetics, laboratory phonetics). The phonetician
can study either the sound-producing movements of the
speaker (physiological phonetics) or the resulting sound-waves
(physical or acoustic phonetics); we have as yet no means for studying
the action of the hearer's ear-drum.

Physiological phonetics begins with inspection. The laryngoscope,
for instance, is a mirror-device which enables an observer to see
another person's (or his own) vocal chords. Like other devices of
the sort, it interferes with normal speech and can serve only for
very limited phases of observation. The x-ray does good service
where its limitations can be overcome; tongue-positions can be
photographed, for instance, if one lays a thin metal strip or chain
along the upper surface of the tongue. Other devices give a transferred
record. For instance, a false palate covered with coloring-matter
is put into the mouth; after the speaker utters a sound,
the places where the tongue has touched the palate are recognizable
by the removal of the coloring-matter. In most devices of
this sort a bulb is attached to some part of the speaker's vocal
organs, say to the adam's-apple; the mechanism transforms the
movement into up-and-down movements of a pen-point which
touches a strip of paper. The strip of paper is kept moving at an
75even rate of speed, so that the up-and-down movement of the
pen-point appears on the paper as a wavy line. This recording
device is called a kymograph. In acoustic phonetics one secures
imprints of the sound-waves. Records of this kind are familiar to
us in the form of phonograph-disks; phoneticians have not yet
succeeded in analyzing most features of such records.

A considerable part of our information about speech-sounds is
due to the methods we have just outlined. However, laboratory
phonetics does not enable us to connect speech-sounds with meanings;
it studies speech-sounds only as muscular movements or as
disturbances in the air, without regard to their use in communication.
On this plane we find that speech-sounds are infinitely
complex and infinitely varied.

Even a short speech is continuous: it consists of an unbroken
succession of movements and sound-waves. No matter into how
many successive parts we break up our record for purposes of
minute study, an even finer analysis is always conceivable. A
speech-utterance is what mathematicians call a continuum; it
can be viewed as consisting of any desired number of successive

Speech-utterances are infinitely varied. Everyday experience
tells us that different persons speak differently, for we can recognize
people by their voices. The phonetician finds that no two
utterances are exactly alike.

Evidently the working of language is due to a resemblance between
successive utterances. Utterances which in ordinary life
we describe as consisting of “the same” speech-forms — say,
successive utterances of the sentence I'm hungry — evidently
contain some constant features of sound-wave, common to all
utterances of this “same” speech-form. Only on this assumption
can we account for our ordinary use of language. The phonetician,
however, cannot make sure of these constant features, as long as
he ignores the meaning of what is said. Suppose, for instance,
that he had records of an utterance which we could identify as
representing the syllable man, spoken on two different pitch-schemes;
If the language of these utterances were English, we
should say that both contained the same speech-form, namely,
the word man, but if the language were Chinese, the two records
might represent two different speech-forms, since in Chinese differences
of pitch-scheme are connected with different meanings:
76the word man with a high rising pitch, for instance means ‘deceive,’
and the word man with a falling pitch means ‘slow.’ As long as
we pay no attention to meanings, we cannot decide whether two
uttered forms are “the same” or “different.” The phonetician
cannot tell us which features are significant for communication
and which features are immaterial. A feature which is significant
in some languages or dialects, may be indifferent in others.

5. 3. The fact that two utterances of the syllable man with
different pitch-schemes are “the same” speech-form in English,
but “different” speech-forms in Chinese, shows us that the working
of language depends upon our habitually and conventionally
discriminating some features of sound and ignoring all others.
The features of sound in any utterance, as they might be recorded
in the laboratory, are the gross acoustic features of this utterance.
Part of the gross acoustic features are indifferent (non-distinctive),
and only a part are connected with meanings and essential to
communication (distinctive). The difference between distinctive
and non-distinctive features of sound lies entirely in the habit of
the speakers. A feature that is distinctive in one language, may
be non-distinctive in another language.

Since we can recognize the distinctive features of an utterance
only when we know the meaning, we cannot identify them on the
plane of pure phonetics. We know that the difference between
the English forms man and men is distinctive, because we know
from ordinary life that these two forms are used under different
circumstances. It is possible that some science other than linguistics
may define this difference in accurate terms, providing
even for the case where we use man for more than one individual
(man wants but little here below). In any case, however, this difference
cannot be recognized by purely phonetic observation: the
difference between the vowel sounds of man and men is in some
languages non-distinctive.

To recognize the distinctive features of a language, we must
leave the ground of pure phonetics and act as though science had
progressed far enough to identify all the situations and responses
that make up the meaning of speech-forms. In the case of our
own language, we trust to our everyday knowledge to tell us
whether speech-forms are “the same” or “different.” Thus, we
find that the word man spoken on various pitch-schemes is in
English still “the same” word, with one and the same meaning,
77but that man and men (or pan and pen) are “different” words,
with different meanings. In the case of a strange language we
have to learn such things by trial and error, or to obtain the meanings
from someone that knows the language.

The study of significant speech-sounds is phonology or practical
. Phonology involves the consideration of meanings.
The meanings of speech-forms could be scientifically defined only
if all branches of science, including, especially, psychology and
physiology, were close to perfection. Until that time, phonology
and, with it, all the semantic phase of language study, rests upon
an assumption, the fundamental assumption of linguistics: we must
assume that in every speech-community some utterances are alike
in form and meaning

5. 4. A moderate amount of experimenting will show that the
significant features of a speech-form are limited in number. In
this respect, the significant features contrast with the gross acoustic
features, which, as we have seen, form a continuous whole and
can be subdivided into any desired number of parts. In order to
recognize the distinctive features of forms in our own language,
we need only determine which features of sound are “different”
for purposes of communication. Suppose, for instance, that we
start with the word pin: a few experiments in saying words out
loud soon reveal the following resemblances and differences:

(1) pin ends with the same sound as fin, sin, tin, but begins
differently; this kind of resemblance is familiar to us because of
our tradition of using end-rime in verse;

(2) pin contains the sound of in, but adds something at the

(3) pin ends with the same sound as man, sun, hen, but the
resemblance is smaller than in (1) and (2);

(4) pin begins with the same sound as pig, pill, pit, but ends
differently ;

(5) pin begins with the same sound as pat, push, peg, but the
resemblance is smaller than in (4);

(6) pin begins and ends like pen, pan, pun, but the middle part
is different;

(7) pin begins and ends differently from dig, fish, mill, but the
middle part is the same.

In this way, we can find forms which partially resemble pin,
by altering any one of three parts of the word. We can alter first
78one and then a second of the three parts and still have a partial
resemblance: if we alter the first part and then the second, we get
a series like pin-tin-tan; if we alter the first part and then the
third, we get a series like pin-tin-tick; if we alter the second part
and then the third, we get a series like pin-pan-pack: and if we
alter all three parts, no resemblance is left, as in pin-tin-tan-tack.

Further experiment fails to reveal any more replaceable parts
in the word pin: we conclude that the distinctive features of this
word are three indivisible units. Each of these units occurs also
in other combinations, but cannot be further analyzed by partial
resemblances: each of the three is a minimum unit of distinctive
sound-feature, a phoneme
. Thus we say that the word pin consists
of three phonemes: the first of these occurs also in pet, pack, push,
and many other words; the second also in fig, hit, miss, and many
other words; the third also in tan, run, hen, and many other words,
In the case of pin our alphabetic writing represents the three
phonemes by three letters, p, i, and n, but our conventions of
writing are a poor guide; in the word thick, for instance, our writing
represents the first phoneme by the two-letter group th and
the third by the two-letter group ck.

A little practice will enable the observer to recognize a phoneme
even when it appears in different parts of words, as pin, apple, mop.
Sometimes our stock of words does not readily bring out the
resemblances and differences. For instance, the word then evidently
consists of three phonemes, but (especially under the influence
of our way of writing) we might question whether the
initial phoneme was or was not the same as in thick; once we hit
upon the pair thigh and thy, or upon mouth and mouthe, we see
that they are different.

5. 5. Among the gross acoustic features of any utterance, then,
certain ones are distinctive, recurring in recognizable and relatively
constant shape in successive utterances. These distinctive
features occur in lumps or bundles, each one of which we call a
phoneme. The speaker has been trained to make sound-producing
movements in such a way that the phoneme-features will be
present in the sound-waves, and he has been trained to respond
only to these features and to ignore the rest of the gross acoustic
mass that reaches his ears.

It would be useless to try to produce the distinctive features in a
pure state, free from non-distinctive accompaniments. For example,
79an English word, as such, has no distinctive pitch-scheme
— the features of pitch which appear in any utterance of it are
non-distinctive — but of course we cannot speak a word like man
without any features of pitch: in any one utterance of it there will
be some pitch-scheme — even, rising, falling, high, middle, low,
and so on. The phonemes of a language are not sounds, but merely
features of sound which the speakers have been trained to produce
and recognize in the current of actual speech-sound — just as
motorists are trained to stop before a red signal, be it an electric
signal-light, a lamp, a flag, or what not, although there is no
disembodied redness apart from these actual signals.

In fact, when we observe closely, especially in a language foreign
to us, we often notice the wide range of non-distinctive features
and the relatively slight consistency of the distinctive features.
The Menomini Indian, in a word like that for ‘water,’ which I
shall here render as nipēw, seems to us to be speaking the middle
consonant sometimes as a p and sometimes as a b. For his language,
the phonemic (that is, essential) feature is merely a closure of the
lips without escape of breath through the nose. Everything else,
including the features by which English distinguishes between p
and b, is non-distinctive. On the other hand, a slight puff of
breath before the consonant, or else a slight catch in the throat —
either of which will probably escape the ear of an English hearer —
would produce in the Menomini language two entirely different
phonemes, each of which contrasts with the plain p-b phoneme.

In the same way, a Chinese observer who had not been forewarned,
would probably have some trouble before he realized that
English words have the same meaning (are “the same”) regardless
of their pitch-scheme.

In part, the non-distinctive features receive a fairly conventional
treatment. When a foreign speaker reproduces the phonemic
values of our language so as to make himself understood, but does
not distribute the non-distinctive features in accordance with our
habit, we say that he speaks our language well enough, but with a
foreign “accent.” In English, for instance, we produce the initial
phonemes of words like pin, tin, kick with a slight puff of breath
(aspiration) after the opening of the closure, but when an s precedes,
as in spin, stick, skin, we usually leave off this puff of breath.
As this difference is not distinctive, a foreign speaker who fails to
reproduce it, is still intelligible, but his speech will seem queer to
80us. Frenchmen are likely to fail in this matter, because in French
the phonemes which resemble our p, t, k are spoken always without
aspiration. On the other hand, an Englishman or American who
speaks French well enough to be understood, is likely still to displease
his hearers by using the aspiration after p, t, k.

Non-distinctive features occur in all manner of distributions. In
most types of American English, the t-phoneme in words like
water or butter is often reduced to an instantaneous touch of the
tongue-tip against the ridge behind the upper gums: in our habit,
the sound so produced suffices to represent the phoneme. In
England this variant is unknown, and is likely to be interpreted as
a variant of the phoneme d, — so that the American may find that
he is not understood when he asks for water.

In the ordinary case, there is a limit to the variability of the non-distinctive
features: the phoneme is kept distinct from all other
phonemes of its language. Thus, we speak the vowel of a word like
pen in a great many ways, but not in any way that belongs to the
vowel of pin, and not in any way that belongs to the vowel of
pan: the three types are kept rigidly apart.

5. 6. The fact that distinctions which are phonemic in one
language or dialect are indifferent in others, and the fact that the
borders between different phonemes differ in different languages
and dialects, appears most clearly when we hear or try to speak a
foreign language or dialect. We have just seen an instance of how
American English may be misunderstood in England. The vowel
of words like fob, bomb, hot is in American English much closer
than in British English to the vowel of words like far, balm, pa;
in some kinds of American English the two sets of words have in
fact the same vowel. The Englishman of the south, moreover, has
lost the r-sound in words like far. A London cabman did not
understand me when I asked to be driven to the Comedy Theatre:
I had forgotten myself and spoken the American form of the first
vowel in comedy, and this the Englishman could take only as a
representative of the vowel phoneme in a word like car — so that
I was really asking for a Carmody Theatre, which does not exist.

When we try to speak a foreign language or dialect, we are likely
to replace its phonemes by the most similar phonemes of our own
language or dialect. Sometimes our native phoneme and the foreign
one overlap, so that part of the time our reproduction is correct,
but part of the time it falls outside the range of the foreign sound.
81Thus, an American who pronounces the French word même
(‘same’) with the vowel of the English word ma'm, will only part
of the time produce a sound which meets the conventional requirements
of the French phoneme; most of the time he will be producing
a sound which differs decidedly from the vowel which the
Frenchman is accustomed to hear.

What saves the situation in such cases is the native's complementary
inaccuracy. When we hear foreign speech-sounds we respond
to them as if they contained the characteristics of some acoustically
similar phoneme of our native language. The discrepancy disturbs
us, and we say that the foreigner speaks indistinctly or with a
strange “accent,” but we do not know where the difference lies.
In our example, accordingly, the Frenchman will mostly understand
the American's pronunciation of même, even when it contains
a vowel sound that would never occur in the Frenchman's
own pronunciation. However, if our rendition deviates too far
from the foreign phoneme, and especially if it comes close to some
other phoneme of the foreign language, we shall be misunderstood;
thus, some varieties of the American's ma'm which he uses for
French même, will be unintelligible because the Frenchman accepts
them as renditions of a different phoneme which occurs, for instance,
in words like lame (‘blade’).

The confusion is more serious when two or three of the foreign
phonemes resemble some one native phoneme of ours. Our infantile
language-learning trains us to ignore differences that are not
phonemic in our language. The English-speaker will not hear any
difference between the Menomini forms a' kāh ‘yes, indeed,’ and
ahkāh ‘kettle,’ and the first part of the word akāhsemen ‘plum.’
In the first of these forms, the phoneme which resembles our k
is preceded by a slight catch in the throat (a glottal stop) which
I have designated here by an apostrophe; in the second, the k
is preceded by a puff of breath (aspiration), which I have designated
by h; in the third form these features are absent. The
English-speaker was trained in childhood not to respond to a
catch in the throat or a slight huskiness before a consonant sound:
if a fellow-speaker occasionally produces such a noise, we pay no
attention to it.

The Menomini, for his part, cannot distinguish differences like
that of our t and d. Words like bad and bat sound alike to him.
This appears, for instance, in the fact that the Menomini have
82translated the word Swede into their language as if it were sweet,
by the term sayēwenet ‘one who is sweet.’ There is a Menomini
phoneme which resembles both our t and d, and doubtless the
Menomini speaker often utters variants of this phoneme which
fall within the range of our t-phoneme, and occasionally variants
which fall within the range of our d-phoneme, but his infantile
training taught him to ignore these differences of sound.

When we try to speak a foreign language, we reproduce, in such
cases, several foreign phonemes by one single phoneme of our own.
The native speaker, in turn, responds to our phoneme as if it
were one of his. Thus, the German hears no difference between
the initial phoneme of tin and that of thin, since both of them
resemble one of his native phonemes. When he speaks English, he
uses this German phoneme. Hearing him, we respond to it as
though it were our t-phoneme; we are right, at any rate, in concluding
that he does not distinguish between tin and thin. In quite
the same way, when the English-speaker hears German, he will
respond to two different phonemes of that language as though
they were identical with the English phoneme that is initial in
words like cat, and he will fail, in consequence, to distinguish between
some words that are quite different in the habits of the

In other cases, the one phoneme which we substitute for several
phonemes of the foreign language, is acoustically intermediate,
and to the native speaker we seem to be interchanging the sounds.
For instance, many Germans (such as Alsatians) have only one
phoneme, of intermediate acoustic quality, in the sphere of our
p and b, and in speaking our language they use this for both of
our phonemes. When they do this in a word like pie, we are struck
by the deviation in the direction of b and respond as though to
the word buy; on the other hand, when they use their intermediate
phoneme in a word like buy, we are struck by the deviation in the
direction of p, and respond as though we had heard pie. Hence
it seems to us (or to a Frenchman) that the German can pronounce
both p and b, but perversely keeps interchanging the two.

The greatest difficulty arises where a language makes significant
use of features that play no such part in our language. An
English-speaker who hears Chinese (or any of quite a few other
languages), will fail to understand or to speak intelligibly, until
he discovers and trains himself to hear and to reproduce the distinctions
83of relative pitch which are significant in every syllable.
He does not respond to them at first, because as an infant he
was trained not to notice the different pitch-schemes which occur
in successive utterances of a word like man; the Chinese infant,
on the other hand, was trained to respond to several types of such

When the foreign language has only one phoneme in a general
acoustic type where our language has more than one, it often
seems to us as if the foreigner were using very different sounds
without a reasonable distinction. Thus, the Menomini's or the
Alsatian's one p-b phoneme will strike our ears now as p and
now as b.

Some persons have an aptitude for hearing and reproducing
foreign speech-sounds; we say that such persons are good imitators
or have a “good ear.” Most other people, if they hear enough of
a foreign language, or if they are carefully instructed, will in time
learn to understand and make themselves understood. Practical
phoneticians sometimes acquire great virtuosity in discriminating
and reproducing all manner of strange sounds. In this, to be
sure, there lies some danger for linguistic work. Having learned
to discriminate many kinds of sounds, the phonetician may turn
to some language, new or familiar, and insist upon recording all
the distinctions he has learned to discriminate, even when in this
language they are non-distinctive and have no bearing whatever.
Thus, having learned, say in the study of Chinese, to hear the
difference between an aspirated p, t, k, (as we usually have it
in words like pin, tin, kick) and a similar sound without aspiration
(as a Frenchman forms it, and as we usually have it in words like
spin, stick, skin), the phonetician may clutter up his record of
English by marking the aspiration wherever he hears it, while
in reality its presence or absence has nothing to do with the meaning
of what is said. The chief objection to this procedure is its
inconsistency. The phonetician's equipment is personal and accidental;
he hears those acoustic features which are discriminated
in the languages he has observed. Even his most “exact” record
is bound to ignore innumerable non-distinctive features of sound;
the ones that appear in it are selected by accidental and personal
factors. There is no objection to a linguist's describing all the
acoustic features that he can hear, provided he does not confuse
these with the phonemic features. He should remember that his
84hearing of non-distinctive features depends upon the accident of
his personal equipment, and that his most elaborate account cannot
remotely approach the value of a mechanical record.

Only two kinds of linguistic records are scientifically relevant.
One is a mechanical record of the gross acoustic features, such as
is produced in the phonetics laboratory. The other is a record
in terms of phonemes, ignoring all features that are not distinctive
in the language. Until our knowledge of acoustics has progressed
far beyond its present state, only the latter kind of record can
be used for any study that takes into consideration the meaning of
what is spoken.

In fact, the laboratory phonetician usually knows, from other
sources, the phonemic character of the speech-sounds he is studying;
he usually formulates his problems not in purely acoustic
terms, but rather in terms which he has borrowed from practical

5. 7. In order to make a record of our observations, we need
a system of written symbols which provides one sign for each
phoneme of the language we are recording. Such a set of symbols
is a phonetic alphabet, and a record of speech in the shape of these
symbols is a phonetic transcription (or, simply, a transcription).

The principle of a symbol for each phoneme is approached by
our traditional alphabetic writing, but our traditional writing does
not carry it out sufficiently for the purposes of linguistic study.
We write sun and son differently, although the phonemes are the
same, but lead (noun) and lead (verb) alike, though the phonemes
are different. The words oh, owe, so, sew, sow, hoe, beau, though all
end with the same phoneme, variously represented in writing; the
words though, bough, through, cough, tough, hiccough end with different
phonemes but are all written with the letters -ough. Our
letter x is superfluous because it represents the same phonemes
as ks (as in tax) or gz (as in examine); our letter c is superfluous
because it represents the same phoneme as k (in cat) or as s (in
cent). Although we have the letter j for the initial phoneme in
jam, we also use the letter g (as in gem) for this same phoneme.
Standard English, as spoken in Chicago, has thirty-two simple
primary phonemes: the twenty-six letters of our alphabet are too
few for a phonetic record. For some phonemes we use combinations
of two letters (digraphs), as th for the initial phoneme in
thin, ch for that in chin, sh for that in shin, and ng for the final
85phoneme in sing. This leads to further inconsistencies: in then
we use th for a different phoneme, and in hothouse for the two
phonemes which are normally represented by the separate letters
t and h; in Thomas the th has the value of the phoneme ordinarily
represented by t. In singer we use ng for a single phoneme,
as in sing, but in finger the letters ng represent this phoneme plus
the phoneme ordinarily represented by the letter g, as in go.
Traditional alphabetic writing is accurate only in the case of a
few languages, such as Spanish, Bohemian, Polish, and Finnish,
where it has been shaped or revised by persons who had worked
out the phonemic system of their language.

5. 8. On account of the imperfections of traditional writing and
the lack of a sufficient number of characters in our (so-called
“Latin”) alphabet, scholars have devised many phonetic alphabets.

Some of these schemes depart entirely from our traditional
habits of writing. Bell's “Visible Speech” is the best-known of
these, chiefly because Henry Sweet (1845-1912) used it. The symbols
of this alphabet are simplified and conventionalized diagrams
of the vocal organs in position for the utterance of the various
phonemes. Visible Speech is hard to write and very costly to

Another system which departs from the historical tradition is
Jespersen's “Analphabetic Notation.” Here every phoneme is
represented by a whole set of symbols which consist of Greek
letters and Arabic numerals, with Latin letters as exponents.
Each Greek letter indicates an organ and each numeral a degree
of opening; thus, a indicates the lips and 0 indicates closure, so
that α0 will appear in the formula for any phoneme during
whose utterance the lips are closed, such as our p, b, and m
phonemes. The formula for the English m phoneme, as in man,
is α0 δ2 ε1, where δ2 means that the back of the palate is lowered,
and ε1 means that the vocal chords are in vibration. The advantages
of this notation are evident, but of course it is not intended
for the recording of whole utterances.

Most phonetic alphabets are modifications of the traditional
alphabet. They supplement the ordinary letters by such devices
as small capitals, letters of the Greek alphabet, distorted forms of
conventional letters, and letters with little marks, diacritical signs,
attached to them (e.g. ā and ä). There are many alphabets of this
86type, such as that of Lepsius, used for African languages; of
Lundell, used for Swedish dialects; of Bremer, used for German
dialects; of the American Anthropological Association, used for
American Indian languages. In this book we shall use the alphabet
of the International Phonetic Association; this alphabet was developed
by Ellis, Sweet, Passy, and Daniel Jones. A crude form
of phonetic alphabet appears in the “keys to pronunciation” of
most dictionaries. Similar devices have grown up in the traditional
writing of some languages, devices such as the two dots over
vowel letters in German writing (ä, ö, ü) or the diacritical marks in
Bohemian writing (č for our ch, š for our sh); the Russian and
Serbian alphabets supplement the Greek alphabet with a number
of extra letters.

In principle, one phonetic alphabet is about as good as another,
since all we need is a few dozen symbols, enough to supply one for
each phoneme of whatever language we are recording. In their
application, however, all phonetic alphabets suffer from serious
drawbacks. When they were invented, the principle of the phoneme
had not been clearly recognized. The inventors meant their alphabets
to be rich and flexible enough to offer a symbol for every
acoustic variety that could be heard in any language. It is evident,
today, that a record of this kind would amount to nothing less than
a mechanical recording of the sound-waves, which would be the
same for no two utterances. In practice, the phonemic principle
somehow slipped in: usually one wrote a symbol for each phoneme,
but these symbols were highly differentiated and cluttered up with
diacritical marks, for the purpose of indicating “exact” acoustic
values. The varieties that were in this way distinguished, were
merely those which phoneticians happened to have noticed.
Henry Sweet devised a relatively simple system, based on the
Latin alphabet, which he called Romic, for use alongside of Visible
Speech. When the phonemic principle became clear to him, he
realized that his Romic notation would still be sufficient if one
greatly simplified it. Accordingly he used a simplified form, with
a symbol for each phoneme, and called it Broad Romic; he still
believed, however, that the more complex form, Narrow Romic,
was somehow “more accurate” and better suited to scientific

Out of Sweet's Romic there has grown the alphabet of the International
Phonetic Association, which consists, accordingly, of
87the Latin symbols, supplemented by a number of artificial letters,
and a few diacritical marks. In a modified form, we shall use it in
this book, placing between square brackets, as is customary, everything
that is printed in phonetic symbols.

5. 9. The principle on which the International Alphabet is
based, is to employ ordinary letters in values approximating the
values they have in some of the chief European languages, and to
supplement these letters by artificial signs or by the use of diacritical
marks whenever the number of phonemes of a type exceeds
the number of ordinary letters. Thus, if a language has one
phoneme of the general type of our t-sound, we symbolize this
phoneme by the ordinary letter [t], regardless of whether this
phoneme is acoustically quite like the English or the French t-sound,
but if the language has two phonemes of this general type,
we can symbolize only one of them by [t], and for the second one
we must resort to the use of a capital [t], or an italic [t], or some
other similar device. If a language has two phonemes of the general
type of our e-sound as in pen, we use the letter [e] for one of them,
and the supplementary symbol [ɛ] for the other, as in pan [pɛn].

These principles, which the International Phonetic Association
formulated as early as 1912, have been neglected even by its
members; most students have failed to break away from the tradition
of the time when the phonemic principle had not yet been
recognized. Thus, we find most writers using queer symbols for
English phonemes because it has been recognized that English
phonemes differ from the most similar types of French phonemes.
For instance, having pre-empted the symbol [o] for the phoneme of
French eau [o] (‘water’), these authors do not use this letter for
recording the English vowel in son, because this English phoneme is
unlike the French phoneme. In this and some other respects, I
shall depart in this book from the usage (but not from the principles)
of the International Phonetic Association.

Where several languages or dialects are under discussion, each
one must be recorded in terms of its own phonemes; the differences,
so far as we are able to state them, may deserve a verbal
description, but must not be allowed to interfere with our symbols.
Thus, even a phonetician who thinks he can describe in accurate
terms the differences between the phonemes of standard English as
spoken in Chicago and as spoken in London, will add nothing to
the value of his statements by using queer symbols for one or the
88other of these two sets of phonemes, and he will only make things
still harder if he uses outlandish symbols for both of them, because
he happens to know that the ordinary letters have been used for
recording the somewhat different phonemes of some other language.

The principle of a single symbol for a single phoneme may be
modified without harm only where no ambiguity can result. It
may be advisable, where no ambiguity can result, to depart from
the strict principle when this saves the use of extra symbols that
might be disturbing to the reader or costly to print. In some languages,
sounds like our [p, t, k] with a slight puff of breath after
them, are distinct from sounds like the French [p, t, k] without this
aspiration; if the language has no phoneme designated by [h], or if
it has such a phoneme but this phoneme never occurs after [p, t,
k], then it is safe and economical to use the compound symbols
[ph, th, kh] for the former type.

5. 10. The matter of recording languages is complicated not
only by the existence of several phonetic alphabets and by inconsistencies
in their application, but also by the frequent use of
two other devices alongside phonetic transcription.

One of these devices is the citation of forms in their traditional
orthography. This is often done where the language in question
uses the Latin alphabet. The author either supposes that his
reader knows the pronunciation, or else, in the case of ancient
languages, he may not care to guess at the pronunciation. Citation
is often helpful to readers who are familiar with the ordinary
orthography; it is only fair, however, to add a transcription, e.g.
French eau [o] ‘water.’ Even in the case of ancient languages it is
often useful to add a guess at the pronunciation, e.g. Old English
geoc [jok] ‘yoke.’ Only in the case of languages like Bohemian or
Finnish, whose traditional orthography is entirely phonetic, can
one dispense with a transcription. In the case of Latin, a citation
with a macron over long vowels is sufficient (e.g. amāre ‘to love’),
since, so far as we know, Latin orthography was phonetic except
that it failed to indicate the distinction between long and short

For languages which use alphabets other than the Latin, citation
is less often employed. It is customary in the case of Greek, less
often of Russian, but is in every way to be deplored. Some luxurious
publications indulge even in Hebrew, Arabic, and Sanskrit type for
citing these languages. The only reasonable exceptions here are
89forms of writing like the Chinese and the ancient Egyptian, whose
symbols, as we shall see, have meaning-values that cannot be
represented in phonetic terms.

For languages which use writing of some form other than the
Latin alphabet, transliteration is often employed instead of transcription.
Transliteration consists in assigning some letter of the
Latin alphabet (or some group of letters or some artificial symbol)
to each character of the original alphabet, and thus reproducing
the traditional orthography in terms of Latin letters. Unfortunately,
different traditions have grown up for transliterating different
languages. Thus, in transliterating Sanskrit, the Latin letter
c is used to represent a Sanskrit letter which seems to have designated
a phoneme much like our initial phoneme in words like
chin, but in transliterating the Slavic alphabet, the letter c is
used to represent a letter which designates a phoneme resembling
our ts combination in hats. For most linguistic purposes it would
be better to use a phonetic transcription.

5. 11. It is not difficult (even aside from the help that is afforded
by our alphabetic writing) to make up a list of the phonemes
of one's language. One need only proceed with a moderate
number of words as we did above with the word pin, to find that
one has identified every phoneme. The number of simple primary
in different languages runs from about fifteen to about
fifty. Standard English, as spoken in Chicago, has thirty-two.
Compound phonemes are combinations of simple phonemes which
act as units so far as meaning and word-structure are concerned.
Thus, the diphthong in a word like buy can be viewed as a combination
of the vowel in far with the phoneme that is initial in
yes. Standard English has eight such combinations.

It is somewhat harder to identify the secondary phonemes. These
are not part of any simple meaningful speech-form taken by itself,
but appear only when two or more are combined into a larger
form, or else when speech-forms are used in certain ways — especially
as sentences. Thus, in English, when we combine several
simple elements of speech into a word of two or more syllables,
we always use a secondary phoneme of stress which consists in
speaking one of these syllables louder than the other or others:
in the word foretell we speak the tell louder than the fore, but in
foresight the fore is louder than the sight. The noun contest has
the stress on the first syllable, the verb contest on the second. Features
90of pitch appear in English as secondary phonemes chiefly
at the end of sentences, as in the contrast between a question (at
four o'clock?
) and an answer (at four o'clock). It is worth noticing
that Chinese, as well as many other languages, uses features of
pitch as primary phonemes. The secondary phonemes are harder
to observe than the primary phonemes, because they occur only
in combinations or in particular uses of simple forms (e.g. John?
in contrast with John).

The principles we have outlined would probably enable anyone
familiar with the use of writing to work out a system of transcribing
his language. In this book the English examples will be
transcribed, unless otherwise indicated, according to the pronunciation
of standard English that prevails in Chicago. This requires
thirty-two symbols for simple primary phonemes and nine
for secondary phonemes.

Primary phonemes

tableau [a] | [ɑ] | [b] | [č] | [d] | [ð] | [e] | [ɛ] | [f] | [g] | [h] | [i] | [j] | [ǰ] | [k] | [l] | [m] | [n] | [ŋ] | [o] | [ɔ] | [p] | [r] | [s] | [š] | [t] | [θ] | [u] | [v] | [w] | [z] | [ž]

Compound primary phonemes

tableau [aj] | [aw] | [ej] | [ij] | [juw] | [ow] | [ɔj] | [uw]

Secondary phonemes

[ˈˈ], placed before primary symbols, loudest stress: That's mine!
[ðɛt s ˈˈmajn!].

[ˈ], placed before primary symbols, ordinary stress: forgiving
[forˈgiviŋ]; I've seen it [aj v ˈsijn it].91

[ˌ], placed before primary symbols, less loud stress: dining-room
[ˈdajniŋ ˌruwm]; Keep it up [ˌkijp it ˈop].

[ˌ], placed under one of the primary symbols [l, m, n, r], a slight
stress which makes this primary phoneme louder than what
precedes and what follows: coral [ˈkɑrl̩], alum [ˈɛlm̩], apron
[ˈejprn̩], pattern [ˈpɛtr̩n]. 113

[.], placed after primary symbols, the falling pitch at the end of
a statement: I've seen it [aj v ˈsijn it.].

[¿], placed after primary symbols, the rising-falling pitch at the
end of a question to be answered by speech-forms other than
yes or no: Who's seen it? [ˈhuw z ˈsijn it¿].

[?], placed after primary symbols, the rising pitch at the end of
a yes-or-no question: Have you seen it? [hɛv juw ˈsijn it?].

[!], placed after primary symbols, the distortion of the pitchscheme
in exclamations: It's on fire! [it s ɑn ˈfajr!], Seven
[ˈsevn̩ o ˈˈklɑk?!].

[,], placed between primary symbols, the pause, often preceded
by rising pitch, that promises continuation of the sentence:
John, the older boy, is away at school [ˈǰɑn, ðij ˈowldr ˈbɔj, iz
eˈwej et ˈskuwl.].92

Chapter 6
Types of Phonemes

6. 1. While the general principles which we surveyed in the last
chapter will enable an observer to analyze the phonetic structure of
his own speech, they yield very little help, at the start, for the
understanding of a strange language. The observer who hears a
strange language, notices those of the gross acoustic features which
represent phonemes in his own language or in other languages he
has studied, but he has no way of knowing whether these features
are significant in the language he is observing. Moreover, he fails
to notice acoustic features which are not significant in his own
language and in the other languages he has studied, but are significant
in the new language. His first attempts at recording contain
irrelevant distinctions, but fail to show essential ones. Even a
mechanical record will not help at this stage, since it would register
the gross acoustic features, but would not tell which ones were
significant. Only by finding out which utterances are alike in
meaning, and which ones are different, can the observer learn to
recognize the phonemic distinctions. So long as the analysis of
meaning remains outside the powers of science, the analysis and
recording of languages will remain an art or a practical skill.

Experience shows that one acquires this skill more easily if one
is forewarned as to the kinds of speech-sounds that are distinctive
in various languages — although it is true that any new language
may show some entirely unforeseen distinction. This information
is most easily acquired if it is put into the form of a rough description
of the actions of the vocal organs. This rough description is
what we mean by the term practical phonetics. After the observer
has found out which of the gross acoustic features are significant
in a language, his description of the significant features can be
illustrated by a mechanical record.

6. 2. We have no special organs for speech; speech-sounds are
produced by the organs that are used in breathing and eating.
Most speech-sounds are produced by interference with the outgoing
breath. Exceptions to this are suction-sounds or clicks. As
93a non-linguistic sign of surprised commiseration (and also as a
signal to urge horses), we sometimes make a click — the novelist
represents it by tut, tut! — with the tongue against the ridge just
back of the upper teeth. As speech-sounds, various clicks, formed
in different parts of the mouth, are used in some African languages.

6. 3. The first interference which the outgoing breath may meet,
is in the larynx. The larynx is a box of cartilage at the head of the
wind-pipe, visible from the outside as the adam's-apple. Within
the larynx, at the right and left, are two shelf-like muscular protuberances,
the vocal chords. The opening between them, through
which the breath passes, is called the glottis. In ordinary breathing
the vocal chords are relaxed and the breath passes freely through
the glottis. At the rear of the larynx, the vocal chords are attached
to two movable cartilaginous hinges, the arytenoids. Thanks to
delicate muscular adjustments, both the vocal chords and the
arytenoids can be set into a number of positions. The extreme
positions are the wide-open position of ordinary breathing and the
firmly closed position which occurs when one holds one's breath
with the mouth wide open. Various languages make use of various
intermediate positions of the glottis.

One of these positions is the position for voicing. In voicing, the
vocal chords are drawn rather tightly together, so that the breath
can get through only from instant to instant. In getting through,
the breath-stream sets the vocal chords into vibration; the frequency
ranges from around eighty to around one-thousand vibrations
per second. These vibrations, communicated to the outer air,
strike our ears as a musical sound, which we call the voice. The
voice does not play a part in all speech-sounds: we distinguish
between voiced and unvoiced (or breathed) speech-sounds. If one
places a finger on the adam's-apple, or, better, if one presses one's
palms tightly over one's ears, and then utters a voiced sound, such
as [v] or [z], the voice will be felt as a trembling or vibration, while
unvoiced sounds, such as [f] or [s] will lack this buzzing accompaniment.
It seems that in every language at least a few phonemes have
lack of voicing among their fixed characteristics. During the
production of most unvoiced sounds the glottis is wide open, as in
ordinary breathing.

Various adjustments enable us to alter the loudness and the
pitch of the voice-sound as well as its quality of resonance. These
last variations, such as the “head register,” “chest register,”
94“muffled sound,” “metallic sound,” and the like, have not been
physiologically analyzed.

Among the positions intermediate between breathing and voicing,
several deserve mention. If the vocal chords are so far separated
that the voice no longer sounds pure, but is accompanied by the
friction-sound of the breath passing through the glottis, we get a
murmur. In English, the unstressed vowels are often spoken with
murmur instead of voice. As a phoneme, the murmur occurs in
Bohemian, where it may be transcribed by the symbol [h], which
is used in the conventional orthography of this language. If the
glottis is still farther opened, the voice ceases and only a friction-sound
remains; this friction-sound characterizes our phoneme [h],
as in hand [hend]. Another intermediate position is the whisper,
in which only the cartilage-glottis — that is, the space between the
arytenoids — is open, but the vocal chords are in contact. In
what we ordinarily call “whispering,” the whisper is substituted
for the voice and the unvoiced sounds are produced as in ordinary

The sound-waves produced by the vibration of the vocal chords
in voicing, are modified by the shape and by the elasticity of the
channel through which they pass before they reach the outer air.
If we compare the vocal chords to the reeds of a wind-instrument,
we may view the mouth, or rather, the whole cavity from the vocal
chords to the lips, including, in some cases the nasal cavity, as a
resonance-chamber. By setting the mouth into various positions,
by cutting off the exit either through the mouth or through the
nose, and by tightening or loosening the muscles of this region, we
vary the configuration of the outgoing sound-waves.

In contrast with musical sound, noises, which consist of irregular
combinations of sound-waves, can be produced by means of the
glottis, the tongue, and the lips. Some voiced sounds, such as
[a, m, l], are purely musical, that is, relatively free from noise,
while others, such as [v, z], consist of a noise plus the musical sound
of voicing. Unvoiced sounds consist merely of noises; examples
are [p, f, s].

6. 4. When the breath leaves the larynx, it passes, in normal
breathing, through the nose. During most speech, however, we
cut off this exit by raising the velum. The velum is the soft, movable
back part of the palate; at the rear it ends in the uvula, the
little lobe that can be seen hanging down in the center of the mouth.
95If one stands before a mirror, breathing quietly through nose and
mouth, and then speaks a clear [a], one can see the raising of the
velum, especially if one watches the uvula. When the velum is
raised, its edge lies against the rear wall of the breath-passage,
cutting off the exit of the breath through the nose. Most sounds of
speech are purely oral; the velum is completely raised and no
breath escapes through the nose. If the velum is not completely
raised, some of the breath escapes through the nose and the
speech-sounds have a peculiar resonance; such sounds are called
nasalized sounds. In English the difference between purely oral
and nasalized sounds is not distinctive; we often nasalize our
vowels before and after the phonemes [m, n, n], and we nasalize
more than usual when we are tired or relaxed. In some languages,
however, nasalized sounds, most commonly vowels, are separate
phonemes, distinct from similar sounds without nasalization. The
usual symbols for nasalization are a small hook under a letter (this
is used in the traditional orthography of Polish), or a tilde over a
letter (Portuguese orthography and International Phonetic Association),
or an exponent [n] after a letter (used in this book,
because easier to print). French has four nasalized vowels as
phonemes, distinct from the corresponding purely oral vowels:
bas [ba] ‘stocking,’ but banc [ban] ‘bench.’

If the velum is not raised and the exit of the breath through the
mouth is in any way cut off, then, as in ordinary breathing, all the
breath escapes through the nose. Phonemes where this is the case
are nasal. In English we have three nasals: [m], in which the lips
are closed; [n], in which the tongue is pressed against the gums; and
[ŋ], as in sing [siŋ], in which the back of the tongue is pressed
against the palate. These are purely musical sounds, characterized
by the resonances which the different shapes of the oral-nasal
cavity give to the musical sound of the voice. Some languages,
however, have unvoiced nasals as phonemes; these are audible not
so much by the very slight friction-noise of the breath-stream, as
by the contrast with preceding or following sounds and by the
intervening non-distinctive glide-sounds that are produced while
the vocal organs change their position.

A good test of nasalization is to hold a card horizontally with
one edge pressed against the upper lip and the opposite edge against
a cold pane of glass if one now produces a purely oral sound, such
as [a], the pane will be misty only under the card; if one produces
96a nasalized sound, such as [an], the moisture will appear both above
and below the card; and if one produces a purely nasal sound, such
as [m], the moisture on the pane appears only above the card.

6. 5. We change the shape of the oral cavity by placing the lower
jaw, the tongue, and the lips into various positions, and we affect
the resonance also by tightening or loosening the muscles of the
throat and mouth. By these means every language produces, as
phonemes, a number of musical sounds, such as our [a] in palm
[pam], our [i] in pin [pin], our [u] in put [put], our [r] in rubber
[ˈrobr̩], and so on. In some of these the tongue actually touches
the roof of the mouth, but leaves enough room at one or both sides
for the breath to escape without serious friction-noise; such sounds
are laterals, of the type of our [l], as in little [ˈlitl̩]. In unvoiced
laterals, which occur in Welsh and in many American languages,
the friction-noise of the breath-stream is more audible than in
unvoiced nasals.

We make noises in the mouth by movements of the tongue and
lips. If we place these organs (or the glottis) so as to leave a very
narrow passage, the outgoing breath produces a friction-noise:
phonemes characterized by this noise are spirants (fricatives). They
may be unvoiced, as are our [f] and [s], or voiced, like our [v] and
[z]. Since the amount of friction can be varied to any degree, there
is no real boundary between spirants and musical sounds such as
[i] or [l]; especially the voiced varieties occur in different languages
with many degrees of closure.

If we place the tongue or the lips (or the glottis) so as to leave
no exit, and allow the breath to accumulate behind the closure, and
then suddenly open the closure, the breath will come out with a
slight pop or explosion; sounds formed in this way are stops
(plosives, explosives), like our unvoiced [p, t, k] and our voiced
[b, d, g]. The characteristic feature of a stop is usually the explosion,
but the making of the closure (the implosion) or even the
brief period of time during closure, may suffice to characterize
the phoneme; thus, in English we sometimes leave off the explosion
of a final [p, t, k]. These varieties are audible by contrast with
what precedes or follows (as a sudden stoppage of sound or as a
moment of silence), or else through the transitional sounds during
the movement of tongue or lips; also, during the closure of a voiced
stop one can hear the muffled sound of the voice.

Since lips, tongue, and uvula are elastic, they can be placed so
97that the breath sets them into vibration, with alternate moments of
contact and opening. Such trills occur in many languages; an
example is the British English “rolled r,” as in red or horrid.

We shall take up the chief types of phonemes in the following


musical sounds:

6. 6. Stops occur as phonemes in perhaps every language.
English distinguishes three types as to position: labial (more
exactly, bilabial), in which the two lips form the closure [p, b];
dental (more exactly, alveolar, or better gingival), in which the
tip of the tongue makes closure against the ridge just back of
the upper gums [t, d]; and velar (in older writings mis-called
guttural), in which the back of the tongue is pressed against the
velum [k, g].

These last two types occur in many varieties, thanks to the
mobility of the tongue. Contact can be made by the tip of the
tongue (apical articulation) or by a larger area, the blade, round
the tip (coronal articulation); it can be made against the edges of
the upper teeth (interdental position), against the backs of the
upper teeth (postdental position), against the ridge back of the
upper teeth (gingival position), or against points still higher up on
the palate (cerebral or cacuminal or, better, inverted or domal
position). Thus, apical articulation in the domal position (the
tip of the tongue touching almost the highest point in the roof of
the mouth) occurs as a non-distinctive variant alongside the
gingival [c, d] in American English. In French the nearest
sounds to our [t, d) are pronounced not gingivally but as postdentals
(the tip or blade touching the back of the teeth). In
Sanskrit and in many modern languages of India, postdentals [t, d]
and domals (usually transcribed by a letter with a dot under it,
or by italics, or, as in this book, by small capitals [t, d]) are distinct

Similarly, different parts of the back of the tongue (dorsal
98articulation) may be raised so as to touch different parts of the
palate; one distinguishes, usually, between anterior or palatal
position and posterior or velar position, and, still farther back,
uvular position. In English the velars [k, g] are closed farther
forward before some sounds, as in kin, give, and farther backward
before others, as in cook, good — both types in contrast with, say,
calm, guard — but these variants are not distinctive. In some
languages, such as Hungarian, there are separate phonemes of the
palatal and velar types, which we distinguish in transcription by
such devices as [c] for the palatal and [k] for the velar unvoiced
stop. In Arabic a velar unvoiced stop [k] and a uvular unvoiced
stop [q] are distinct phonemes.

A glottal or laryngal stop is produced by bringing the vocal
chords tightly together and then letting them spring apart under
the pressure of the breath. We sometimes produce this sound
before an initial stressed vowel when speaking under a strain, and
in German this is the normal usage; as a phoneme, the glottal stop
occurs in many languages, as, for instance, in Danish, where there
is a distinctive difference, for example, between hun [hun] ‘she’
and hund [hunʔ] ‘dog.’

As to the manner of forming the closure, aside from the difference
of unvoiced and voiced, the amount of breath-pressure and the
vigor of action in the lips or tongue may be variously graded:
pressure and action are gentle in lenes, vigorous in fortes; in solution-lenes
the opening-up is relatively slow, so as to weaken the
explosion. The unvoiced stops may be followed by a puff of
unvoiced breath (aspiration) or preceded by one (pre-aspiration);
the voiced stops, similarly, may be preceded or followed by unvoiced
breath or by a murmur. The closure may be made simultaneously
in two positions, as in the [gb] stops of some African
languages; many languages have glottalized oral stops, with a
glottal stop occurring simultaneously, or just before, or just after
the opening of the [p, t, k]. In English the unvoiced stops are
aspirated fortes, but other types occur as non-distinctive variants,
notably the unaspirated lenis type after [s], as in spin, stone, skin.
Our voiced stops are lenes; at the beginning or at the end of a word
they are not voiced through their whole duration. In French the
unvoiced stops [p, t, k] are fortes and, as a non-distinctive variant,
may be accompanied by a simultaneous glottal stop, but are never
aspirated; the voiced [b, d, g] are lenes, more fully voiced than in
99English. In North Chinese, aspirated and unaspirated unvoiced
stops are different phonemes, e.g. [pha] versus [pa], and voiced
stops occur only as non-distinctive variants of the latter. Many
South-German dialects distinguish unvoiced unaspirated fortes
and lenes, which we may transcribe by [p, t, k] and [b, d, g]; voiced
variants are not distinctive. Sanskrit had four such types of
stops: unvoiced unaspirated [p], aspirated [ph], and voiced unaspirated
[b], aspirated [bh].

6. 7. The commonest trill is the apical or tongue-tip trill, in
which the tongue-tip vibrates in a few rapid strokes against the
gums; this is the “rolled” r of British English, Italian, Russian,
and many other languages. Bohemian distinguishes two phonemes
of this type, the one accompanied by a strong friction sound. The
uvular trill, in which the uvula vibrates against the uplifted back
of the tongue, occurs in Danish, in the commoner pronunciation of
French, German, and Dutch, and in varieties of English (the
“Northumbrian burr”); in these languages, as well as in Norwegian
and Swedish, the uvular and the tongue-tip trill are geographic
variants of the same phoneme. The phonetic symbol for a
trill is [r]; if a language has more than one trill phoneme, [r] is a
handy character.

If the tongue-tip is allowed to make only a single swing, with one
rapid contact against the gums or palate, we have a tongue-flip.
In the Central-Western type of American English, a voiced gingival
tongue-flip occurs as a non-distinctive variant of [t] in forms like
water, butter, at all; different types of tongue-flip occur in Norwegian
and Swedish dialects.

6. 8. The positions in which spirants are formed in English
differ from those of the stops. In one pair, the labiodentals [f, v],
the breath-stream is forced to pass between the upper teeth and the
lower lip. In the dentals [θ, ð], as in thin [θin], then [ðen], the
blade of the tongue touches the upper teeth. Our gingival spirants
[s, z] are hisses or sibilants: that is, the tongue is constricted,
so as to bulge up at the sides and leave only a narrow channel
along the center, through which the breath is forced sharply
against the gums and teeth, giving a sonorous hiss or buzz. If
we draw the tongue a little ways out of this position — in English
we draw it back — the breath is directed less sharply against the
gums and teeth, and seems to eddy round before finding an exit:
in English these hushes or abnormal sibilants are separate phonemes
100[š, ž], as in shin [šin], vision [ˈvižn̩]. 114 In each of these positions we
have a pair, voiced and unvoiced. Many other varieties occur,
such as bilabial spirants, in which the narrowing is made between
the two lips (an unvoiced variety in Japanese, a voiced in Spanish).
In French the hisses are formed postdentally; to our ears the
Frenchman seems to have a slight lisp. German, which has no
[ž], protrudes the lips for [š], so as to accentuate the eddying sound.
Swedish has a [š] with very wide opening, which sounds queer to
English ears.

English has no dorsal spirants, but they occur in many languages,
in a great variety of positions, including lateral types. German
has an unvoiced palatal spirant, in which the middle of the tongue
is raised against the highest part of the palate; as a non-distinctive
variant of this, it uses a velar type, an unvoiced spirant
in the position of our [k, g, ŋ]. The customary transcription of
German uses two symbols, [ç] for the palatal variety, as in ich
[iç] ‘I,’ and [x] for the velar variety, as in ach [ax] ‘oh,’ but only
one symbol is needed, since the varieties depend upon the preceding
phoneme. Voiced spirants [ɣ] of the same position occur
in some types of German pronunciation as variants of the stop [g];
in Dutch and in modern Greek they occur as separate phonemes.
Uvular spirants occur in Danish as variants of the uvular trill,
in other languages as distinct phonemes.

In English we have an unvoiced glottal spirant, [h] as in hit [hit],
when [hwen], hew [hjuw], in which friction is produced by the passage
of the breath through the slightly opened glottis; Bohemian
has a similar sound in which the friction is accompanied by voice
vibrations (murmur). A further pair of glottal spirants, unvoiced
(“hoarse h”) and voiced (“ayin”), occurs in Arabic; their characteristic
feature is said to be a tightening of the throat-muscles.

As to manner, spirants show perhaps less variety than stops.
Among languages which distinguish two varieties of manner,
French voices its [v, z, ž] more completely than does English. Some
languages have glottalized spirants (preceded, accompanied, or
followed by a glottal stop).

6. 9. The positions of nasals are much like those of stops; in
English [m, n, ŋ] are spoken in the same three positions as the
101stops. On the same principle, French speaks its [n] in postdental
position, like its [t, d]. On the other hand, French has no velar
nasal, but has a palatal nasal, in which the closure is made by
raising the middle of the tongue against the highest part of the
palate, as in signe [siɲ] ‘sign.’ As in the stops, Sanskrit and modern
Indian languages distinguish between a dental [n] and a
domal [ɴ].

6. 10. In English the lateral [l] is apical, in gingival position;
at the end of words we use a non-distinctive variety in which the
middle of the tongue is excessively lowered; contrast less with
well. In German and French the [l] is spoken with the surface of
the tongue more raised; the acoustic impression is quite different;
in French, moreover, the contact is postdental. Italian has a
palatal lateral, distinct from the dental, with the back of the tongue
touching the highest point of the palate but leaving free passage
for the breath at one or both sides: figlio [ˈfiλo] ‘son.’ Some American
languages have a whole series of laterals, with differences of
position, glottalization, or nasalization. Unvoiced laterals, especially
if the contact is extensive, take on a spirant character;
voiced laterals, especially if the point of contact is minute, merge
with vowels; thus, one of the two lateral phonemes of Polish
strikes our ear almost as a [w]. On the other hand, the Central-Western
American English vowel [r], as in red [red], fur [fr], far
[far], is closely akin to a lateral: the tip of the tongue is raised to
domal (inverted) position, but does not quite make a. contact. In
transcription we use the same symbol [r] as for the trill of other
languages; this is convenient, because our sound and the British
English trill in red are geographic variants of the same phoneme.

6. 11. Vowels are modifications of the voice-sound that involve
no closure, friction, or contact of the tongue or lips. They are
ordinarily voiced; some languages, however, distinguish different
voice-qualities, such as muffled vowels, murmured vowels, with
slow vibration of the vocal chords, or whispered vowels, in which
friction between the arytenoids replaces vibration of the vocal
chords. 115102

Every language distinguishes at least several different vowel
phonemes. The differences between these phonemes seem to be
largely differences of tongue-position, and to consist, acoustically,
of differences in the distribution of overtones. Even these principles
are disputed; in what follows I shall state the tongue-positions
according to the generally accepted scheme, which has this merit,
that it agrees with the relations of the vowels that are exhibited in
the phonetic and grammatical systems of many languages. Other
factors that enter into the distinction of vowel phonemes, are the
tenseness and looseness of the tongue and other muscles, and
different positions of the lips, such as protrusion and retraction.

The Central-Western type of American English distinguishes
nine vowel phonemes. One of these, [r], which we have already
discussed, is peculiar in its inverted tongue-position. The other
eight form what we may call a two-four system. As to position,
they occur in pairs; each pair consists of a front vowel, formed by
raising the middle of the tongue toward the highest part of the
palate, and a back vowel, formed by raising the back of the tongue
toward the velum. The four pairs differ as to nearness of the
tongue to the palate; thus we have four degrees of raising: high,
higher mid, lower mid, and low. Instead of the terms high and
low, some writers use close and open. This gives us the following

tableau front back | high i u | higher mid e o | lower mid e d | low a a

Examples: in, inn [in], egg [eg], add [ed], alms [amz], put [put],
up [op], ought [ot], odd [ad]. These phonemes are subject to a good
deal of non-distinctive variation, some of-which depends upon the
surrounding phonemes and will interest us later.

Southern British English has much the same system, but the
distribution of the back-vowel phonemes is different, in that the
degrees of closure of the vowels in words like up and odd are the
reverse of ours: higher mid in odd [od], low in up [ɑp]. However,
there has arisen a convention of transcribing British English, not
by the symbols here indicated in accord with the principles of the
IPA alphabet, but by means of queer symbols which are intended
103to remind the reader, irrevelantly enough, of the difference between
English and French vowel phonemes:

tableau Chicago pronunciation according to IPA principles | British pronunciation according to IPA principles | British pronunciation, actual practitce | inn | egg | add | alms | put | odd | ought | up

The ninth vowel phoneme, which we transcribe for Central-Western
American English by [r], as in bird [br̩d], has no uniform
correspondent in Southern British English or in New-England or
Southern American English. Before vowels, British English has a
tongue-tip trill, which we transcribe by [r], as in red [red]; where
Central-Western American has [r] after vowels, British has merely
a modification (in some cases, a lengthening) of the vowel, which
is indicated by a colon [:], as in part [pɑ:rt], form [fɔ:rm]; where in
Central-Western American the [r] is neither preceded nor followed
by a vowel, British English uses a mixed vowel, intermediate between
front and back positions, which is transcribed by [ə:] or
[ə], as in bird [bə:d] or bitter [ˈbitə].

6. 12. Some Central-Western types of American English lack the
distinction of [a] and [ɑ]. The low vowel of such speakers strikes
my ear as an [a], both in alms and in odd; in their phonemic system,
however, its position is neither “front,” nor “back,” but indifferent,
since this pronunciation has only one low-vowel phoneme. A
similar system, without the eccentric [r] vowel, occurs also in
Italian. We may call this a seven-vowel system:

tableau front | indifferent | back | high | higher mid | lower mid | low

Italian examples are: si [si] ‘yes,’ pesca [ˈpeska] ‘fishing,’ pesca
[ˈpɛska] ‘peach,’ tu [tu] ‘thou,’ pollo [ˈpollo] ‘chicken,’ olla [ˈɔlla]
‘pot,’ ama [ˈama] ‘loves.’104

Some languages have simpler systems, such as the five-vowel
system of Spanish or Russian:

tableau Front | Indifferent | Back | high | mid | low

Spanish examples: si [si] ‘yes,’ pesca [ˈpeska] ‘fishing,’ tu [tu]
‘thou,’ pomo [ˈpomo] ‘apple,’ ama [ˈama] ‘loves.’

Even simpler is the three-vowel system which appears in some
languages, such as Tagalog:

tableau front | indifferent | back | high | low

The fewer the phonemes in a vowel-system, the more room is
there for non-distinctive variation of each phoneme. In Spanish
the mid vowels, for instance, vary, to our ear, between higher and
lower positions, with much the same acoustic qualities as in
Italian, where these differences represent different phonemes. The
Russian vowels are subject to wide variation, which depends
chiefly on the preceding and following phonemes; especially one
variant of the high front vowel, as in [sin] ‘son,’ strikes our ear
very strangely, because in this variant the tongue is drawn back
much farther than in any variant of the English high front vowel.
The three-vowel system of Tagalog, finally, allows each phoneme
a range that seems enormous to our hearing; the variants of the
Tagalog phonemes symbolized above by the characters [i] and
[u], range all the way from positions like those of our high vowels
to positions like those of our lower mid vowels.

6. 13. Different positions of the lips play no part in American
English vowels, except for one minor fact which we shall take up
later. In many languages, however, lip-positions accentuate the
quality of different vowels: the front vowels are supported by
retraction of the lips (drawing back the corners of the mouth), and
the back vowels by protrusion or rounding of the lips. In general,
the higher the vowel, the more pronounced is the action of the
lips. These features appear in most European languages and contribute
to the difference between their and our vowels. Even
here we find decided differences; the Scandinavian languages,
105especially Swedish, round their back vowels more than do the
other European languages: a Swedish [o], as in bo [bo:] ‘to dwell,’
has about the tongue-position of a German or French [o], as in
German so [zo:] ‘thus’ or French beau [bo] ‘beautiful,’ but it has
the extreme lip-rounding of a German or French high vowel [u], as
in German du [duː] ‘thou’ or French bout [bu] ‘end;’ it strikes us
as a kind of intermediate sound between an [o] and an [u].

The languages just named make use of lip-positions also for the
distinction of phonemes. The commonest distinction of this kind
is that between the ordinary front vowels (with retracted lip-position)
and rounded front vowels, with the lip-position of the
corresponding back vowels. Thus, French, beside eight vowel
phonemes in a distribution like that of American English, has
three rounded front vowels:

tableau front | back (rounded) | unrounded | rounded | high | higher mid | lower mid | low

Examples :

fini [fini] ‘done,’ été [ete] ‘summer,’ lait [\e] ‘milk,’ bat [ba]

rue [ry] ‘street,’ feu [fø] ‘fire,’ peuple [pœpl] ‘people,’

roue [ru] ‘wheel,’ eau [o] ‘water,’ homme [ɔm] ‘man,’ bas [bɑ]

To these are added four nasalized vowels (see above, § 6.4), as
distinct phonemes: pain [pɛn] ‘bread,’ bon [bon] ‘good,’ unn]
‘one,’ banc [bɑn] ‘bench.’ Furthermore, French has a shorter
variety of [œ], which is transcribed [ə], as in cheval [šəval] ‘horse.’

The symbols [y, ø] are taken from the traditional orthography
of Danish; that of German (and of Finnish) uses the symbols u
and 6.

One can learn to produce rounded front vowels by practising
lip-positions before a mirror: after learning to produce front vowels
of the types [i, e, ɛ] with the corners of the mouth drawn back>
and back vowels of the types [u, o, ɔ] with the lips protruded and
rounded, one speaks an [i] and then tries to keep the tongue-position
unchanged while rounding the lips as for an [u]; the result is an
106[y]. In the same way one passes from [e] to [ø] and from [ɛ]
to [œ].

A further distinction is created by the use of unrounded back
, in contrast with rounded. This additional factor produces
in Turkish a three-dimensional vowel system: each vowel phoneme
is either front or back, high or low, rounded or unrounded:

tableau front | back | unrounded | rounded | high | low

6. 14. Another factor in vowel-production is the tense or loose
position of the muscles: to our ears, vowels of the former type
sound clearer and perhaps excessively precise, since the English
vowels are all loose. Some authors use the terms narrow and wide
instead of tense and loose. The most striking characteristic, ;o
our ear, of the French vowels is their tense character. It is relative
tenseness, too, which in addition to lip-action, makes the
Italian vowels very different from those of English, although the
two languages make the same number of distinctions.

Tenseness and looseness are utilized for distinctions of phonemes
in German and Dutch. In German, and, to a lesser extent, in
Dutch, the tense vowels are also of longer duration (a factor
which will concern us later) than the loose. If we indicate tenseness,
combined with greater length, by a colon after the symbol,
we obtain for these languages the following system, with a pair
of phonemes in each position 116:

tableau front | indifferent | back (rounded) | unrounded | rounded | high | mid | low

German examples:

ihn [iːn] ‘him,’ in [in] ‘in,’ Beet [be:t] ‘flower-bed,’ Bett [bet] ‘bed,’

Tür [ty:r] ‘door,’ hübsch [hypš] ‘pretty,’ König [ˈkø:nik] ‘king,’
zwölf [tsvølf] ‘twelve,’

Fusz [fuːs] ‘foot,’ Flusz [flus] ‘river,’ hoch [ho:x] ‘high,’ Loch
[lox] ‘hole,’ kam [ka:m] ‘came,’ Kamm [kam] ‘comb.’

The differences between the vowel phonemes of different languages
107are not sufficiently understood. It is likely, moreover,
that one and the same phoneme may often be produced, in the
same language, by very different actions of the vocal organs,
but with similar, and for the native hearer identical, acoustic
effects: it is supposed that in such cases the deviation of one organ
(say, a different tongue-position) is compensated by different
action of some other organ (such as a different action of the larynx).108

Chapter 7

7. 1. The typical actions of the vocal organs described in the
last chapter may be viewed as a kind of basis, which may be modified
in various ways. Such modifications are: the length of time
through which a sound is continued; the loudness with which it
is produced; the musical pitch of the voice during its production;
the position of organs not immediately concerned in the characteristic
action; the manner of moving the vocal organs from one
characteristic position to another. This distinction between basic
speech-sounds and modifications is convenient for our exposition,
but it is not always recognized in the phonetic system of languages;
many languages place some of the latter features quite on a par
with phonemes of the former sort. We have seen, for instance,
that features of pitch are utilized as primary phonemes in Chinese,
and features of duration distinguish primary phonemes in German.
On the other hand, most languages do recognize the distinction
to this extent, that they use some of the modifying features as
secondary phonemes — phonemes which are not part of the
simplest linguistic forms, but merely mark combinations or particular
uses of such forms.

7. 2. Duration (or quantity) is the relative length of time through
which the vocal organs are kept in a position. Some languages
distinguish between two or more durations of speech-sounds. Thus,
we have seen (§ 6.14) that in German the tense vowels are longer
than the loose; this difference of length is more striking than that
of tenseness. The sign for a long phoneme is a colon after the
symbol for the sound, as German Beet [be:t] ‘flower-bed,’ in contrast
with Beit [bet] ‘bed.’ If more degrees of length are to be indicated,
a single dot or other signs can be used. Another method
of indicating long quantity is to write the symbol twice; this is
done in Finnish orthography, e.g. kaappi ‘cupboard’ with long
[a] and long [p].

In American English, vowel-quantity is not distinctive. The low
and lower mid vowels, as in pan, palm, pod, pawn, are longer than
109the other vowels, as in pin, pen, pun, pull. All our vowels, moreover,
are longer before voiced sounds than before unvoiced; thus,
the [ɛ] in pan, pad is longer than in pat, pack and the [i] in pin,
bid longer than in pit, bit. These differences are, of course, not
distinctive, since they depend upon the height of the vowel and
upon the following phonemes.

In dealing with matters of quantity, it is often convenient to
set up an arbitrary unit of relative duration, the mora. Thus, if we
say that a short vowel lasts one mora, we may describe the long
vowels of the same language as lasting, say, one and one-half morae
or two morae.

In French, the distinction between long and short vowels works
in a peculiar way. Long vowels occur only before the last consonant
or consonant-group of a word: the mere presence of a long
vowel in French thus indicates that the next consonant or consonant-group
ends a word. In this position, moreover, the length
of a vowel is for the most part determined entirely by the nature
of the phonemes themselves. The nasalized vowels [ɑn, ɛn, on, œn]
and the vowels [o, ø] are in this position always long: tante [tɑnt]
‘aunt,’ faute [fo:t] ‘fault.’ The remaining vowels are always long if
the final consonant is [j, r, v, vr, z, ž], as in cave [ka:v] ‘cellar,’
vert [vɛ:r] ‘green.’ Only in the cases not covered by these two rules,
is the vowel-quantity ever distinctive, as in bete [bɛ:t] ‘beast’
versus bette [bɛt] ‘beet.’

Long consonants occur in English in phrases and compound
words, such as pen-knife [ˈpen ˌnajf] or eat two [ˈijt ˈtuw]; within a
single word [nn] occurs in a variant pronunciation of forms like
meanness [ˈmijnnes] beside [ˈmijnes]. A distinction of two consonant-quantities
within simple words is normal in Italian, as in
fatto [ˈfatto] ‘done,’ but fato [ˈfato] ‘fate,’ in Finnish, and in many
other languages. In Swedish and Norwegian a consonant is long
always and only after a stressed short vowel; the difference of
consonant-quantities, accordingly, is not distinctive. In Dutch
there are no long consonants; even when like consonants meet in a
phrase, only one consonant mora is spoken, so that the phrase
consisting of dat [dat] ‘that’ and tal [tal] ‘number’ is pronounced
[ˈda ˈtal].

7. 3. Stress — that is, intensity or loudness — consists in greater
amplitude of sound-waves, and is produced by means of more
energetic movements, such as pumping more breath, bringing the
110vocal chords closer together for voicing, and using the muscles
more vigorously for oral articulations. In English we have three
secondary phonemes which consist of increased stress, in contrast
with what we may call unstressed passages of phonemes. Our
highest stress [ˈˈ] marks emphatic forms, usually in contrast or
contradiction; our high stress or ordinary stress [ˈ] appears normally
on one syllable of each word; our low stress or secondary stress [ˌ]
appears on one or more syllables of compound words and long
words. In phrases, the high stress of certain words is replaced by a
low stress or entirely omitted. Examples:

This is my parking-place [ˈðis iz ˈˈmaj ˈparkiŋ ˌplejs!]

It isn't my fault and it is your fault [it ˈˈiz nt ˈˈmaj ˈfolt en it
ˈˈiz ˈˈjuwr ˈfolt.]

insert, verb [inˈsr̩t]; noun [ˈinsr̩t]

I'm going out [aj m ˌgowiŋ ˈawt.]

Let's go back [ˈlet s ˌgow ˈbɛk.]

business-man [ˈbiznes ˌmɛn]

gentleman [ˈǰentl̩mn̩]

dominating [ˈdɑmiˌnejtiŋ]

domination [ˌdɑmiˈnejšn̩]

This system is paralleled in all the Germanic languages, and in
many others, such as Italian, Spanish, the Slavic languages, Chinese.
In stress-using languages like these, the stress characterizes
combinations of linguistic forms; the typical case is the use of one
high stress on each word in the phrase, with certain unstressed or
low-stressed words as exceptions. However, some languages of this
type contain simple linguistic forms (such as unanalyzable words)
of more than one syllable, which may be differentiated, accordingly,
by the place of the stress; thus Russian [ˈgorot] ‘city’ and [moˈros]
‘frost’ are both simple words, containing no prefix or suffix; here,
accordingly, the place of stress has the value of a primary phoneme.

Other languages use degrees of loudness as non-distinctive
features. In the Menomini language a sentence sounds, as to ups
and downs of stress, quite like an English sentence, but these ups
and downs are determined entirely by the primary phonemes and
bear no relation to the meaning. In French the distribution of stress
serves only as a kind of gesture: ordinarily the end of a phrase
is louder than the rest; sometimes, in emphatic speech, some other
syllable is especially loud; often enough one hears a long succession
of syllables with very little fluctuation of stress.111

7. 4. Among stress-using languages there are some differences in
the manner of applying stress. In English there is a non-distinctive
variation by which the vowels of unstressed words and syllables
appear in a “weakened” form: they are shorter and formed with
looser muscles, the voice is sometimes reduced to a murmur, and
the tongue-positions tend toward a uniform placing, somewhere
near higher mid position. The degree of weakening varies from
utterance to utterance, and differs a great deal in different geographic
and social types of standard English. Phoneticians often
use special symbols for the weakened vowels, but this is unnecessary,
since the differences are not distinctive, but depend merely
upon the unstressed position. The unstressed vowel is a shorter,
looser, less extremely formed variant of the stressed vowel. Compare
the full [e] of test [ˈtest] with the weakened [e] of contest
[ˈkɑntest]; this weakened [e] appears also, in American English, in
forms like glasses [ˈglɛsez], landed [ˈlɛnded]; in all these cases
British English seems to use a somewhat higher vowel. Similarly,
we may compare the full [o] in seen and unseen [ˈˈonˈsijn] with its
weakened variant in undo [onˈduw]; this weakened [o] appears also
in forms like cautious [ˈkɔšos], parrot [ˈpɛrot].

In other cases the weakened syllables actually show a loss of
phonemes, or substitution of one vowel phoneme for another;
usually various grades of weakening exist side by side:

tableau concert | concerted | address, noun | address, verb | relay | return | vacate | vacation | protest, noun | protest, verb | rebel, noun | rebel, verb | atom | atomic | maintenance | maintain

In cases like these, various grades of weakening exist side by
side and are used according to the speed and the mood (formal,
familiar, and so on) of utterance. There are also local and social
differences. American English says dictionary [ˈdikšn̩ˌejrij], secretary
[ˈsekreˌtejrij] (compare secretarial [ˌsekreˈtejrijl̩]); British
English uses weaker forms, saying [ˈdikšn̩ri, ˈsekritri]. On the
other hand, in forms like Latin [ˈlɛtn̩], Martin [ˈmartn̩] this degree
of weakening is decidedly sub-standard in England, where the
standard forms are [ˈlɛtin, ˈma:tin].112

Not all languages that use stress as a distinctive feature weaken
their unstressed vowels. The Germanic languages other than English
produce the vowels of unstressed syllables quite like those of
stressed syllables. The unstressed vowels in German Monat
[ˈmo:nat] ‘month,’ Kleinod [ˈklajno:t] ‘gem,’ Armut [ˈarmuːt]
‘poverty,’ are quite like the stressed vowels in hat [hat] ‘has,’
Not [no:t] ‘distress,’ Mut [muːt] ‘courage,’ In these languages only
one vowel, the short [e], appears in a weakened variant when it
is unstressed. Thus, in German hatte [ˈhate] ‘had’ or gebadet
[geˈba:det] ‘bathed,’ the [e]-vowel is spoken shorter and with the
tongue less raised and fronted than in a form like Belt [bet] ‘bed,’
and in a form like baden [ˈba:den] ‘to bathe,’ the second syllable is
acoustically quite like the second syllable of an English form like
sodden[ˈsɑ:dn̩], and very different from a German denn [den]
‘then.’ Phoneticians often indicate this weakening by using the
character [ə] for the unstressed form of [e], transcribing hatte
[ˈhatə], baden [ˈba:dən] or [ˈba:dn̩], but this is unnecessary, since
the accent-mark suffices to indicate the weakening.

Other stress-using languages, such as Italian, Spanish, Bohemian,
Polish, do not use special variants for any of the unstressed
vowels; compare, for instance, our restitution [ˌrestiˈtuwšn̩] with an
Italian restituzione [restituˈtsjone]. In a Bohemian word like
kozel [ˈkozel] ‘goat,’ the [e] is just as fully formed as in zelenec
[ˈzelenets] ‘evergreen.’

7. 5. Another difference between stress-using languages concerns
the point at which the increase of loudness sets in. In
English, if the first syllable of a word has a stress, the increase of
loudness begins exactly at the beginning of the word. Accordingly,
there is a difference between pairs like the following:

tableau a name | an aim | that sod | that's odd | that stuff | that's tough

The same habit prevails in German and Scandinavian; German,
in fact, marks the onset of stress so vigorously that it often takes
the shape of a (non-distinctive) glottal stop before the initial
vowel of a stressed word or element, as in ein Arm [ajn ˈarm]
‘an arm,’ or in Verein [fer-ˈajn] ‘association,’ where the ver- is an
unstressed prefix.

In many stress-using languages, on the other hand, the point
of onset of a stress is regulated entirely by the character of the
113primary phonemes. In Dutch, for instance, when there is a single
consonant before the vowel of a stressed syllable, this consonant
always shares in the loudness, regardless of word-division or other
factors of meaning: een aam ‘an aam’ (measure of forty gallons)
and een naam ‘a name’ are both [eˈna:m], and a phrase like het
ander oog
‘the other eye’ is [eˈtandeˈro:x]. The same habit prevails
in Italian, Spanish, and the Slavic languages.

7. 6. Differences of pitch, that is, frequency of vibration in the
musical sound of the voice, are used in English, and perhaps in
most languages, as secondary phonemes. The actual acoustic
forms are highly variable; there is also some geographic variation.
The Englishman's rising pitch in Thank you! is striking to American
ears, and his rising pitch in some statements often makes them
sound to us like a yes-or-no question. Moreover, we use features
of pitch very largely in the manner of gestures, as when we talk
harshly, sneeringly, petulantly, caressingly, cheerfully, and so on.
In English, and in the languages of Europe generally, pitch is the
acoustic feature where gesture-like variations, non-distinctive but
socially effective, border most closely upon genuine linguistic
distinctions. The investigation of socially effective but non-distinctive
patterns in speech, an investigation scarcely begun,
concerns itself, accordingly, to a large extent with pitch. For the
same reason, it is not easy to define the cases where features of
pitch have in our language a genuine status as secondary phonemes.

It is clear that the end of a sentence (a term we shall have to
define later) is always marked by some special distribution of
pitch. We can speak the words It's ten o'clock, I have to go home,
as a single sentence, with a final-pitch only at the end, or as two
sentences, with a final-pitch on clock and another at the end:
It's ten o'clock. I have to go home. After a final-pitch we may
pause for any length of time, or stop talking.

Within the domain of final-pitch we can distinguish several
phonemic differences. It's ten o'clock, as a statement, differs from
It's ten o'clock? as a question; the latter ends with a rise, instead
of a fall. Among questions, there is a difference of pitch-scheme
between a yes-or-no question, such as It's ten o'clock? or Did you
see the show?
and a supplement-question, which is to be answered
by some special word or phrase, as What time is it? or Who saw the
with a lesser rise at the end. In transcription we may indicate
the latter type by placing the question-mark upside down
114[¿]. The distinction appears plainly in the contrast between a
supplement-question and a yes-or-no question which asks whether
this supplement-question is to be answered: Who saw the show?
[ˈhuw ˈsɔðe ˈšow¿] asks for the person, but [ˈhuw ˈsɔ ðe ˈšow?]
means ‘Is this what you were asking about?’

These three types of final-pitch appear side by side in the following
example. If someone said I'm the man whowho —, his
interlocutor might help him out by saying, with the final-pitch
of a statement, Who took the money [huw ˈtuk ðe ˈmonij.]. This
contrasts with the supplement-question Who took the money?
[ˈhuw ˈtuk ðe ˈmonij¿], to which an interlocutor who wanted to
make sure that this was the question, or to use it as a formal
starting-point, might answer by a yes-or-no question, Who took
the money?
[ˈhuw ˈtuk ðe ˈmonij?] (I'll tell you who took it…).

It appears, further, that sentences of all three of these types
may be distorted as to pitch, and also as to stress, when the speaker
is responding to a strong stimulus. We are doubtless justified
in setting up a single secondary phoneme of exclamatory pitch,
symbol [!], for this type, and in supposing that the varieties within
this type, such as the intonations of anger, surprise, call, sneer,
and the like, are non-distinctive, gesture-like variations. The
exclamatory phoneme appears in conjunction with all three of
the final-pitch phonemes. Contrast John [ˈǰɑn.] as an answer to
a question, with John! [ˈǰɑn!] as a call for the hearer's (John's)
presence or attention; similarly John? [ˈǰɑn?] as a simple question
(‘Is that John?’) contrasts with the same question accompanied
by exclamatory pitch: John?! [ˈǰɑn?!] (‘It isn't John, I hope!’);
finally, Who was watching the door [¿] contrasts with the exclamatory
Who was watching the door [¿!] in an emergency or a calamity.

As a fifth secondary phoneme of pitch in English we must recognize
pause-pitch or suspension-pitch [,], which consists of a rise
of pitch before a pause within a sentence. It is used, in contrast
with the final-pitches, to show that the sentence is not ending at
a point where otherwise the phrasal form would make the end of
a sentence possible: I was waiting there [,] when in came the man.
John [,] the idiot [,] missed us. (Contrast: John the Baptist was
) The man [,] who was carrying a bag [,] came up to our
. Only one man is in the story; contrast: The man who was
carrying a bag came up to our door
, which implies that several men
are in the story.115

7. 7. In English both stress and pitch, then, are used only as
secondary phonemes, but there are some differences between the
functions of the two. The stress phonemes step in only when two
or more elements of speech are joined into one form: a simple word,
like John, contains no distinctive feature of stress; to hear a distinctive
feature of stress we must take a phrase or a compound
word or, at least, a word containing two or more parts, such as contest.
The pitch phonemes, on the other hand, occur in every utterance,
appearing even when a single word is uttered, as in John!
John? John. On the other hand, the pitch phonemes in English
are not in principle attached to any particular words or phrases,
but vary, with differences of meaning, in otherwise identical forms.

Many languages differ from English in using secondary phonemes
of pitch as we use those of stress, in words and phrases that consist
of more than one element. In Swedish and Norwegian, a
word of two syllables, for instance, has an ordinary high stress
on one of them, quite as it would in English, but, in addition to
this, the stressed syllables are distinguished by two different
schemes of pitch. The stress may be accompanied by a rising
pitch, giving much the same acoustic impression as an English
high stress, as in Norwegian [ˈbøner] ‘peasants’ or [ˈaksel] ‘shoulder,’
or, with a distinctive difference, it may be accompanied by
a falling pitch, as in [˅bøner] ‘beans’ or [˅aksel] ‘axle.’ This distinctive
word-pitch is all the more remarkable because in all other
respects Swedish and Norwegian closely resemble English in
their use of secondary phonemes of pitch and stress.

The Japanese language is said to distinguish two relative pitches,
normal and higher; thus, [hana] ‘nose’ has normal pitch on both
syllables, [ˈhana] ‘beginning’ has higher pitch on the first syllable,
and [haˈna] ‘flower’ on the second; there seem to be no secondary
phonemes of word-stress.

In still other languages features of pitch are used as primary
phonemes. North Chinese distinguishes four of these, which we
may symbolize by numbers:

[1] high level: [ma1] ‘mother’

[2] high rising: [ma2] ‘hemp’

[3] low rising: [ma3] ‘horse’

[4] low falling: [ma4] ‘scold.’

Cantonese is said to have nine such tones. Primary phonemes of
pitch, in fact, appear in very many languages, either in a few simple
116types, as in Lithuanian, Serbian, and ancient Greek, or in what
seems to us a bewildering variety, as in some African languages.

It is worth noticing that we have in American English a non-distinctive
variation of pitch on our stressed vowels: before an
unvoiced sound, as in map or mat, the pitch-scheme is simple, but
before a voiced sound, as in mad or man, we have ordinarily, and
under loud stress quite clearly, a rising-falling pitch.

7. 8. Once we have obtained some notion of how a phoneme is
formed, we may observe various modifications in the way it is
produced. The English phonemes [k, g], for instance, are made by
closure of the back of the tongue against the velum: if we observe
carefully, we find that the closure is made farther forward when
the next phoneme is a front vowel, as in kin [kin], keen [kijn],
give fgiv], gear [gijr], and farther backward before a back vowel,
as in cook [kuk], coop [kuwp], good [gud], goose [guws], in contrast
with what we may call the normal position, as in car [kar], cry
[kraj], guard [gard], gray [grej]. The English phoneme [h] is formed
with the oral position of the following vowel. These variants are
not distinctive, since they depend entirely upon the following
phoneme. In languages where differences of this sort are distinctive,
we have really no right to call them “modifications,” for in
these languages they are essential features of the phoneme. We
might just as well use the term “modification” of the action or
inaction of the voice during the production of a noise-sound, or of
the presence or absence of nasalization, or of the rounding or retraction
of the lips during the production of a vowel. Nevertheless, it
is convenient to view in this way some less familiar features which
are phonemic in certain languages.

The most important of these is palatalization: during the production
of a consonant the tongue and lips take up, so far as is compatible
with the main features of the phoneme, the position of a
front vowel, such as [i] or [e]. Thus, we may say that in English
[k] and [g] are subject to a non-distinctive palatalization before a
front vowel. Palatalization occurs as a distinctive feature notably
in some of the Slavic languages. In Russian, for instance, most
consonant phonemes occur in pairs, with the distinctive difference
of plain versus palatalized. For the transcription of the latter,
various devices have been used, such as a dot, curve, or caret-sign
over the symbol, or an exponent i or an accent-mark after it, or
the use of italic letters. We shall adopt the last-named device, as
117the most convenient for printing. In a Russian word like [pat]
‘five’ the corners of the mouth are retracted and the tongue is
raised into front-vowel position during the formation of both
consonants. In the case of the [t] this means, of course, that while
the tip and edge of the tongue are making closure against the
backs of the upper teeth, the blade of the tongue is raised toward
the palate; similarly in words like [ˈdada] ‘uncle’ or [ˈnana] ‘nurse.’
The distinctive character of the difference appears in cases like
[bit] ‘way of being,’ [bit] ‘to be,’ [bit] ‘to beat.’

Some languages distinguish velarized consonants, in which the
tongue is retracted as for a back vowel. If the lips are rounded
during the production of a consonant, it is said to be labialized.
These two modifications appear together in labiovelarized consonants.

7. 9. The manner in which the vocal organs pass from inactivity
to the formation of a phoneme, or from the formation of one
phoneme to that of the next, or from the formation of a phoneme
to inactivity, will often show varieties which we label as transitions.
This term is fair enough when the differences are not distinctive,
but when they are distinctive, we have really no right to describe
some of the essential features of the phonemes as basic and others
as transitional.

In passing from silence to a voiced stop, as in bay, day, gay,
we begin the voicing gradually, and in passing from these sounds
to silence, as in ebb, add, egg, we gradually lessen the voicing. This
contrasts with the French manner, where the stops in these positions
are fully voiced, from the very beginning to the very end.
In passing from silence to a stressed vowel, we usually make a
gradual onset of the voice, while the North German first closes the
glottis and then suddenly begins full voicing, so as to produce a
(non-distinctive) glottal stop. Occasionally, as a non-distinctive
variant, we start in the German style and the German in ours. In
French and in sub-standard southern English a third variety of
onset is non-distinctive, in which the glottis passes through the
[h]-position. In standard English and in German this variety is
distinctive, as in English heart [hart] versus art [art]. In passing
from a vowel to silence, the languages so far named use a gentle
off-glide, but others pass through the [h]-position or end sharply
with a glottal stop, and in still others these differences are phonemic.
In passing from an unvoiced stop to a voiced sound,
118especially a vowel, one may begin the voicing at the very moment
of explosion, or the voicing may lag for an instant; in either case it
may begin gently or with a glottal stop; these differences are
phonemic in some languages, and were discussed in § 6.6. Before
or after palatalized consonants there may be a glide resembling a
front vowel; velarized consonants, similarly, may be accompanied
by a back-vowel glide.

In successions of consonants the chief transitional feature seems
to be the difference between close and open transition. In English
we use close transition. When we pass from one stop to another,
we form the second closure before opening the first: in a word like
actor [ˈɛktr̩], for instance, the tip of the tongue touches the gums
for the [t] before the back of the tongue is removed from the velum
to release the [k]. French uses open transition: in a word like
acteur [aktœ:r] ‘actor,’ the [k] is opened before the tongue-tip
touches the teeth for the [t]. Similarly, combinations of stop plus
spirant in English have close transition, as in Betsy, cupful, it shall:
before the stop is opened, the organs are already placed, as far as
possible, into the position of the following spirant, so that the
explosion of the stop is incomplete. This contrasts with the open
transition of French, where the stop is fully exploded before the
spirant begins, as in cette scene [sɛt sɛ:n] ‘this scene,’ etappe facile
[etap fasil] ‘easy stage,’ cette chaise [set ̴šɛ:z] ‘this chair.’ The same
difference appears in so-called double consonants, combinations in
which the same consonant phoneme appears twice in succession.
In English, forms like grab-bag [ˈgrɛb ˌbɛg], hot time [ˈhɑt ˈtajm],
pen-knife [ˈpen ˌnajf] show only one closure for the groups [bb, tt,
nn]; this closure merely lasts longer than the closure of a single
consonant. The double consonant is marked also by the difference
of stress between the implosion (in our examples, weak) and the
explosion (in our examples, strong). In French, similar groups, as
in cette table [sɛt tabl] ‘this table,’ normally show two openings, with
an implosion and an explosion for each of the two consonant units.

If both types of transition occur in a language, the difference
may be utilized as a phonemic distinction. Thus, Polish has mostly
open transition, like that of French, as in trzy [tši] ‘three,’ but the
combination of [t] and [š] occurs also with close transition, as a
separate phoneme, which we may designate by [č], as in czy [či]
‘whether.’ There is also, again as a separate phoneme, a palatalized
variety of this, [č], as in ci [či] ‘to thee.’119

This last example shows us compound phonemes — that is, sounds
resembling a succession of two or more phonemes of the same
language, but in some way distinguished from such a succession,
and utilized as separate phonemes. Many compound phonemes
consist, like those in our example, of a stop plus a spirant or other
open consonant; phonemes of this sort are called affricates. In
English, where all consonant groups have close transition, this
could not be used as a phonemic feature. Nevertheless, English
has two affricate phonemes, [č] as in church [čr̩č], and [ǰ] as in
judge [ǰoǰ]. These affricates are always palatalized, and it is this
feature which distinguishes them from combinations of [t] plus
[š], as in beet-sugar [ˈbijt ˌšugr̩], it shall [it ˈšɛl], and of [d] plus [ž],
as in did Jeanne [did ˈžan]. 117

7. 10. The treatment of successions of vowels and predominantly
musical sounds shows great variety, and many types of
transition are distinctive in one or another language.

In any succession of sounds, some strike the ear more forcibly
than others: differences of sonority play a great part in the transition
effects of vowels and vowel-like sounds. Thus, other things
(especially, the stress) being equal, a low vowel, such as [a], is
more sonorous than a high vowel, such as [i]; any vowel is more
sonorous than a consonant; a nasal, trill, or lateral more than a
stop or spirant; a sibilant [s, z], with its concentration of the breath-stream
into a narrow channel, more than another spirant; a spirant
more than a stop; a voiced sound more than an unvoiced. In any
succession of phonemes there will thus be an up-and-down of
sonority. In a series like [tatatata], the [a]'s will be more sonorous
than the [t]'s. In the following example four degrees of sonority
are distinguished by means of numbers:

Jack caught a red bird
[ǰɛk kɔt e red brd]
314 414 1213 323.

Evidently some of the phonemes are more sonorous than the
phonemes (or the silence) which immediately precede or follow.
This is true of the phonemes marked 1 in our example and, in one
case, of a phoneme marked 2, namely the [r] in bird, but not of the
[r] in red. Any such phoneme is a crest of sonority or a syllabic; the
other phonemes are non-syllabic. Thus the [e] in red and the [r] in
120bird are syllabics, but the [r] in red and the [d] in red and bird are
non-syllabics. An utterance is said to have as many syllables (or
natural syllables) as it has syllabics. The ups and downs of syllabication
play an important part in the phonetic structure of all languages.

In every language, only certain ones of the phonemes ever occur
as syllabics, but in principle any sound may be more sonorous than
its surroundings. The interjections pst! [pst!] and sh! [š!] with which
we demand silence, differ from ordinary English words in using
[s] and [š] as syllabics. Actually, most of the phonemes in any
language are used only as non-syllabics, as, in English, [p, t, k];
we call these consonants. Other phonemes, fewer in number, occur
only as syllabics, as, in English, [e, o, a]; we call these vowels. In
most languages there is a third, intermediate group of sonants,
phonemes which occur in both syllabic and non-syllabic positions;
thus, in American English, of the Central-Western type, [r] is
syllabic in bird [brd], but non-syllabic in red [red].

Whether a sonant in any word is syllabic or non-syllabic, is
determined in different ways in different languages. If the syllabic
or non-syllabic character of a sonant depends entirely upon
the surrounding phonemes (as in bird versus red), then the difference
is not distinctive, and, so far as transcription is concerned,
we do not need more than one symbol. In many cases, however,
the syllabic or non-syllabic character of the sonant is determined
arbitrarily, and constitutes a phonemic difference. Thus, in
stirring [ˈstr̩iŋ] the [r] is syllabic, but in string [striŋ] it is non-syllabic;
in the second syllable of pattern [ˈpɛtr̩n] the [r] is syllabic
and the [n] is non-syllabic, but in the second syllable of patron
[ˈpejtrn̩] the [r] is non-syllabic and the [n] is syllabic. In such cases
we need separate symbols for the two phonemes. Unfortunately,
our habits of transcription in this regard are neither uniform nor
consistent. In a few cases we use different symbols: [i, u, y] are generally
used for syllabic values, and [j, w, ɥ], respectively, for the
corresponding non-syllabics; many transcribers, however, use the
former symbols also for certain non-syllabic occurrences. Another
device is to place a little curve above or below symbols like [i, u, y,
e, o, a] to indicate non-syllabic function. On the other hand, the
symbols [r, l, m, n] usually have a dot, circle, or vertical line placed
under them to denote syllabic function.

When the syllabic or non-syllabic function of a sonant is determined
121by the surrounding phonemes (or silence), the distribution
is natural. Thus, in standard German, the phonemes [i, u] are
non-syllabic when they precede or follow a vowel, and in all other
positions they are syllabic. Non-syllabic [u] occurs only after
[a], as in Haus [haws] ‘house;’ non-syllabic [i] occurs after [a],
as in Ei [aj] ‘egg,’ after [o] (or [ø]), as in neu [noj, nøj] ‘new,’
and before vowels and [u], as in ja [ja:] ‘yes,’ jung [juŋ] ‘young.’
The variants after a vowel are decidedly lowered, and the non-syllabic
[i] before syllabics is spoken with close contact, so as to
give a decided friction-sound, but these differences are not distinctive;
traditionally, transcribers use the symbols [i, u] for the
former type, but [j] for the latter.

Where the syllabic or non-syllabic function of sonants is not
determined by the surrounding phonemes, the difference is phonemic.
Some languages use a slight increase of stress to make
a sonant syllabic. In English this syllabic-stress acts as a secondary
phoneme. In Central-Western American English, syllabic-stress
makes an [r] syllabic in stressed syllables in cases like stirring
[ˈstr̩iŋ], in contrast with string [striŋ], or erring [ˈr̩iŋ] in contrast
with ring [riŋ]. In unstressed syllables, the sonants [r, l, m, n]
are often syllabic by natural distribution, as in butter [ˈbotr̩],
bottle [ˈbɑtl̩], bottom [ˈbɑtm̩], button [ˈbotn̩], but in other cases
their syllabic value is determined by the use of syllabic-stress.
Thus, syllabic-stress marks off an [r] from a preceding [r], as in
error [ˈerr̩] bearer [ˈbejrr̩], or from a preceding [r], as in stirrer
[ˈstr̩r̩], and it often determines which of successive sonants is

tableau apron | pattern | pickerel | minstrel | coral | Carl | char 'em | charm | maintenance | penance

The syllabic-stress may even make [r, l, m, n] syllabic before a
more open phoneme:

tableau battery | pantry | hastily | chastely | anatomy | met me | botany | chutney

The syllabic-stress, then, has in this type of English the value
of a secondary phoneme. If we omitted the little vertical stroke
122under syllabic [r, l, m, n], as we properly should, in all cases where
the syllabic value is due merely to the character of the surrounding
phonemes, this stroke would serve in the remaining cases as a
consistent sign of syllabic-stress.

By the use of syllabic-stress some languages reverse the relations
of natural sonority; thus, South German dialects have the
[i, u, y] syllabic and the [a] non-syllabic in forms like [liab] ‘dear,’
[guat] ‘good,’ [gryan] ‘green.’

Another type of distribution is the use of articulatory differences
to set off the syllabic and non-syllabic functions of the sonants.
Usually this consists in forming the non-syllabic variety with more
closure than the syllabic variety. In English, the sonants [i] and
[u] occur as non-syllabics before and after vowels; symbolizing
these non-syllabic occurrences by [j] and [w], we have [j] in yes
[jes], say [sej], buy [baj], boy [bɔj] and [w] in well [wel], go [gow],
now [naw]. In these examples the non-syllabic function of [j, w]
is sufficiently determined by natural sonority, since a more open
vowel precedes or follows. Therefore the actual variations in the
manner of forming the sounds are here non-distinctive: the [j, w]
after vowels, especially in the types [aj, ɔj, aw] are very open, and
the [a] also is quite different from an ordinary [a]; before a vowel,
as in yes, well, the [j] has a higher and more fronted tongue-position
than a syllabic [i], and the [w] has a higher tongue-position
than a syllabic [u] and is formed with a slight contraction
of the lips. Now, these latter differences are utilized, in English,
as phonemic differences: even where the function is not determined
by natural sonority, we distinguish the closer non-syllabic [j, w]
as separate phonemes, from the more open syllabic [i, u]. Thus,
we distinguish between [uw] in ooze [uwz] and [wu] in wood [wud],
and between [ij] in ease [ijz] and a rare [ji], as in slang yip [jip]
‘to squeal,’ and we have even groups like [jij, wuw], as in yeast
[jijst], woo [wuw]. When two different members of the set,
[i, u, r] come together in a stressed syllable, the first is non-syllabic:
you [juw], yearn [jr̩n], win [win], work [wr̩k], rid [rid],
roof [ruf, ruwf]. In unstressed syllables we have, however, distinctions
like hire [hajr] versus higher [ˈhajr̩], pair [pejr] versus
payer [ˈpejr̩], sore [sowr] versus sower [ˈsowr̩]. A non-syllabic
sonant- which, thanks to some modification, is phonemically
distinct from the corresponding syllabic sonant is called a semivowel.123

In the same way, French produces its high vowels [i, u, y] with
greater closure and tensity when they are non-syllabic, as in
hier [jɛ:r] ‘yesterday,’ oie [wa] ‘goose,’ ail [a:j] ‘garlic,’ huile
[ɥil] ‘oil,’ and treats these types as separate semivowel phonemes,
distinguishing, for instance, between out [wi] ‘yes’ and houille
[uːj] ‘anthracite,’ and employing the sequence [ij], as in fille
[fiːj] ‘daughter.’

7. 11. Vowels and sonants combine into compound phonemes,
which are known as diphthongs, or, if there are three components,
as triphthongs. Whether a succession of phonemes is to be viewed
as a compound phoneme, depends entirely upon the phonetic
structure of the language. In English, successions like [je] in yes
or [we] in well are treated as two phonemes, like any sequence of
consonant plus vowel, but combinations of vowel plus semivowel
are treated as compound phonemes. We have seven
such combinations, as well as one triphthong of semivowel-vowel-semivowel:

tableau see | seeing | say | saying | buy | buying | boy | boyish | do | doing | go | going | bow | bowing | few | fewer

We shall see in the next chapter that in the phonetic structure
of our speech-forms, these groups play the same part as simple
vowel phonemes. The peculiar non-distinctive modifications of
the components, especially of [a, j, w], which we noticed above,
often appear in diphthongs, but this is of secondary importance;
the essential feature is the peculiar structural treatment. Another
non-distinctive peculiarity of our diphthongs is their divergent
sound, in most American types of pronunciation, before [r]: in
this position they approach the character of a single long and
rather tense vowel:

tableau gear | sure | air | oar | fire | hour124.

In some pronunciations these modified varieties differ from
any simple vowel, witness:

tableau mary | wore, hoarse | merry | horse | marry | war

Many types of pronunciation, however, lack some or all of these
differences; in these types either some of the diphthongs or some of
the simple vowels do not occur before [r].

Diphthongs occur also in languages that do not treat syllabic
and non-syllabic vowels as separate phonemes. In German the
combinations [aj] as in Eis [ajs] ‘ice,’ [oj] as in neu [noj] ‘new,’
and [aw], as in Haus [haws] ‘house,’ are treated, structurally, as
unit phonemes. As in English, the constituents differ greatly from
their ordinary form: the non-syllabics have mid-vowel quality
rather than high, and the [oj], especially, exists in several varieties,
resembling, in some pronunciations, rather a combination of
rounded front vowels, say [øɥ].

Diphthongs like the English and German, where the syllabic
part precedes, are called falling diphthongs, in contrast with rising
diphthongs, in which the non-syllabic part precedes. Thus, in
French, combinations like [jɛ], as fier [fjɛ:r] ‘proud,’ and [wa],
as in moi [mwa] ‘I,’ are treated structurally as unit phonemes; in
Italian, the combinations [jɛ, wɔ] are treated as diphthongs; the
same is true of [je, we] in Spanish.

Some languages have compound phonemes of syllabic vowels and
non-syllabic consonants. In Lithuanian the phonemes [l, r, m, n]
are never syllabic, but combinations like [al, ar, am, an] are treated
structurally and accentually as diphthongs, quite on a par with
[aj] or [aw].

7. 12. Since syllabication is a matter of the relative loudness of
phonemes, it can be re-enforced or opposed by adjustments of
stress. The re-enforcing habit prevails probably in most languages.
In French, where stress is not distinctive, every syllable is re-enforced
by a slight increase of stress on its syllabic; if there is
only one non-syllabic before the syllabic, the rise begins on this
non-syllabic; if there are two, different groups are treated differently:
pertinacité [pɛr-ti-na-si-te] ‘pertinacity,’ patronnesse [patrɔ-nɛs]
‘patroness.’ This distribution of minute rises and falls
of stress is non-distinctive, since it is determined entirely by the
125character of the primary phonemes. It gives the language, to our
ears, a rapid, pattering or drumming sound. The same habit
prevails also in many stress-using languages, such as Italian,
Spanish, Polish, Bohemian, and even in Russian, which not only
has distinctive stress, but also weakens the unstressed vowels.
Thus, in Italian pertinacia [per-ti-ˈna-ča] ‘stubbornness’ or patronessa
[pa-tro-ˈnes-sa] ‘patroness,’ the syllables are divided by
ups and downs of stress, which are well-marked in the accented
syllables, and slight in the others.

English and the other Germanic languages do not mark off the
unstressed syllables by ups and downs of stress. In a word like
dimity [ˈdimitij] or patroness [ˈpejtrones], the stress merely drops
off after its high point on the first syllable. Evidently there are
three syllables, because there are three crests of natural sonority,
but it would be impossible to say where one syllable ends and
the next begins. In forms like pertinacity [pr̩tiˈnɛsitij] or procrastination
[proˌkrɛstiˈnejšn̩], the beginnings of the stressed syllables
are plainly marked by the onset of stress, but no other syllable-boundaries
are in any way marked off.

The distribution of stress may create crests of sonority which
are independent of the natural sonority of the phonemes. We
have seen that in English the phonemes [r, l, m, n] may be louder
than the surrounding phonemes, and therefore syllabic, thanks to a
slight increase of stress.

The distribution of stress may even overcome relations of
natural sonority. In a combination like [dzd], the [z] is more
sonorous than the [d]'s, and in [kst] the [s] is more sonorous than the
stops, but in English our single high stress on forms like adzed
[edzd], text [tekst], step [step] is so loud that it drowns out these
small differences of sonority. Some stress-using languages in this
way drown out even the sonority of predominantly musical sounds:
thus, Russian speaks the following, thanks to stress, as one-syllable
words: [lba] ‘of the forehead,’ [rta] ‘of the mouth;’ Polish, similarly
trwa [trva] ‘it lasts,’ msza [mša] ‘mass.’126

Chapter 8
Phonetic Structure

8. 1. Descriptions of speech-sounds like those in the last two
chapters, are due merely to chance observation. These descriptions
are made in terms of a speaker's movements: more refined
physiological observation may show that some of them are wrong.
What is more serious, the differences and varieties that are observed,
such as, say, the difference between French and English
unvoiced stops [p, t, k], are not selected by any fixed principles
(such as acoustic phonetics may some day give us), but owe their
currency to the chance that some observer with a good ear had
heard both of the languages concerned. Just as observation of
South German dialects or of certain American Indian languages
adds to the varieties of unvoiced stops that could be gathered from
standard English and standard French, so the study of almost any
new dialect will increase the repertoire of differences which a
phonetician can hear. The extent of observation is haphazard, its
accuracy doubtful, and the terms in which it is reported are vague.
Practical phonetics is a skill, for the student of languages often a
very useful skill, but it has little scientific value.

For this reason it is beyond our power to analyze the general
acoustic effect of a language. We can explain certain superficial
effects: the “pattering” run of Italian (to English ears) is due to
the syllable-division; the “guttural” sound of Dutch (to our
sense), to the use of a uvular trill (§ 6.7) and of velar spirants
(§ 6.8). In general, however, such observations of the “basis of
articulation” are bound to be vague. English (in contrast, say,
with French or German) retracts the jaw; the Central and Western
type of American English adds a tendency to raise the tip of the
tongue. German and French (in contrast with English) advance the
jaw and use the muscles more vigorously — German in large,
sweeping movements, French in smaller and more precise ones,
especially in the front of the mouth. Danish draws the muscles in
toward the median line. Such observations are often helpful
toward understanding or imitating a pronunciation, but they are
127hazy and inaccurate. We must wait for laboratory phonetics to
give us precise and trustworthy statements.

The important thing about language, however, is not the way
it sounds. The speaker's movement, the disturbance in the air,
and the hearer's ear-drum vibrations (the B of § 2.2) are, in
themselves, of very little moment. The important thing about
language is its service in connecting the speaker's stimulus (A in
§ 2.2) with the hearer's response (C in § 2.2). This connection
depends, as we have seen (§ 5.4), upon only a relatively few features
of the acoustic form, upon the features which we call phonemes.
For the working of language, all that is necessary is that each
phoneme be unmistakably different from all the others. Except
for this differentiation, its range of variety and its acoustic character
are irrelevant. Any language can be replaced, for all its essential
values, by any system of sharply distinct signals, provided that one
signal is made to replace each phoneme of the language. Such a
replacement is made in a correct phonetic transcription — one
which satisfies the demands of accuracy and relevancy by using
one and only one symbol for each phoneme. Imperfectly and yet
sufficiently well for practical purposes, such a replacement is made
in traditional alphabetic writing. The importance of a phoneme,
then, lies not in the actual configuration of its sound-waves, but
merely in the difference between this configuration and the configurations
of all the other phonemes of the same language.

For this reason even a perfected knowledge of acoustics will not,
by itself, give us the phonetic structure of a language. We shall
always have to know which of the gross acoustic features are, by
virtue of meanings, “the same,” and which “different” for the
speakers. The only guide to this is the speaker's situation and the
hearer's response. Any description which fails to discriminate the
distinctive features from the non-distinctive, can tell us little or
nothing about the structure of a language. In this respect, a
mechanical record has at least the virtue of not distorting the
acoustic facts. The “exact” freehand records of zealous phonetic
experts are likely to insist upon irrelevant acoustic differences that
owe their notation merely to the circumstance that the observer
has learned to respond to them. On this basis, it is possible to find
the same set of “sounds” in languages of entirely different phonemic
structure. For instance, both languages might show seven
similar vowel “sounds,” but in Language B these might be seven
128different phonemes, while in Language A [ɛ] and [o] might be non-distinctive
variants of [a], and [e, o] respectively of [i, u]. Both
languages might seem to show two durations of vowels, but these
might be phonemic in Language A (as in German), while in
Language B they might be non-distinctive variants. Both might
show plain and aspirated unvoiced stops, as different phonemes in
Language A and as mere non-distinctive variants in Language B.
Both might have a series of voiced spirants, but these might be
distinctive in Language B, while in Language A they existed
merely as variants of stops between vowels.

Only the phonemes of a language are relevant to its structure
— that is, to the work it does. A description of the non-distinctive
features might be of great interest, but for this it would have to be
more complete and more copious than any that have so far been

8. 2. A list or table of the phonemes of a language should therefore
ignore all non-distinctive features. Such lists or tables are
usually made on the basis of practical-phonetic classifications,

American English (Chicago)

image stops, unvoiced voiced | affricate, unvoiced voiced | spirants, unvoiced voiced | nasals | lateral | inverted | semivowels | vowels, high higher mid lower mid low | secondary phonemes: stress syllabic-stress pitch

Tables like these, even when they exclude non-distinctive features,
are nevertheless irrelevant to the structure of the language,
129because they group the phonemes according to the linguist's notion
of their physiologic character, and not according to the parts which
the several phonemes play in the working of the language. Our
table does not show, for instance, that two of the nasals, [m] and
[n], sometimes serve as syllables in unstressed syllables as in
bottom [ˈbɑtm̩], button [ˈbotn̩], while the third one, [ŋ], does not.
It fails to show that [l] serves as a syllabic in unstressed syllables
only, as in bottle [ˈbɑtl̩], while [r] may serve as a syllabic regardless
of stress, as in learner [ˈlr̩nr̩]. It does not show which vowels and
semivowels combine into compound phonemes. To show these
structural facts, we should need a supplementary table something
like this:

I. Primary phonemes:

A. Consonants, always or sometimes non-syllabic:

1. Mutes, always non-syllabic: [p t k b d g č ǰ f θ s š h v ð z ž ŋ]

2. Sonants, sometimes syllabic:

a. Semi-consonants, syllabicity determined by surroundings
and by syllabic-stress:

(1) Consonantoids, syllabic only in unstressed syllables:
[m n l]

(2) Vocaloid, syllabic also in stressed syllables: [r]

b. Semivowels, syllabicity determined also by manner
of articulation; diphthong-forming:

(1) Non-syllabic: [j w]

(2) Syllabic: [i u]

B. Vowels, always syllabic:

1. Diphthongs and triphthong, compound phonemes:
[ij uw ej ow aj aw ɔj juw]

2. Simple vowels: [e o ɛ ɔ a ɑ]

II. Secondary phonemes:

A. Syllabic-stress, applied to semi-consonants: [ˌ]

B. Form-stress, applied to meaningful forms: [ˈˈ ˈ ˌ]

C. Pitch, relating to end of utterance:

1. Medial: [,]

2. Final: [. ¿ ? !]

8. 3. The parts which our phonemes play in the structure of our
language are in reality much more diverse than this; in fact, we can
easily show that no two of them play exactly the same part.130

Since every utterance contains, by definition, at least one syllabic
phoneme, the simplest way to describe the phonetic structure of a
language is to state which non-syllabic phonemes or groups of
non-syllabic phonemes (clusters) appear in the three possible positions:
initial, before the first syllabic of an utterance; final, after
the last syllabic of an utterance; and medial, between syllabics.

In this respect the diphthongs and triphthong play in English
the same part as do the simple vowels; it is precisely this fact that
compels us to class them as compound phonemes and not as mere
successions of phonemes.

For convenience, I shall place a number before each phoneme
or group of phonemes that shows any peculiarity in its structural

Taking first the initial non-syllabics, we find at the outset that
two phonemes never begin an utterance; they are (1) [ŋ, ž]. We
ignore foreign forms, such as the French name Jeanne [žan].

Further, six of the non-syllabics that occur in initial position
never appear as members of an initial cluster: (2) [v, ð, z, č, ǰ, j].

The initial clusters all begin with one of the following non-syllabics:
(3) [p, t, k, b, d, g, f, θ, s, š, h]. Here we find an accord
between the structural grouping and our physiologic description,
since our structural group (3) embraces exactly the physiologic
groups of stops and unvoiced spirants.

If the first consonant of the cluster is (4) [s], it may be followed
by one of the set (5) [p, t, k, f, m, n], as in spin, stay, sky, sphere,
small, snail.

All the initials of group (3) and the combinations of (4) [s]
with (6) [p, t, k] may be followed by one of the set (7) [w, r, l],
with the following restrictions :

(8) [w] never comes after (9) [p, b, f, š], and never after the
combination of (4) [s] with (10) [t]. The actual clusters, then,
are illustrated by the words twin, quick, dwell, Gwynne, thwart,
swim, when [hwen], squall.

(11) [r] never comes after (12) [s, h]. The clusters, therefore,
are those which begin the words pray, tray, crow, bray, dray, gray,
fray, three, shrink, spray, stray, scratch.

(13) [1] never comes after (14) [t, d, θ, š, h], and never after
the combination of (4) [s] with (15) [k]. The clusters, accordingly,
are those which appear in play, clay, blue, glue, flew, slew, split.

8. 4. We come now to the final clusters. These are subject to
131the general rule that the same phoneme never occurs in two adjoining
positions: there are no such final groups as [ss] or [tt].
This rule holds good also for initial clusters and is implied by our
description of them, but it does not hold good, as we shall see,
for medial clusters.

We have undertaken to view combinations of vowel plus [j]
or [w] as compound phonemes (diphthongs) and accordingly cannot
count the semivowels in these combinations as final non-syllabics
or parts of clusters. If, accordingly, we eliminate these
cases (e.g. say [sej], go [gow]), we find that (16) [h, j, w] do not
occur as final non-syllabics or members of final clusters. All the
remaining non-syllabics occur in both of these functions.

English final clusters consist of two, three, or four non-syllabics.
One can describe the combinations most simply by saying that
each cluster consists of a main final consonant, which may be
preceded by a pre-final, which in turn may be preceded by a
second pre-final; further, the main final may be followed by a
post-final. This gives us six possibilities:

tableau without post-final | with post-final | main final alone | pre-final plus main final second pre-final plus pre-final plus main final

The consonants which occur as post-finals are (17) [t, d, s, z].
In a form like test or text we call the [-t] a main final, because there
exist forms like tests, texts, in which a further consonant (a postfinal)
is added, but in a form like wished [wišt] we call the [-t]
a post-final because the cluster [-st] is not paralleled by any cluster
with the addition of a further consonant: we have no such final
cluster as, say, [-šts].

The occurrence of the post-finals is limited by three important
restrictions. The post-finals (18) [t, s] are the only ones that occur
after the main finals (19) [p, t, k, c, f, 6, s, s]; these same post-finals
never occur after any other sounds; and the post-finals
(20) [t, d] are the only ones that occur after the main finals (21)
[č, ǰ, s, z, š, ž]. It is worth noticing that set (19) agrees, except for
the absence of [h], with the physiological class of unvoiced sounds,
132and that set (21) embraces the physiological classes of affricates
and sibilants. These restrictions group the main finals into six

Those in (19) but not in (21) may be followed by [t, s], as [p]
in help, helped, helps;

those in neither (19) nor (21) may be followed by [d, z], as [b]
in grab, grabbed, grabs;

those in (19) and (21) may be followed only by [t], as [č] in
reach, reached;

those in (21) but not in (19) may be followed only by [d], as
[ǰ] in urge, urged;

[t] in (19) but not in (21), owing to the rule of no doubling, may
be followed only by [s], as in wait, waits;

[d] in neither (19) nor (21), owing to the same rule, may be
followed only by [z], as infold, folds.

We turn now to the pre-finals. The main consonants (22) [g, ð, ž,
ŋ, r] are never accompanied by a pre-final, and the consonants (23)
[b, g, č, ǰ, v, š] never occur as pre-finals. The combinations that
remain are subject to the following further restrictions:

The pre-finals (24) [l, r] do not occur before the main final (25)
[z]. Their combinations, accordingly, are those which appear in the
following examples: harp, barb, heart, hard, hark, march, barge,
scarf, carve, hearth, farce, harsh, arm, barn; help, bulb, belt, held,
milk, filch, bilge, pelf, delve, wealth, else, Welsh, elm, kiln.

The pre-final (25) [n] occurs only before the main finals (27)
[t, d, č, ǰ, θ, s, z], as in ant, sand, pinch, range, month, once, bronze.

The pre-final (28) [m] occurs only before the main finals (29)
[p, t, f, θ], as in camp, dreamt, nymph; the combination with
(30) [θ] occurs with the second pre-final (11) [r]: warmth.

The pre-final (31) [ŋ] occurs only before (32) [k, θ], as in link,

The pre-final (4) [s] occurs only before (6) [p, t, k], as in wasp,
test, ask. Before (10) [t] it may be preceded by the second pre-final
(15) [k], as in text.

The pre-finals (33) [ð, z] occur only before the main final (28)
[m], as in rhythm, chasm.

The pre-final (10) [t] occurs only before the main finals (34)
[θ, s], as in eighth [ejtθ], Ritz (compare, with post-final [t] added,
the slang ritzed [ritst] ‘snubbed’). The combination with the main
final (4) [s] occurs also with second pre-final (11) [r] in quartz.133

The pre-final (35) [d] occurs only before (36) [θ, z], as in width,

The pre-finals (37) [p, k] occur only before the main finals (18)
[t, s], as in crypt, lapse, act, tax. Of these two, the pre-final (15) [k]
before the main final (4) [s] occurs also with the second pre-final
(31) [ŋ], as in minx (compare, with a post-final [t] added, the
slang jinxed [ǰiŋkst] ‘gave bad luck’); the other, [p], occurs with
the second pre-final (28) [m]: glimpse, tempt.

The pre-final (38) [f] occurs only before (10) [t], as in lift.

The medial non-syllabics of English consist of all the combinations
of final plus initial, ranging from hiatus, complete lack of a
non-syllabic, as in saw it [ˈsɔ it], to such clusters as in glimpsed
[-mpst str-], including repetitions of the same phoneme, as
in that time [-t t-] or ten nights [-n n-].

8. 5. A survey of the 38 functional sets of non-syllabics will
show that this classification suffices to define every non-syllabic
phoneme in our language. In the same way, most or possibly all
of our syllabic phonemes could be defined by the parts they play
in the structure of our language. Since different types of standard
English differ in the distributions of the syllabic phonemes, I
shall mention only a few of the pattern features.

Of the semi-consonants, only [r̩] occurs in stressed syllables;
it never occurs before [r]. The syllabic semivowel [u] is distinguished
by the fact that it does not occur initially, and occurs
medially only before [t, k, d, s, š, l], as in put, took, wood, puss,
push, pull; it occurs also before [f, m], as in roof, room, but here
always beside a more elegant variant with [uw]. Neither [i] nor
[u] occurs in final position.

Of the vowels, [e, a] do not occur before semivowels (in diphthong
combinations) and [o] does not occur before [w]. Only
[ɔ, a] occur in final position, as in saw, ma. The vowel [a] occurs
only before [ž , m, r], as in garage, calm, far, and before medial
[ð], as in father. The phonemes [i, e, ɛ, a] occur before [r] only if
another vowel follows, as in spirit, herring, marry, sorry; [ɔ] occurs
before [r] only when the [r] is a pre-final, as in horn, horse, north;
in many types of pronunciation the combination [or] is entirely
lacking. The vowel [ɔ] occurs before [r] only if [w] precedes, as in
war, dwarf. The vowel [ɑ] occurs before [g] only as a less common
variant of [ɔ], as in log, fog.

Of the diphthongs, only [ij, ej, ow] occur before [rs], as in fierce,
134scarce, course; before the other combinations of pre-final [r] the
only permitted diphthongs are [ow], as in cord, fork, torn, and, in
only a few dialectal-sounding words [ej]: laird, cairn. Before pre-final
[1] the only permitted diphthongs are [ij, aj, ow], and the first
two occur only when [d] follows, as in field, mild, old, colt. Before
pre-final [n] only [ɔj, aw] occur with any freedom, as in pint,
mount, bind, bound; [ɔj, ej] occur when [t] follows, as in paint,
point. The diphthongs do not occur before [ŋ].

The triphthong [juw] differs from ordinary combinations of
[j] plus vowel or diphthong (yank, year, Yale) in that it occurs
after initial consonants; it occurs after [p, k, b, g, f, h, v, m, n]
as in pew, cue, beauty, gules, few, view, muse, new, and after the
clusters [sp, sk], as in spew, skew; after [n] there is a less elegant
variant with [uw] instead of [juw], but, on the other hand, [juw]
occurs in an elegant pronunciation after [t, d, θ, s, l, st], where
[uw] is the commoner variant, as in tune, dew, thews, sue, lute,

We shall find that the grammatical structure of a language
implies groupings of the phonemes which supplement the groups
definable on the basis of succession (§ 13.6).

8. 6. The structural pattern differs greatly in different languages,
and leads us to recognize different types of compound phonemes.
German, for instance, has, on the whole, a structural scheme
much like that of English, but with some striking differences.
The voiced stops and spirants [b, d, g, v, z] never occur in final
position. The initial groups can be simply described only if one
takes the affricate combinations [pf, ts] as compound phonemes,
as in Pfund [pfunt] ‘pound,’ zehn [tse:n] ‘ten,’ zwei [tsvaj] ‘two.’
The only diphthongs are [aj, aw, oj]; the simplicity of structure in
this respect, leads phoneticians to transcribe them rather by [ai,
au, oi], since no ambiguity can arise. The French system differs
not only as to the particular clusters, but also in more general
respects. The diphthongs are rising, such as [jɛ, wa]. The greatest
difference is in the use of the vowel phoneme [ə], whose occurrence
is governed largely by the phonetic pattern, so that it may be said
to play the part of a secondary rather than of a primary phoneneme
The phoneme [ə] occurs wherever without it there would arise
an unpermitted cluster of consonants. Thus, it occurs in le chat
[lə ša] ‘the cat,’ because [Iš] is not permitted as an initial cluster,
but not in l'homme [l ɔm] ‘the man,’ where no cluster arises. It appears
135in cheval [šəval] ‘horse,’ since the cluster [šv] is not permitted
initially, but since this cluster is permitted in medial position,
one says un chevaln šval] ‘a horse.’ The medial clusters are
limited, for the most part to two consonants; thus, [rt] is permitted
as a final cluster, as in porte [pɔrt] ‘carries,’ but if an initial
consonant follows, [ə] is inserted, as in porte bien [pɔrtə bjɛn]
‘carries well.’ An entirely different system appears in a language
like Plains Cree. The structure groups the phonemes into five
sets: (1) the vowels [a, a:, e:, i, iː, u, o:]; these are the only syllabic
phonemes; (2) consonants of four types: stops [p, t, k],
including the affricate [č]; spirants [s, h]; nasals [m, n]; semivowels
[j, w]. The initial possibilities are: no consonant; any one consonant;
stop, spirant, or nasal plus semivowel. The medial possibilities
are: any one consonant; stop, spirant, or nasal plus semivowel;
spirant plus stop; spirant plus stop plus semivowel. The
only final possibility is one consonant. The Fox language, with a
somewhat similar patterning, permits of no final consonant: every
utterance ends in a short vowel.

While English is especially rich in consonant clusters, it is easy
to find others, such as initial [pf-, pfl-, pfr-, ts-, tsv-, šv-, kn-, gn-]
in German, e.g. Pflaume [ˈpflawme] ‘plum,’ schwer [šve:r] ‘heavy,’
Knie [kniː] ‘knee,’ or the clusters in Russian [tku] ‘I weave,’ [mnu]
‘I squeeze,’ [šči] ‘cabbage-soup,’ [lšču] ‘I flatter.’ Final clusters
foreign to English appear, for example, in German Herbst [herpst]
‘autumn’ and Russian [boršč] ‘beet-soup.’

8. 7. Once we have defined the phonemes as the smallest units
which make a difference in meaning, we can usually define each
individual phoneme according to the part it plays in the structural
pattern of the speech-forms. We observe, especially, that
the structural pattern leads us to recognize also compound phonemes,
which resemble successions of other phonemes, but play
the part of a simple phoneme, and that very slight acoustic differences,
such as, in English, the syllabic-stress on [r, l, m, n],
or the greater tensity of [j, w] compared to syllabic [i, u], may give
rise to separate phonemes.

The phonemes so defined are the units of signaling; the meaningful
forms of a language can be described as arrangements of
primary and secondary phonemes. If we take a large body of
speech, we can count out the relative frequencies of phonemes
and of combinations of phonemes. This task has been neglected
136by linguists and very imperfectly performed by amateurs, who
confuse phonemes with printed letters. Taking the total number
of phonemes in the text used as 100 per cent, a recent count for
English shows the following percentage frequencies for consonant

tableau n | t | r | s | d | l | ð | z | m | k | v | w | p | f | b | h | ŋ | š | g | j | č | ǰ | θ | ž

The figures for [r, l, m, n] include the occurrences in syllabic
function; those for [j] and [w] do not include the occurrences of
these phonemes as parts of diphthongs or triphthong. The count
of vowel phonemes is too confused to allow of plain reading. Apparently,
[e] is the most-used, with a frequency of over 8 per cent;
next comes [ij], with over 6 per cent; then [ɛ], with 3.5 per cent.
The figures for groups of phonemes are unusable. From this and
similar counts it is evident that the phonemes of a language perform
very different roles as to frequency. Moreover, there seems
to be some resemblance between languages; thus, in languages
which use two types of stops, such as our [p, t, k] versus [b, d, g],
the stop of the unvoiced type in each pair is more frequent than
its voiced mate, — for instance, [t] more frequent than [d]. A
serious study of this matter is much to be desired.

8. 8. We have seen three ways of studying the sounds of speech.
Phonetics in the strict sense — that is, laboratory phonetics —
gives us a purely acoustic or physiological description. It reveals
only the gross acoustic features. In practice, the laboratory
phonetician usually singles out for study some feature which his
lay knowledge recognizes as characteristic of a phoneme. Practical
phonetics is an art or skill, not a science; the practical phonetician
frankly accepts his everyday recognition of phonemic units and
tries to tell how the speaker produces them. The term phonology
is sometimes placed in contrast with the two forms of phonetics:
phonology pays no heed to the acoustic nature of the phonemes,
but merely accepts them as distinct units. It defines each phoneme
by its role in the structure of speech-forms. It is important to
remember that practical phonetics and phonology presuppose a
137knowledge of meanings: without this knowledge we could not
ascertain the phonemic features.

The description of a language, then, begins with phonology, which
defines each phoneme and states what combinations occur. Any
combination of phonemes that occurs in a language, is pronounceable
in this language, and is a phonetic form. The combination
[mnu], for instance is unpronounceable in English, but the combination
[men] is pronounceable and is a phonetic form.

When the phonology of a language has been established, there
remains the task of telling what meanings are attached to the
several phonetic forms. This phase of the description is semantics.
It is ordinarily divided into two parts, grammar and lexicon.

A phonetic form which has a meaning, is a linguistic form. Thus,
any English sentence, phrase, or word is a linguistic form, and so
is a meaningful syllable, such as, say, [mɛl] in maltreat, or [mon]
in Monday; a meaningful form may even consist of a single phoneme,
such as the [s] which means ‘more than one’ in plural-forms
like hats, caps, books. In the following chapters we shall see how
meanings are connected with linguistic forms.138

Chapter 9

9. 1. The study of speech-sounds without regard to meanings is
an abstraction: in actual use, speech-sounds are uttered as signals.
We have defined the meaning of a linguistic form as the situation
in which the speaker utters it and the response which it calls forth
in the hearer. The speaker's situation and the hearer's response
are closely co-ordinated, thanks to the circumstance that every one
of us learns to act indifferently as a speaker or as a hearer. In the
causal sequence

image speaker's situation → speech → hearer's response,

the speaker's situation, as the earlier term, will usually present a
simpler aspect than the hearer's response; therefore we usually
discuss and define meanings in terms of a speaker's stimulus.

The situations which prompt people to utter speech, include
every object and happening in their universe. In order to give a
scientifically accurate definition of meaning for every form of a
language, we should have to have a scientifically accurate knowledge
of everything in the speakers' world. The actual extent of
human knowledge is very small, compared to this. We can define
the meaning of a speech-form accurately when this meaning has to
do with some matter of which we possess scientific knowledge.
We can define the names of minerals, for example, in terms of
chemistry and mineralogy, as when we say that the ordinary
meaning of the English word salt is ‘sodium chloride (NaCl),’ and
we can define the names of plants or animals by means of the
technical terms of botany or zoology, but we have no precise way
of defining words like love or hate, which concern situations that
have not been accurately classified — and these latter are in the
great majority.

Moreover, even where we have some scientific (that is, universally
recognized and accurate) classification, we often find
that the meanings of a language do not agree with this classification.
The whale is in German called a ‘fish’: Walfisch [ˈval-ˌfiš]
139and the bat a ‘mouse’: Fledermaus [ˈfle:der-ˌmaws]. Physicists
view the color-spectrum as a continuous scale of light-waves of
different lengths, ranging from 40 to 72 hundred-thousandths of a
millimetre, but languages mark off different parts of this scale quite
arbitrarily and without precise limits, in the meanings of such
color-names as violet, blue, green, yellow, orange, red, and the color-names
of different languages do not embrace the same gradations.
The kinship of persons seems a simple matter, but the terminologies
of kinship that are used in various languages are extremely
hard to analyze.

The statement of meanings is therefore the weak point in
language-study, and will remain so until human knowledge advances
very far beyond its present state. In practice, we define the
meaning of a linguistic form, wherever we can, in terms of some
other science. Where this is impossible, we resort to makeshift
devices. One is demonstration. If someone did not know the meaning
of the word apple, we could instruct him by handing him an
apple or pointing at an apple, and continuing, as long as he made
mistakes, to handle apples and point at them, until he used the
word in the conventional way. This is essentially the process by
which children learn the use of speech-forms. If a questioner
understood enough of our language, we could define the word
apple for him by circumlocution — that is, in the manner of our
dictionaries, by a roundabout speech which fitted the same situations
as does the word apple, saying, for instance: “The well-known,
firm-fleshed, smooth-skinned, round or oblong pome fruit of the
trees of the genus Malus, varying greatly in size, shape, color, and
degree of acidity.” Or else, if we knew enough of the questioner's
language, we could answer him by translation — that is, by uttering
a roughly equivalent form of his language; if he were a Frenchman,
for instance, we could give pomme [pɔm] as the meaning of
apple. This method of definition appears in our bilingual dictionaries.

9. 2. The situations which prompt us to utter any one linguistic
form, are quite varied; philosophers tell us, in fact, that no two
situations are ever alike. Each one of us uses the word apple, in
the course of a few months, of many individual pieces of fruit
which differ in size, shape, color, odor, taste, and so on. In a
favorable case, such as that of the word apple, all the members of
the speech-community have been trained, from childhood, to use
140the speech-form whenever the situation (in this case, the object)
presents certain relatively definable characteristics. Even in
cases like this, our usage is never quite uniform, and most speech-forms
have less clear-cut meanings. Nevertheless, it is clear that
we must discriminate between non-distinctive features of the situation,
such as the size, shape, color, and so on of any one particular
apple, and the distinctive, or linguistic meaning (the semantic
features) which are common to all the situations that call forth
the utterance of the linguistic form, such as the features which are
common to all the objects of which English-speaking people use
the word apple.

Since our study ordinarily concerns only the distinctive features
of form and meaning, I shall henceforth usually omit the qualification
linguistic or distinctive, and speak simply of forms and meanings,
ignoring the existence of non-distinctive features. A form is
often said to express its meaning.

9. 3. Even if we had an accurate definition of the meaning that
is attached to every one of the forms of a language, we should still
face a difficulty of another sort. A very important part of every
situation is the state of the speaker's body. This includes, of
course, the predisposition of his nervous system, which results
from all of his experiences, linguistic and other, up to this very
moment — not to speak of hereditary and pre-natal factors. If we
could keep an external situation ideally uniform, and put different
speakers into it, we should still be unable to measure the equipment
each speaker brought with him, and unable, therefore, to predict
what speech-forms he would utter, or, for that matter, whether he
would utter any speech at all.

If we had perfect definitions, we should still discover that during
many utterances the speaker was not at all in the situation which
we had defined. People very often utter a word like apple when
no apple at all is present. We may call this displaced speech. The
frequency and importance of displaced speech is obvious. We recall
the infant “asking for” his doll (§ 2.5). Relayed speech embodies
a very important use of language: speaker A sees some
apples and mentions them to speaker B, who has not seen them;
speaker B relays this news to C, C to D, D to E, and so on, and it
may be that none of these persons has seen them, when finally
speaker X goes and eats some. In other ways, too, we utter linguistic
forms when the typical stimulus is absent. A starving beggar
141at the door says I'm hungry, and the housewife gives him food:
this incident, we say, embodies the primary or dictionary meaning
of the speech-form I'm hungry. A petulant child, at bed-time,
says I'm hungry, and his mother, who is up to his tricks, answers
by packing him off to bed. This is an example of displaced speech.
It is a remarkable fact that if a foreign observer asked for the
meaning of the form I'm hungry, both mother and child would still,
in most instances, define it for him in terms of the dictionary meaning.
Lying, irony, jesting, poetry, narrative fiction, and the like,
are probably as old and certainly as widespread as language. As
soon as we know the dictionary meaning of a form, we are fully
able to use it in displaced speech; our dictionaries and handbooks
of foreign languages need tell us only the dictionary meaning. The
displaced uses of speech are derived in fairly uniform ways from its
primary value, and require no special discussion; nevertheless, they
add to our uncertainty as to the forms that a given speaker will
utter (if he speaks at all) in a given situation.

9. 4. Adherents of mentalistic psychology believe that they
can avoid the difficulty of defining meanings, because they believe
that, prior to the utterance of a linguistic form, there occurs within
the speaker a non-physical process, a thought, concept, image, feeling,
act of will, or the like, and that the hearer, likewise, upon receiving
the sound-waves, goes through an equivalent or correlated
mental process. The mentalist, therefore, can define the meaning
of a linguistic form as the characteristic mental event which
occurs in every speaker and hearer in connection with the utterance
or hearing of the linguistic form. The speaker who utters the word
apple has had a mental image of an apple, and this word evokes a
similar image in a hearer's mind. For the mentalist, language is
the expression of ideas, feelings, or volitions.

The mechanist does not accept this solution. He believes that
mental images, feelings, and the like are merely popular terms for
various bodily movements, which, so far as they concern language,
can be roughly divided into three types:

(1) large-scale processes which are much the same in different
people, and, having some social importance, are represented by
conventional speech-forms, such as I'm hungry (angry, frightened,
sorry, glad; my head aches, and so on);

(2) obscure and highly variable small-scale muscular contractions
and glandular secretions, which differ from person to person,
142and, having no immediate social importance, are not represented
by conventional speech-forms;

(3) soundless movements of the vocal organs, taking the place
of speech-movements, but not perceptible to other people (“thinking
in words,” § 2.4).

The mechanist views the processes in (1) simply as events which
the speaker can observe better than anyone else; the various
problems of meaning, such as that of displaced speech (the naughty
child saying I'm hungry), exist here no less than elsewhere. The
mechanist believes that the processes in (2) are private habits left
over, as traces, from the vicissitudes of education and other experience;
the speaker reports them as images, feelings, and so on,
and they differ not only for every speaker, but for every occasion of
speech. The speaker who says, “I had the mental image of an
apple,” is really saying, “I was responding to some obscure internal
stimuli of a type which was associated at some time in my past
with the stimuli of an apple.” The sub-vocal speech in (3) seems
to the mechanist merely a derivative of the habit of actual speech-utterance;
when we are assured that a speaker has inaudibly
performed the speech-movements of a certain utterance (“thought
it in words”), we face exactly the same problem as when he has
audibly uttered the same speech-form. In sum, then, the “mental
processes” seem to the mechanist to be merely traditional names
for bodily processes which either (1) come within the definition of
meaning as speaker's situation, or (2) are so distantly correlated
with speech-utterance as to be negligible factors in the speaker's
situation, or (3) are mere reproductions of the speech-utterance.

Although this difference of opinion plays a decisive part in our
views about the fundamentals of language, as of other human
activities, and although mentalists lean heavily upon their terminology
in all discussion of meaning, the dispute has really very little
to do with problems of linguistic meaning. The events which the
mentalist designates as mental processes and the mechanist classifies
otherwise, affect in every case only one person: every one of
us responds to them when they occur within him, but has no way
of responding to them when they occur in anyone else. The
mental processes or internal bodily processes of other people are
known to each one of us only from speech-utterances and other
observable actions. Since these are all we have to work with, the
mentalist in practice defines meanings exactly as does the mechanist,
143in terms of actual situations; he defines apple not as “the image
of the well-known, firm-fleshed, etc… fruit,” but, like the
mechanist, omits the first three of these words, and, in fact, for
all speakers except himself, merely infers that the image was present,
either from the fact that the speaker used the word apple, or
from some more definite utterance of the speaker's (“I had a mental
image of an apple”). In practice, then, all linguists, both mentalists
and mechanists, define meanings in terms of the speaker's
situation and, whenever this seems to add anything, of the hearer's

9. 5. Linguistic meanings are more specific than the meanings
of non-linguistic acts. A great deal of human co-operation is
effected without language, by such means as gestures (for instance,
pointing at something), the handling of objects (placing an object
into someone's hand, dashing an object to the ground), contact
(nudging, caressing), non-linguistic sounds, both non-vocal (snapping
the fingers, applause) and vocal (laughing, crying), and so on.
We must mention especially, in this last connection, the nonlinguistic
(non-distinctive) features of speech-sound, such as
plaintive, angry, commanding, drawling “tones of voice;” the
manner of speech, in fact, is, next to speech itself, our most effective
method of signaling. Linguistic forms, however, result, for the
most part, in far more accurate, specific, and delicate co-ordination
than could be reached by non-linguistic means; to see this, one
need only listen to a few chance speeches: Four feet three and a half
. — If you don't hear from me by eight o'clock, go without me.
Where's the small bottle of ammonia? Apparent exceptions, such
as elaborate systems of gesture, deaf-and-dumb language, signaling-codes,
the use of writing, telegraphy, and so on, turn out, upon
inspection, to be merely derivatives of language.

Since we have no way of defining most meanings and of demonstrating
their constancy, we have to take the specific and stable
character of language as a presupposition of linguistic study, just
as we presuppose it in our everyday dealings with people. We
may state this presupposition as the fundamental assumption of
(§ 5.3), namely:

In certain communities (speech-communities) some speech-utterances
are alike as to form and meaning

This virtue of speech-forms is bought at the cost of rationality.
The non-linguistic modes of communication are based directly
144upon our bodily make-up, or else arise directly from simple social
situations, but the connection of linguistic forms with their meanings
is wholly arbitrary. What we call horse, the German calls
Pferd [pfe:rt], the Frenchman cheval [šəval], the Cree Indian
[misatim], and so on; one set of sounds is as unreasonable as any

Our fundamental assumption implies that each linguistic form
has a constant and specific meaning. If the forms are phonemically
different, we suppose that their meanings also are different — for
instance, that each one of a set of forms like quick, fast, swift, rapid,
speedy, differs from all the others in some constant and conventional
feature of meaning. We suppose, in short, that there are no
actual synonyms. On the other hand, our assumption implies also
that if the forms are semantically different (that is, different as to
linguistic meaning), they are not “the same,” even though they
may be alike as to phonetic form. Thus, in English, the phonetic
form [bejr] occurs with three different meanings: bear ‘to carry;
to give birth to,’ bear ‘ursus,’ and bare ‘uncovered.’ Similarly,
[pejr] represents two nouns (pear and pair) and a verb (pare), and
many other examples will occur to the reader. Different linguistic
forms which have the same phonetic form (and differ, therefore,
only as to meaning) are known as homonyms. Since we cannot with
certainty define meanings, we cannot always decide whether a
given phonetic form in its various uses has always the same meaning
or represents a set of homonyms. For instance, the English
verb bear in bear a burden, bear troubles, bear fruit, bear offspring,
can be viewed as a single form or as a set of two or perhaps even
more homonyms. Similarly, charge, in charge the cannon with
, charge the man with larceny, charge the gloves to me, charge
him a stiff price
, can be viewed in several ways; the infantry will
charge the fort
seems to be different. The quality sloth and the
animal sloth probably represent a pair of homonyms to some speakers
and a single meaning to others. All this shows, of course, that
our basic assumption is true only within limits, even though its
general truth is presupposed not only in linguistic study, but by
all our actual use of language.

9. 6. Although the linguist cannot define meanings, but must
appeal for this to students of other sciences or to common knowledge,
yet, in many cases, having obtained definitions for some
forms, he can define the meanings of other forms in terms of
145these first ones. The mathematician, for instance, who is here
acting as a linguist, cannot define such terms as one and add,
but if we give him a definition of these, he can define two (‘one
added to one’), three (‘one added to two’), and so on, without end.
What we see plainly in mathematical language, where the denotations
are very precise, appears also in many ordinary speech-forms.
If the meanings of the English past tense and of the word go are
defined, the linguist can define went as ‘the past of go.’ If the
difference male : female is defined for the linguist, he can assure
us that this is the difference between he : she, lion : lioness, gander
: goose, ram
: ewe. The linguist has this assurance in very many
cases, where a language, by some recognizable phonetic or grammatical
feature, groups a number of its forms into form-classes:
in any one form-class, every form contains an element, the class-meaning,
which is the same for all forms of this form-class. Thus,
all English substantives belong to a form-class, and each English
substantive, accordingly, has a meaning, which, once it is defined
for us (say, as ‘object’), we can attribute to every substantive
form in the language. English substantives, further, are subdivided
into the two classes of singular and plural; granted a definition of
the meanings of these two classes, we attribute one of these meanings
to every substantive.

In every language we find certain forms, substitutes, whose
meaning consists largely or entirely of class-meanings. In English,
the pronouns are the largest group of substitutes. The pronouns
show us a very interesting combination of meanings. The principal
features are class-meanings; thus, somebody, someone have the class-meanings
of substantives, singulars, personals; he has the class-meanings
of substantives, singulars, personals, males; it has the
class-meanings of substantives, singulars, non-personals; they has
the class-meanings of substantives and plurals. In the second place,
a pronoun may contain an element of meaning which makes the
pronoun represent some particular substantive form of the language.
Thus, the pronouns some and none tell us that the particular
substantive is one which has been recently mentioned (Here are
: take some); in contrast with this, something, somebody,
someone, nothing, nobody, no one tell nothing about the species.
Thirdly, some pronouns contain an element of meaning which tells
us which particular objects in a species are concerned. Thus, he,
she, it, they imply that not only the species (say, policeman) has
146been mentioned, but also that the particular object of this species
(say, Officer Smith, or the one at this corner) has been identified.
This feature of meaning, once defined, will be found in various
other forms of our language; it occurs, apparently without admixture,
as the meaning of the article the, for this little word tells us
only that the following substantive denotes an identified individual
of a species.

In sum, then, we may say that certain meanings, once they are
defined, can be recognized as recurring in whole series of forms.
In particular, the last-named type, which has to do with the identification
of individual objects of a species, in the way of selection,
inclusion, exclusion, or numbering, elicits very uniform responses
from different persons, and recurs with relative uniformity in
different languages; these types of meaning, accordingly, give rise
to the specially accurate form of speech which we call mathematics.

9. 7. Vocal gestures, serving an inferior type of communication,
occur not only outside of speech, as in an inarticulate outcry,
but also in combination with speech-forms, in the disposition of
non-distinctive features of speech-sound, such as the “tone of
voice.” Some conventional speech-forms, in fact, seem to lie on
the border-line; thus, we have seen that, in English, the exclamations
pst [pst] and sh [š], with which we demand silence, violate
the phonetic pattern by the use as syllables of the relatively
un-sonorous phonemes [s, š]. Less striking deviations from the
phonetic pattern sometimes occur in words whose meaning
resembles that of a pointing gesture. In English the initial phoneme
[ð] occurs only in words of demonstrative and related meanings,
such as this, that, the, then, there, though; in Russian, the
phoneme [e] occurs initially in none but demonstrative words, such
as[ˈeto] ‘this.’

Non-phonemic, gesture-like features may become fairly fixed.
In Plains Cree the word [e:] ‘yes’ is ordinarily spoken with a diphthongal
glide in the vowel and a final glottal stop, somewhat
as [ɛe:ʔ], although neither of these features is phonemic in the language.
In our slang fashions, peculiar pitch-schemes occasionally
become fixed for certain values; in the last years, Yeah? and Is that
with a peculiar modification of the question-pitch, have been
used as facetious vulgarisms, expressing disbelief.

The latter expression has also a form Is zat so? which illustrates
another phase of unusual linguistic features, facetious mispronunciation.
147To say Please, oxcuse me, for instance, is a form of tired
wit. These distortions get their value from a resemblance to other
linguistic forms (as in our example, the word ox) or to the speech-forms
of foreigners, sub-standard speakers, and children, as in
the facetious use of [ɔj] for [r] in words like bird (imitating the substandard
speech of New York City), or in the use of baby-talk
(Atta boy! Atta dirl!).

Certain expressions have slurred and shortened by-forms in
which the phonetic pattern is lost; these are common formulas of
social intercourse, such as greetings and terms of address. Thus,
How do you do? is shortened in all manner of ways into forms which
cannot be recorded in terms of English phonemes, but only suggested
by such sketches as [ǰˈduw¿] or [dˈduw¿]; How are you?
is something like [hwaj¿ haj¿]; madam appears as [m] in Yes'm.
These by-forms occur only in the formula; in asking How do you do
[ˈhaw ǰu ˈduw it¿] for example, we do not use the over-slurred
form. These shortened forms occur in various languages; their
relation to normal speech is obscure, but evidently they represent
a kind of sub-linguistic communication, in which the ordinary
meaning of the forms plays no part.

We can mention any sound by means of a rough imitation in
terms of vocal sound, as when we tell the calls of animals, or when
we report the noise of an engine. In this way we can also mention
speech-sounds; talking about a person who lisps, for instance,
someone may say, “I am tired of his eternal yeth, yeth.” The commonest
case is hypostasis, the mention of a phonetically normal
speech-form, as when we say, “That is only an if,” or “There is
always a but,” or when we talk about “the word normalcy” or
“the name Smith.” One may even speak of parts of words, as I
shall speak in this book of “the suffix -ish in boyish.” Hypostasis
is closely related to quotation, the repetition of a speech.

9. 8. The peculiarities of the forms discussed in the last paragraph
consist in deviations from the ordinary tie-up of phonetic
form with dictionary meaning. When there is no such deviation,
and only a normal phonetic form with a dictionary meaning is
to be considered, the latter will still exhibit great complexity.
We have already seen that present-day knowledge does not suffice
to unravel all the entanglements of meaning, but there are two
main features of the dictionary meaning of speech-forms which
demand such comment as we are able to make.148

Very many linguistic forms are used for more than one typical
situation. In English, we speak of the head of an army, of a procession,
of a household, or of a river, and of a head of cabbage;
of the mouth of a bottle, cannon, or river; of the eye of a needle,
and of hooks and eyes on a dress; of the teeth of a saw; of the
tongue of a shoe or of a wagon; of the neck of a bottle and of a
neck of the woods; of the arms, legs, and back of a chair; of the
foot of a mountain; of hearts of celery. A man may be a, fox, an ass,
or a dirty dog; a woman, a peach, lemon, cat, or goose; people are
sharp and keen or dwZZ, or else bright or foggy, as to their wits;
warm or coZa7 in temperament; crooked or straight in conduct; a
person may be wp w iAe air, at sea, off the handle, off his base, or
even beside himself, without actually moving from the spot. The
reader will be able to add examples practically without limit;
there is no greater bore than the enumeration and classification of
these “metaphors.”

The remarkable thing about these variant meanings is our assurance
and our agreement in viewing one of the meanings as
normal (or central) and the others as marginal (metaphoric or
transferred). The central meaning is favored in the sense that we
understand a form (that is, respond to it) in the central meaning
unless some feature of the practical situation forces us to look
to a transferred meaning. If we hear someone say There goes a
we look for a real fox, and if this is out of the question, we are
likely to take the utterance as displaced speech (say, as make-believe
or as part of a fairy-tale). Only if some situational feature
forces us — say, if the speaker is pointing at a man — do-we take
the form in the transferred sense. Even if we heard someone say,
The fox promised to help her, we should think of a fairy-tale rather
than of fox ‘unscrupulous and clever person.’ Sometimes the
practical feature that forces us to take a form in transferred meaning,
has been given by speech: Old Mr. Smith is a fox is bound to
be taken in transferred meaning, because we do not call real foxes
“Mr.” or give them family-names. He married a lemon forces us
to the transferred meaning only because we know that men do
not go through a marriage ceremony with a piece of fruit. On
the other hand, special practical situations may change all this.
People who lived close to the Fox Indians might, without special
constraint, take fox in our examples in the transferred sense ‘member
of the Fox nation.’149

In some cases a transferred meaning is linguistically determined
by an accompanying form. The word cat always has a transferred
meaning when it is accompanied by the suffix -kin (catkin), and
the word pussy when it is compounded with willow (pussy-willow);
similarly, the word eye when it has the suffix -let (eyelet). The words
dog, monkey, beard when they appear with the marks of verb derivation
(say, with a preceding to), always have transferred meaning
(to dog someone's footsteps; don't monkey with that; to beard a
lion in his den
). These linguistic features may be purely negative:
give out, used without an object (his money gave out; our horses
gave out
), always has a transferred meaning (‘become exhausted’).
In these cases the structure of the language recognizes the transferred
meaning. Even a linguist who made no attempt to define
meanings would have to specify that give out, intransitive, meant
something different (was a different form) from give out, transitive
(he gave out tickets).

In many cases we hesitate whether to view the form as a single
form with several meanings or as a set of homonyms. Examples
of this are air ‘atmosphere; tune, melody; manner’ (this last
including airs ‘haughty manners’), key ‘instrument for locking
and unlocking; set of tones in music,’ charge ‘attack; load; accuse;
debit,’ sloth ‘name of an animal; laziness.’

We are likely to make the mistake of thinking that the transferred
meanings of our language are natural and even inevitable
in human speech — the more so, as they appear also in other European
languages. This last, however, is merely a result of our common
cultural traditions; while transferred meanings occur in all
languages, the particular ones in any given language are by no
means to be taken for granted. Neither in French nor in German
can one speak of the eye of a needle or of an ear of grain. To speak
of the foot of a mountain seems natural to any European, but it
would be nonsense in Menomini and doubtless in many other
languages. On the other hand, in Menomini [una:ʔnɛw] ‘he places
him in position’ has also the transferred meaning ‘he picks lice
from him.’ In Russian, [noˈga] ‘leg’ is not used of the leg of a chair
or table; this transferred meaning appears only in the diminutive
[ˈnoška] ‘little leg; leg of a chair or table.’ Accordingly, when the
linguist tries to state meanings, he safely ignores the uses of displaced
speech, but does his best to register all cases of transferred

All this applies also to another type of deviant meaning, the
narrowed meaning, with this difference, that we are far more
ready to accept a form in a narrowed meaning. The practical
situation guides us at once to take car in different narrowed senses
in The diner is the second car forward (‘railroad-carriage’); Does
the car stop at this corner?
(‘ street-car’); Bring the car close to the
(‘motor-car’). When we hear the command to call a doctor,
we take it at once to mean a doctor of medicine. A burner is primarily
a person or instrument that burns things, but usually, in a narrowed
sense, a gas-tap arranged to give a certain kind of flame.
A bulb among gardeners is one thing and among electricians another.
A glass is usually a drinking-glass or a looking-glass;
glasses are usually eye-glasses. Narrowed meanings are hard to
define, because, after all, every occurrence of a form is prompted
by some one practical situation which need not contain all the
possibilities of meaning: apple is used now of a green one, now of a
red one, and so on.

The language itself, by formal characteristics, recognizes narrowed
meanings in certain combinations. For instance, blackbird
is not merely any ‘black bird’: in this combination the meaning
of black is greatly narrowed; similarly blueberry, whitefish, and the

Widened meanings are less common. In general, cat is the domestic
animal, but now and then we use the word to include lions,
tigers, and so on; the word dog, however, is not similarly used to
include wolves and foxes. On the other hand, hound is used poetically
and facetiously of any kind of dog. Often, the widened meaning
is recognized in the structure of the language, and appears
only when certain accompanying forms are present. Thus meat
is edible flesh, but in meat and drink and in sweetmeats it is food
in general; fowl is an edible bird, but in fish, flesh, or fowl or the
fowl of the air
it is any bird.

Often enough the speakers of a language do not distinguish a
central and a marginal meaning in cases where an outsider might
see two situationally different values; thus, day in English means
a period of twenty-four hours (Swedish dygn [dyŋn]) or the light
part of this period (in contrast with night; Swedish dag [da:g]).

9. 9. The second important way in which meanings show instability,
is the presence of supplementary values which we call
connotations. The meaning of a form for any one speaker is nothing
151more than a result of the situations in which he has heard this
form. If he has not heard it very many times, or if he has heard it
under very unusual circumstances, his use of the form may deviate
from the conventional. We combat such personal deviations by
giving explicit definitions of meaning; this is a chief use of our
dictionaries. In the case of scientific terms, we manage to keep the
meaning nearly free from connotative factors, though even here
we may be unsuccessful; the number thirteen, for instance, has
for many people a strong connotation.

The most important connotations arise from the social standing
of the speakers who use a form. A form which is used by a less
privileged class of speakers often strikes us as coarse, ugly, and
vulgar. I ain't got none, I seen it, I done it sound nasty to the
speaker of standard English. This may be offset by some special
factor: the speech-forms of tramps or criminals may bear a connotation
of devil-may-care wit, and those of a rustic type may
strike us as homely but poetic. A form used by a more privileged
class of speakers may strike us as over-formal or prettified and
affected. Most speakers of Central-Western American English
find this connotation in the use of [a] instead of [ɛ] in forms like
laugh, bath, can't and of [juw] instead of [uw] in forms like tune,
sue, stupid.

Connotations of local provenience are closely akin to these; a
Scotch or an Irish locution has its own tang; so have, in America,
certain real or supposed Anglicisms, such as luggage (for baggage)
or old chap, old dear as terms of address.

Even in communities that have no writing, some forms are
recognized (rightly or wrongly) as archaisms; in communities
that have written records, these serve as additional sources of
archaic forms. Examples are, in English, the old second-person
singular forms (thou hast), the third-person forms in -th (he hath),
the old present subjunctive (if this be treason), the pronoun ye,
and many forms like eve, e'en, e'er, morn, anent, and so on. Sometimes
fully current locutions may preserve some special aphoristic
; thus, an old sentence-construction survives in a few proverbs,
such as First come, first served or Old saint, young sinner.

The connotation of technical forms gets its flavor from the standing
of the trade or craft from which they are taken. Sea-terms
sound ready, honest, and devil-may-care: abaft, aloft, the cut of his
, stand by; legal terms precise and a bit tricky: without let or
152hindrance, in the premises, heirs and assigns; criminals' terms crass
but to the point: a stickup, a shot (of whiskey), get pinched.

The connotation of learned forms is vaguer but more frequent:
almost any colloquial form has a parallel form with learned connotation.

tableau normal learned | He came too soon. | He arrived prematurely. | It's too bad. | It is regrettable. | Where're you going? | What is your destination? | now | at present | if he comes | in case (in case that, in the event that, in the contingency that) he comes; should he come,… | so (that) you don't lose it. | in order that you may not lose it, lest you lose it.

As these examples show, the learned, elegant, and archaic types
of connotation merge in many a form. In formal speech and in
writing, we customarily prefer learned forms, up to a certain degree
: he who uses too many learned forms is a stilted speaker or a
tiresome writer.

Foreign speech-forms bear connotations of their own, which
reflect our attitude toward foreign peoples. The foreign features
of form may consist in peculiarities of sound or of phonetic pattern:
garage, mirage, rouge, a je ne sais quoi; olla podrida, chile con came;
dolce far niente, fortissimo; Zeitgeist, Wanderlust; intelligentsia. In
other instances, the foreign feature lies in the construction, as in the
French types marriage of convenience and that goes without saying.
This flavor is turned to facetious use in mock-foreign forms, such as
nix come erouse (mock-German), ish gabibble (‘it's none of my concern,’
supposedly Judeo-German). Schoolboys use mock-Latinisms,
such as the nonsense-form quid sidi quidit, or macaronic verse:
Boyibus kissibus priti girlorum, girlibus likibus, wanti somorum.

Some languages, and most notably, perhaps, English, contain a
great mass of semi-foreign or foreign-learned forms — a class of
forms with a separate style of pattern and derivation. Our textbooks
of rhetoric distinguish these forms, as the “Latin-French”
part of our vocabulary, from the “native” or “Anglo-Saxon”
forms. The connotation, however, does not depend directly upon
the actual provenience of the forms. The word chair, for instance,
153is Latin-French in origin, but does not belong to the foreign-learned
part of our vocabulary. The chief formal characteristics
of our foreign-learned forms is perhaps the use of certain accented
suffixes and combinations of suffixes, such as [-itij] ability; [-ˈejšn̩]
education. Another feature is the use of certain phonetic alternations,
such as [sijv] in receive, but [sep] in reception and [sij] in
receipt, or [vajd] in provide, but [vid] in provident, [viz] in visible,
and [viž] in provision. These peculiarities suffice to mark certain
words and constituents of words as foreign-learned, especially
certain prefixes (ab-, ad-, con-, de-, dis-, ex-, in-, per-, pre-, pro-,
re-, trans-); these prefixes themselves in part show peculiar phonetic
alternations, as in con-tain but collect, correct, and ab-jure but
abs-tain. Semantically, our foreign-learned forms are peculiar in
the capricious and highly specialized meanings of the combinations;
it seems impossible, for instance, to set up any consistent meaning
for elements like [sijv] in conceive, deceive, perceive, receive or [tend]
in attend, contend, distend, pretend, or [d(j)uws] in adduce, conduce,
deduce, induce, produce, reduce. The connotative flavor of these
forms lies in the learned direction: a speaker's ability to use these
forms measures his education. Errors in their use (malapropisms)
mark the semi-educated speaker. The less educated speaker fails
to understand many of these forms, and is to this extent shut out
from some types of communication; he may take vengeance by
using mock-learned forms, such as absquatulate, discombolulate,
rambunctious, scrumptious. Many languages contain a foreign-learned
layer of this kind: the Romance languages have a Latin
type, largely identical with ours; Russian, beside a fair sprinkling
of this type, has learned forms from Old Bulgarian; Turkish has a
stratum of Persian and Arabic words, and Persian of Arabic; the
languages of India similarly use Sanskrit forms.

Opposed to the foreign-learned connotation, the slangy connotation
is facetious and unrestrained: the users of slang forms are
young persons, sportsmen, gamblers, vagrants, criminals, and,
for that matter, most other speakers in their relaxed and unpretentious
moods. Examples are familiar, such as guy, gink, gazebo,
gazook, bloke, bird for ‘man,’ rod or gat for ‘pistol,’ and so on; the
slang form may at the same time be foreign, as loco ‘crazy,’ sabby
‘understand,’ vamoose ‘go away,’ from Spanish. The value is
largely facetious; when the slang form has been in use too long,
it is likely to be replaced by some new witticism.154

9. 10. The varieties of connotation are countless and indefinable
and, as a whole, cannot be clearly distinguished from denotative
meaning. In the last analysis, every speech-form has its own
connotative flavor for the entire speech-community and this, in
turn, is modified or even offset, in the case of each speaker, by the
connotation which the form has acquired for him through his
special experience. It may be well, however, to speak briefly of
two more types of connotation which stand out with at least
relative clearness.

In many speech-communities certain improper speech-forms are
uttered only under restricted circumstances; a speaker who utters
them outside the restriction is shamed or punished. The strictness
of the prohibition ranges from a mild rule of propriety to a severe
tabu. The improper forms belong for the most part to certain
spheres of meaning, but often enough there exist by their side
forms with the same denotation but without the improper connotation,
as prostitute by the side of the improper form whore.

Some improper forms denote objects or persons that are not to
be named in a casual way, or perhaps not to be named at all.
In English, various terms of religion, such as God, devil, heaven,
hell, Christ, Jesus, damn are proper only in serious speech. Violation
of the rule exposes the speaker to reproof or avoidance; on the
other hand, in certain groups or under certain conditions, the
violation connotes vigor and freedom. In many communities the
names of persons are tabu under some circumstances or to some
people. The male Cree Indian, for example, does not speak the
names of his sisters and of some other female relatives; he explains
the avoidance by saying, “I respect her too much.”

Another direction of impropriety is the tabu on so-called obscene
forms. In English there is a severe tabu on some speech-forms
whose meaning is connected with excretory functions, and on some
that deal with reproduction.

A third type of improper connotation is less universal among us;
the avoidance of ominous speech-forms, which name something
painful or dangerous. One avoids the words die and death (if
anything should happen to me
) and the names of some diseases.
Other peoples avoid mention of the left hand, or of thunderstorms.

In some communities one avoids the names of game animals,
either during the hunt or more generally. Under special conditions
155(as, on the war-path), many speech-forms may be avoided, or
inverted speech, saying the opposite of what one means, may be in

9. 11. The second more specialized type of connotation that
here deserves to be pointed out, is intensity. The most characteristic
intense forms are exclamations. For these we have in English
not only a special secondary phoneme [!], but also certain special
speech-forms, interjections, such as oh! ah! ouch! These forms all
reflect a violent stimulus, but differ in connotation from an ordinary
statement in which the speaker merely says that he is undergoing
a strong stimulus.

Certain speech-forms have an animated flavor, akin to the exclamatory,
as, for instance, the placing first of certain adverbs:
Away ran John; Away he ran. In connected narrative a similar
flavor appears in less violent transpositions: Yesterday he came (and
) is more lively than He came yesterday… In English
the historical present, in narrating past events, is either elegant,
as in the summary of a play or story, or, in ordinary speech, slightly
vulgar: Then he comes back and says to me…

English is especially rich in another type of intense forms, the
symbolic forms. Symbolic forms have a connotation of somehow
illustrating the meaning more immediately than do ordinary
speech-forms. The explanation is a matter of grammatical structure
and will concern us later; to the speaker it seems as if the
sounds were especially suited to the meaning. Examples are
flip, flap, flop, flitter, flimmer, flicker, flutter, flash, flush, flare,
glare, glitter, glow, gloat, glimmer, bang, bump, lump, thump, thwack,
whack, sniff, sniffle, snuff, sizzle, wheeze. Languages that have
symbolic forms show some agreement, but probably more disagreement
as to the types of sounds and meanings which are associated.
A special type of symbolic form, which is quite widely
distributed, is the repetition of the form with some phonetic variation,
as in snip-snap, zig-zag, riff-raff, jim-jams, fiddle-faddle,
teeny-tiny, ship-shape, hodge-podge, hugger-mugger, honky-tonk.

Closely akin to these are imitative or onomatopoetic intense forms,
which denote a sound or an object which gives out a sound: the
imitative speech-form resembles this sound: cock-a-doodle-doo,
meeow, moo, baa. Many bird names are of this sort: cuckoo, bobwhite,
whip-poor-will. Doubled forms are common: bow-wow,
ding-dong, pee-wee, choo-choo, chug-chug. These forms differ from
156language to language: the French dog says gnaf-gnaf [ɲaf ɲaf];
the German bell says bim-bam.

Among the forms just cited, some have an infantile connotation;
they are nursery-forms. The most familiar are papa and mama.
In English almost any doubled syllable may be used, in almost
any meaning, as a nursery-word; each family develops its private
supply of the type [ˈdijdij, ˈdajdaj, ˈdajdij, ˈmijmij, ˈwawa].
This custom provides speech-forms which the infant can reproduce
with relative ease, and it helps adults to turn the infant's
utterances into conventional signals.

The pet-name or hypochoristic connotation largely merges with
that of the nursery. In English, relatively few pet-names like Lulu,
have the doubled nursery form; in French this type is common:
Mimi, Nana, and so on. English pet-names are less uniform:
Tom, Will, Ed, Pat, Dan, Mike can be described structurally as
shortenings of the full name; this is not the case in Bob for Robert,
Ned for Edward, Bill for William, Dick for Richard, Jack for John.
Some have the diminutive suffix [-ij], as Peggy, Maggie for Margaret,
Fanny for Frances, Johnny, Willie, Billy.

There is some intensity also in the connotation of nonsense-forms.
Some of these, though conventional, have no denotation
at all, as tra-la-la, hey-diddle-diddle, tarara-boom-de-ay; others
have an explicitly vague denotation, as fol-de-rol, gadget, conniption
. Any speaker is free to invent nonsense-forms; in fact,
any form he invents is a nonsense-form, unless he succeeds in
the almost hopeless task of getting his fellow-speakers to accept
it as a signal for some meaning.157

Chapter 10
Grammatical Forms

10. 1. Our discussion so far has shown us that every language
consists of a number of signals, linguistic forms. Each linguistic
form is a fixed combination of signaling-units, the phonemes. In
every language the number of phonemes and the number of actually
occurring combinations of phonemes, is strictly limited.
By uttering a linguistic form, a speaker prompts his hearers to
respond to a situation; this situation and the responses to it, are
the linguistic meaning of the form. We assume that each linguistic
form has a constant and definite meaning, different from the
meaning of any other linguistic form in the same language. Thus,
hearing several utterances of some one linguistic form, such as
I'm hungry, we assume (1) that the differences in sound are irrelevant
(unphonetic), (2) that the situations of the several speakers
contain some common features and that the differences between
these situations are irrelevant (unsemantic), and (3) that
this linguistic meaning is different from that of any other form in
the language. We have seen that this assumption cannot be verified,
since the speaker's situations and the hearer's responses
may involve almost anything in the whole world, and, in particular,
depend largely upon the momentary state of their nervous
systems. Moreover, when we deal with the historical change of
language, we shall be concerned with facts for which our assumption
does not hold good. In the rough, however, our assumption
is justified by the mere fact that speakers co-operate in a very
refined way by means of language-signals. In describing a language,
we are concerned primarily with the working of this cooperation
at any one time in any one community, and not with
its occasional failures or with its changes in the course of history.
Accordingly, the descriptive phase of linguistics consists in a
somewhat rigid analysis of speech-forms, on the assumption that
these speech-forms have constant and definable meanings (§ 9.5).

Our basic assumption does have to be modified, however, right
at the outset, in a different way. When we have recorded a fair
158number of forms in a language, we always discover a feature
which we have so far ignored in our discussion: the partial resemblance
of linguistic forms. Suppose we hear a speaker say

John ran,

and a little later hear him or some other speaker say

John fell.

We recognize at once that these two forms, John ran and John
, are in part phonetically alike, since both of them contain an
element John [ǰɑn], and our practical knowledge tells us that the
meanings show a corresponding resemblance: whenever a form
contains the phonetic element [ǰɑn], the meaning involves a certain
man or boy in the community. In fact, if we are lucky, we
may hear someone utter the form


all by itself, without any accompaniment.

After observing a number of such cases, we shall be constrained
to modify the basic assumption of linguistics to read: In a speech-community
some utterances are alike or partly alike in sound and

The common part of partly like utterances (in our example,
John) consists of a phonetic form with a constant meaning: it
answers, therefore, to the definition of a linguistic form. The parts
which are not common to the partly-like utterances (in our example,
ran in the one utterance, and fell in the other) may, in
the same way, turn out to be linguistic forms. Having heard the
form John ran, we may later hear the form Bill ran, and perhaps
even (say, in answer to a question) an isolated Ran. The same will
happen with the component fell in John fell: we may hear a form
like Dan fell or even an isolated Fell.

In other cases, we may wait in vain for the isolated form. Knowing
the forms John, Bill, and Dan, we may hear the forms, Johnny,
Billy, and Danny and hope to hear now an isolated -y [-ij] with
some such meaning as ‘little,’ but in this instance we shall be disappointed.
In the same way, familiar with the forms play and
dance, we may hear the forms playing and dancing, and then hope,
in vain, to hear an isolated -ing [-iŋ], which might reassure us as to
the somewhat vague meaning of this syllable. In spite of the fact
that some components do not occur alone, but only as parts of
larger forms, we nevertheless call these components linguistic
159forms, since they are phonetic forms, such as [ij] or [iŋ], with constant
meanings. A linguistic form which is never spoken alone is
a bound form; all others (as, for instance, John ran or John or run
or running) are free forms.

In other cases we wait in vain for the occurrence of a form even
as part of some other form. For instance, having heard the form
cranberry, we soon recognize the component berry in other forms,
such as blackberry, and may even hear it spoken alone, but with
the other component of cranberry we shall have no such luck.
Not only do we wait in vain to hear an isolated *cran, but, listen
as we may, we never hear this element outside the one combination
cranberry, and we cannot elicit from the speakers any other
form which will contain this element cran-. As a practical matter,
observing languages in the field, we soon learn that it is unwise to
try to elicit such forms; our questions confuse the speakers, and
they may get rid of us by some false admission, such as, “Oh, yes,
I guess cran means red.” If we avoid this pitfall, we shall come
to the conclusion that the element cran- occurs only in the combination
cranberry. However, since it has a constant phonetic
form, and since its meaning is constant, in so far as a cranberry
is a definite kind of berry, different from all other kinds, we say
that cran-, too, is a linguistic form. Experience shows that we
do well to generalize this instance: unique elements, which occur
only in a single combination, are linguistic forms.

Sometimes we may be unable to decide whether phonetically
like forms are identical in meaning. The straw- in strawberry is
phonetically the same as the straw- in strawflower and as the isolated
straw, but whether the meanings are “the same,” we cannot say.
If we ask the speakers, they will answer sometimes one way,
sometimes another; they are no more able to tell than we. This
difficulty is part of the universal difficulty of semantics: the
practical world is not a world of clear-cut distinctions.

10. 2. We see, then, that some linguistic forms bear partial
phonetic-semantic resemblances to other forms; examples are,
John ran, John fell, Bill ran, Bill fell; Johnny, Billy; playing,
dancing; blackberry, cranberry; strawberry, strawflower. A linguistic
form which bears a partial phonetic-semantic resemblance to some
other linguistic form, is a complex form.

The common part of any (two or more) complex forms is a
linguistic form; it is a constituent (or component) of these complex
160forms. The constituent is said to be contained in (or to be included
or to enter into) the complex forms. If a complex form, beside
the common part, contains a remainder, such as the cran- in
cranberry, which does not occur in any other complex form, this
remainder also is a linguistic form; it is a unique constituent of the
complex form. The constituent forms in our examples above are:
John, ran, Bill, fell, play, dance, black, berry, straw, flower, cran(unique
constituent in cranberry), -y (bound-form constituent
in Johnny, Billy), -ing (bound-form constituent in playing, dancing).
In any complex form, each constituent is said to accompany
the other constituents.

A linguistic form which bears no partial phonetic-semantic
resemblance to any other form, is a simple form or morpheme.
Thus, bird, play, dance, cran-, -y, -ing are morphemes. Morphemes
may show partial phonetic resemblances, as do, for instance, bird
and burr, or even homonymy, as do pear, pair, pare, but this
resemblance is purely phonetic and is not paralleled by the meanings.

From all this it appears that every complex form is entirely
made up, so far as its phonetically definable constituents are concerned,
of morphemes. The number of these ultimate constituents
may run very high. The form Poor John ran away contains five
morphemes: poor, John, ran, a- (a bound form recurring, for
instance, in aground, ashore, aloft, around), and way. However, the
structure of complex forms is by no means as simple as this; we
could not understand the forms of a language if we merely reduced
all the complex forms to their ultimate constituents. Any English-speaking
person who concerns himself with this matter, is sure to
tell us that the immediate constituents of Poor John ran away are
the two forms poor John and ran away; that each of these is, in
turn, a complex form; that the immediate constituents of ran away
are ran, a morpheme, and away, a complex form, whose constituents
are the morphemes a- and way; and that the constituents
of poor John are the morphemes poor and John. Only in this
way will a proper analysis (that is, one which takes account of the
meanings) lead to the ultimately constituent morphemes. The
reasons for this will occupy us later.

10. 3. A morpheme can be described phonetically, since it
consists of one or more phonemes, but its meaning cannot be
analyzed within the scope of our science. For instance, we have
161seen that the morpheme pin bears a phonetic resemblance to other
morphemes, such as pig, pen, tin, ten, and, on the basis of these
resemblances, can be analyzed and described in terms of three
phonemes (§ 5.4), but, since these resemblances are not connected
with resemblances of meaning, we cannot attribute any meaning
to the phonemes and cannot, within the scope of our science,
analyze the meaning of the morpheme. The meaning of a morpheme
is a sememe. The linguist assumes that each sememe is a
constant and definite unit of meaning, different from all other
meanings, including all other sememes, in the language, but he
cannot go beyond this. There is nothing in the structure of morphemes
like wolf, fox, and dog to tell us the relation between their
meanings; this is a problem for the zoologist. The zoologist's
definition of these meanings is welcome to us as a practical help,
but it cannot be confirmed or rejected on the basis of our science.
A workable system of signals, such as a language, can contain
only a small number of signaling-units, but the things signaled
about — in our case, the entire content of the practical world
— may be infinitely varied. Accordingly, the signals (linguistic
forms, with morphemes as the smallest signals) consist of different
combinations of the signaling-units (phonemes), and each such
combination is arbitrarily assigned to some feature of the practical
world (sememe). The signals can be analyzed, but not the things
signaled about.

This re-enforces the principle that linguistic study must always
start from the phonetic form and not from the meaning. Phonetic
forms — let us say, for instance, the entire stock of morphemes in
a language — can be described in terms of phonemes and their
succession, and, on this basis, can be classified or listed in some
convenient order, as, for example, alphabetically; the meanings
— in our example, the sememes of a language — could be analyzed
or systematically listed only by a well-nigh omniscient observer.

10. 4. Since every complex form is made up entirely of morphemes,
a complete list of morphemes would account for all the
phonetic forms of a language. The total stock of morphemes in a
language is its lexicon. However, if we knew the lexicon of a
language, and had a reasonably accurate knowledge of each sememe,
we might still fail to understand the forms of this language.
Every utterance contains some significant features that are not
accounted for by the lexicon. We saw, for instance, that the five
162morphemes, John, poor, ran, way, a- which make up the form
Poor John ran away, do not fully account for the meaning of this
utterance. Part of this meaning depends upon the arrangement
— for example, upon the order of succession — in which these
morphemes appear in the complex form. Every language shows
part of its meanings by the arrangement of its forms. Thus, in
English, John hit Bill and Bill hit John differ in meaning by virtue
of the two different orders in which the morphemes are uttered.

The meaningful arrangements of forms in a language constitute
its grammar. In general, there seem to be four ways of arranging
linguistic forms.

(1) Order is the succession in which the constituents of a complex
form are spoken. The significance of order appears strikingly
in. contrasts such as John hit Bill versus Bill hit John. On the other
hand, *Bill John hit is not an English form, because our language
does not arrange these constituents in this order; similarly, play-ing
is a form, but ing-play is not. Sometimes differences of order have
connotative values; thus, Away ran John is livelier than John ran

(2) Modulation is the use of secondary phonemes. Secondary
phonemes, we recall (§ 5.11), are phonemes which do not appear in
any morpheme, but only in grammatical arrangements of morphemes.
A morpheme like John [ǰɑn] or run [ron] is really an
abstraction, because in any actual utterance the morpheme is
accompanied by some secondary phoneme which conveys a grammatical
meaning. In English, if the morpheme is spoken alone, it
is accompanied by some secondary phoneme of pitch (§ 7.6): it is
either John! or John? or John [.] — this last with falling final-pitch,
as, in answer to a question — and there is no indifferent or abstract
form in which the morpheme is not accompanied by any final-pitch.
In English complex forms, some of the constituents are
always accompanied by secondary phonemes of stress (§ 7.3); thus,
the difference in the place of stress distinguishes the noun convict
from the verb convict.

(3) Phonetic modification is a change in the primary phonemes
of a form. For instance, when the forms do [duw] and not [nɑt]
are combined into a complex form, the [uw] of do is ordinarily
replaced by [ow], and, whenever this happens, the not loses its
vowel, so that the combined form is don't [dow nt]. In this example
the modification is optional, and we have also the unmodified
163forms in do not, with a difference of connotation. In other cases
we have no choice. Thus, the suffix -ess with the meaning ‘female,’
as in count-ess, is added also to duke [d(j)uwk], but in this combination
the form duke is modified to duch- [doč-], for the word is
duchess [ˈdočes].

Strictly speaking, we should say that the morpheme in such cases
has two (or, sometimes, more) different phonetic forms, such as
not [nɑt] and [nt], do [duw] and [dow], duke and duch-, and that
each of these alternants appears under certain conditions. In our
examples, however, one of the alternants has a much wider range
than the other and, accordingly, is a basic alternant. In other cases,
the alternants are more on a par. In run and ran, for instance,
neither alternant is tied to the presence of any accompanying form,
and we might hesitate as to the choice of a basic alternant. We
find, however, that in cases like keep : kep-t the past-tense form
contains an alternant (kep-) which occurs only with a certain
accompanying form (-t); accordingly, to obtain as uniform as
possible a statement, we take the infinitive form (keep, run) as
basic, and describe the alternant which appears in the past tense
(kep-, ran) as a phonetically modified form. We shall see other
instances where the choice is more difficult; we try, of course, to
make the selection of a basic alternant so as to get, in the long
run the simplest description of the facts.

(4) Selection of forms contributes a factor of meaning because
different forms in what is otherwise the same grammatical arrangement,
will result in different meanings. For instance, some
morphemes spoken with exclamatory final-pitch, are calls for a
person's presence or attention (John! Boy!), while others, spoken
in the same way, are commands (Run! Jump!), and this difference
extends also to certain complex forms (Mr. Smith! Teacher!
versus Run away! Backwater!). The forms which, when spoken
with exclamatory final-pitch, have the meaning of a call, may be
said, by virtue of this fact, to make up a, form-class of the English
language; we may call it the form-class of “personal substantive
expressions.” Similarly, the forms which, when spoken with exclamatory
final-pitch, have the meaning of a command, make up,
by virtue of this fact, the English form-class of “infinitive expressions.”
Whether an exclamation is a call or a command, depends
upon the selection of the form from the one or the other of these
two classes.164

The meaning of a complex form depends in part upon the selection
of the constituent forms. Thus, drink milk and watch John
name actions, and, as we have just seen, are infinitive expressions,
but fresh milk and poor John name objects and are substantive
expressions. The second constituents, milk, and John, are the
same; the difference depends upon the selection of the first constituent.
By virtue of this difference, the forms drink and watch
belong to one English form-class (that of “transitive verbs”)?
and the forms fresh and poor to another (that of “adjectives”).

The features of selection are usually quite complicated, with
form-classes divided into sub-classes. In English, if we combine
a form like John or the boys (form-class of “nominative substantive
expressions”) with a form like ran or went home (form-class
of “finite verb expressions”), the resultant complex form means
that this object ‘performs’ this action (John ran, the boys ran,
John went home, the boys went home). These features of selection,
however, are supplemented by a further habit: we say John runs
but the boys run fast, and we never make the reverse combinations
of John with run fast, or of the boys with runs fast. The form-class
of nominative expressions is divided into two sub-classes
(“singular” and “plural”) and the form-class of finite verb expressions
likewise, into two sub-classes (“singular” and “plural”),
such that in the complex forms which mean that an object performs
an action, the two constituents agree as to the “singular”
or “plural” sub-class. In Latin, the form pater fīlium amat (or
fīlium pater amat) means ‘the father loves the son,’ and the form
patrem fīlius amat (or fīlius patrem amat) means ‘the son loves the
father;’ the forms pater ‘father’ and fīlius ‘son’ belong to a form-class
(“nominative case”) whose forms, in combination with a
verb like amat ‘he loves,’ denote the ‘performer’ of the action;
the forms patrem ‘father’ and fīlium ‘son’ belong to a different
form-class (“accusative case”), whose forms, in combination with
a verb like amat, denote the ‘undergoer’ (‘object’ or ‘goal’)
of the action.

The features of selection are often highly arbitrary and whimsical.
We combine prince, author, sculptor with the suffix -ess in
princess, authoress, sculptress (in this last case with phonetic modification
of [r̩] to [r]), but not king, singer, painter. By virtue of
this habit, the former words belong to a form-class from which
the latter words are excluded.165

10. 5. The features of grammatical arrangement appear in
various combinations, but can usually be singled out and separately
described. A simple feature of grammatical arrangement
is a grammatical feature or taxeme. A taxeme is in grammar what
a phoneme is in the lexicon — namely, the smallest unit of form.
Like a phoneme, a taxeme, taken by itself, in the abstract, is
meaningless. Just as combinations of phonemes, or, less commonly,
single phonemes, occur as actual lexical signals (phonetic forms),
so combinations of taxemes, or, quite frequently, single taxemes,
occur as conventional grammatical arrangements, tactic forms.
A phonetic form with its meaning is a linguistic form; a tactic
form with its meaning is a grammatical form. When we have occasion
to contrast the purely lexical character of a linguistic form
with the habits of arrangement to which it is subject, we shall
speak of it as a lexical form. In the case of lexical forms, we have
defined the smallest meaningful units as morphemes, and their
meanings as sememes; in the same way, the smallest meaningful
units of grammatical form may be spoken of as tagmemes, and
their meanings as episememes.

The utterance Run!, for example, contains two grammatical
features (taxemes), namely, the modulation of exclamatory final-pitch,
and the selective feature which consists in the use of an
infinitive verb (as opposed, for instance, to the use of a noun, as
in John!). Each of these two taxemes happens to be, in English,
a tactic form, since each is currently used as a unit of signaling.
Taking each of them with its meaning, we describe them as units
of grammatical form (tagmemes). The tagmeme of exclamatory
final-pitch occurs with any lexical form and gives it a grammatical
meaning (an episememe) which we may roughly describe, perhaps,
as ‘strong stimulus.’ The tagmeme of selection by which
infinitive forms are marked off as a form-class, has a grammatical
meaning (an episememe) which we may call a class-meaning and
roughly define as ‘action.’

A tagmeme may consist of more than one taxeme. For instance,
in forms like John ran; poor John ran away; the boys are here; I
, we find several taxemes. One constituent belongs to the
form-class of nominative expressions (John, poor John, the boys, I).
The other constituent belongs to the form-class of finite verb expressions
(ran, ran away, are here, know). A further taxeme of selection
assigns certain finite verb expressions to certain nominative
166expressions; thus, the constituents are not interchangeable
in the three examples I am, John is, you are. A taxeme of order
places the nominative expression before the finite verb expression:
we do not say *ran John. Further taxemes of order, in part reversing
the basic one, appear in special cases like did John run?
away ran John; will John? A taxeme of modulation appears only
in special cases, when the nominative expression is unstressed,
as in I know [aj ˈnow]. Taxemes of phonetic modification appear
also in certain special cases, such as John's here, with [z] for is,
or I'd go, with [d] for would. Now, none of these taxemes, taken
by itself, has any meaning, but, taken all together, they make up
a grammatical form, a tagmeme, whose meaning is this, that the
one constituent (the nominative expression) ‘performs’ the other
constituent (the finite verb expression).

If we say John ran! with exclamatory pitch, we have a complex
grammatical form, with three tagmemes. One of these is ‘strong
stimulus,’ the second is ‘(object) performs (action),’ and the third
has the episememe of ‘complete and novel’ utterance, and consists,
formally, in the selective feature of using an actor-action
phrase as a sentence.

10. 6. Any utterance can be fully described in terms of lexical
and grammatical forms; we must remember only that the meanings
cannot be defined in terms of our science.

Any morpheme can be fully described (apart from its meaning)
as a set of one or more phonemes in a certain arrangement. Thus,
the morpheme duke consists of the phonemes, simple and compound,
[d], [juw], [k], in this order; and the morpheme -ess consists
of the phonemes [e], [s], in this order. Any complex form can
be fully described (apart from its meaning) in terms of the immediate
constituent forms and the grammatical features (taxemes)
by which these constituent forms are arranged. Thus, the complex
form duchess [ˈdočes] consists of the immediate constituents
duke [djuwk] and -ess [es], arranged in the following way:

Selection. The constituent duke belongs to a special class of
English forms which combine with the form -ess. This form-class
includes, for instance, the forms count, prince, lion, tiger, author,
waiter, but not the forms man, boy, dog, singer; it is a sub-class of
a larger form-class of male personal nouns. The form -ess constitutes
a little form-class of its own, by virtue of the fact that it
(and it alone) combines with precisely the forms in the class just
167described. All these facts, taken together, may be viewed as a
single taxeme of selection.

Order. The form -ess is spoken after the accompanying form.

Modulation. The form -ess is spoken unstressed; the accompanying
form has a high stress.

Phonetic modification. The [juw] of duke is replaced by [o], and
the [k] by [č].

Given the forms duke and -ess, the statement of these four
grammatical features fully describes the complex form duchess.

Any actual utterance can be fully described in terms of the
lexical form and the accompanying grammatical features. Thus,
the utterance Duchess! consists of the lexical form duchess and the
two taxemes of exclamatory final-pitch and selection of a substantive

If some science furnished us with definitions of the meanings of
the units here concerned, defining for us the meanings (sememes)
of the two morphemes (duke and -ess) and the meanings (episememes)
of the three tagmemes (arrangement of duke and -ess;
use of exclamatory final-pitch; selection of a substantive expression),
then the meaning of the utterance Duchess! would be fully
analyzed and defined.

10. 7. The grammatical forms are no exception to the necessary
principle — strictly speaking, we should call it an assumption —
that a language can convey only such meanings as are attached to
some formal feature: the speakers can signal only by means of
signals. Many students of language have been misled in this matter
by the fact that the formal features of grammar are not phonemes
or combinations of phonemes which we can pronounce or transcribe,
but merely arrangements of phonetic forms. For this our
scholastic tradition may be largely to blame; if it were not for this
tradition, there would perhaps be nothing difficult about the fact,
for instance, that in English, John hit Bill and Bill hit John signal
two different situations, or that convict stressed on the first syllable
differs in meaning from convict stressed on the second syllable, or
that there is a difference of meaning between John! and John?
and John.

A form like John or run, mentioned in the abstract, without, for
instance, any specification as to final-pitch, is, properly speaking,
not a real linguistic form, but only a lexical form; a linguistic
form, as actually uttered, always contains a grammatical form.
168No matter how simple a form we take and how we utter it, we have
already made some selection by virtue of which the utterance
conveys a grammatical meaning in addition to its lexical content,
and we have used some pitch-scheme which, in English at any
rate, lends it a grammatical meaning such as ‘statement’ ‘yes-or-no
question,’ ‘supplement-question,’ or ‘exclamation.’

The grammatical forms of a language can be grouped into three
great classes:

(1) When a form is spoken alone (that is, not as a constituent
of a larger form), it appears in some sentence-type. Thus, in English,
the use of the secondary phoneme [!] gives us the sentence-type of
exclamation, and the use of a substantive expression gives us the
type of a call (John!).

(2) Whenever two (or, rarely, more) forms are spoken together,
as constituents of a complex form, the grammatical features by
which they are combined, make up a construction. Thus, the
grammatical features by which duke and -ess combine in the form
duchess, or the grammatical features by which poor John and ran
combine in the form poor John ran away, make up a construction.

(3) A third great class of grammatical forms must probably be
set up for the cases where a form is spoken as the conventional
substitute for any one of a whole class of other forms. Thus, the
selective feature by which the form he in English is a conventional
substitute for a whole class of other forms, such as John, poor John,
a policeman, the man I saw yesterday, whoever did this, and so on
(which forms, by virtue of this habit, constitute form-class of
“singular male substantive expressions”), must doubtless be
viewed as an example of a third class of grammatical forms, to
which we may give the name of substitutions.169

Chapter 11

11. 1. In any utterance, a linguistic form appears either as a
constituent of some larger form, as does John in the utterance
John ran away, or else as an independent form, not included in any
larger (complex) linguistic form, as, for instance, John in the
exclamation John! When a linguistic form occurs as part of a larger
form, it is said to be in included position; otherwise it is said to be in
absolute position and to constitute a sentence.

A form which in one utterance figures as a sentence, may in
another utterance appear in included position. In the exclamation
just cited, John is a sentence, but in the exclamation Poor John! the
form John is in included position. In this latter exclamation,
poor John is a sentence, but in the utterance Poor John ran away,
it is in included position. Or again, in the utterance just cited,
poor John ran away is a sentence, but in the utterance When the
dog barked poor John ran away
, it is in included position.

An utterance may consist of more than one sentence. This is the
case when the utterance contains several linguistic forms which
are not by any meaningful, conventional grammatical arrangement
(that is, by any construction) united into a larger form, e.g.: How
are you? It's a fine day. Are you going to play tennis this afternoon?

Whatever practical connection there may be between these three
forms, there is no grammatical arrangement uniting them into one
larger form: the utterance consists of three sentences.

It is evident that the sentences in any utterance are marked off
by the mere fact that each sentence is an independent linguistic
form, not included by virtue of any grammatical construction in
any larger linguistic form. In most, or possibly all languages, however,
various taxemes mark off the sentence, and, further, distinguish
different types of sentence.

In English and many other languages, sentences are marked off
by modulation, the use of secondary phonemes. In English,
secondary phonemes of pitch mark the end of sentences, and
distinguish three main sentence-types: John ran away [.] John
170ran away [?] Who ran away [¿]. To each of these, further, we may
add the distortion of exclamatory sentence-pitch, so that we get
in all, six types, as described in § 7.6.

This use of secondary phonemes to mark the end of sentences
makes possible a construction known as parataxis, in which two
forms united by no other construction are united by the use of
only one sentence-pitch. Thus, if we say It's ten o'clock [.] I have
to go home
[.] with the final falling pitch of a statement on o'clock,
we have spoken two sentences, but if we omit this final-pitch
(substituting for it a pause-pitch), the two forms are united, by
the construction of parataxis, into a single sentence: It's ten
[,] I have to go home [.]

Another feature of sentence-modulation in English and many
other languages, is the use of a secondary phoneme to mark
emphatic parts of a sentence. In English we use highest stress
for this (“Now it's my turn,” §7.3). The emphatic element in
English may be marked also by the use of special constructions
(It was John who did that) and by word-order (Away he ran);
in languages where stress is not significant, such methods prevail,
as in French C'est Jean qui l'afait [s ɛ žɑn ki l a fɛ] ‘It is John who
did it.’ Some languages use special words before or after an emphatic
element, as Tagalog [ikaw ˈŋaʔ aŋ nagˈsa:bi nijan] ‘you
(emphatic particle) the one-who-said that,’ i.e. ‘You yourself said
so;’ Menomini [ˈjo:hpeh ˈniw, kan ˈwenah ˈwa:pah] ‘Today (emphatic
particle), not (emphatic particle) tomorrow.’ Our high
stress can even strike forms that are normally unstressed: of, for,
and by the people; immigration and emigration.

11. 2. Beside features of modulation, features of selection may
serve to mark off different sentence-types. This is the case in some
of the examples just given, where a special construction, or the use
of a special particle, marks an emphatic element. In English,
supplement-questions are distinguished not only by their special
pitch-phoneme [¿], but also by a selective taxeme: the form used as
a supplement-question either consists of a special type of word or
phrase, which we may call an interrogative substitute, or else contains
such a word or phrase; Who? With whom? Who ran away?
With whom was he talking?

Perhaps all languages distinguish two great sentence-types
which we may call full sentences and minor sentences. The difference
consists in a taxeme of selection: certain forms are favorite
171sentence-forms; when a favorite sentence-form is used as a sentence,
this is a full sentence, and when any other form is used as
a sentence, this is a minor sentence. In English we have two favorite
sentence-forms. One consists of actor-action phrases —
phrases whose structure is that of the actor-action construction:
John ran away. Who ran away? Did John run away? The other
consists of a command — an infinitive verb with or without modifiers:
Come! Be good! This second type is always spoken with
exclamatory sentence-pitch; the infinitive may be accompanied
by the word you as an actor: You be good! As these examples
show, the meaning of the full sentence-type is something like
‘complete and novel utterance’ — that is, the speaker implies
that what he says is a full-sized occurrence or instruction, and
that it somehow alters the hearer's situation. The more deliberate
the speech, the more likely are the sentences to be of the full
type. The nature of the episememe of full sentences has given
rise to much philosophic dispute; to define this (or any other)
meaning exactly, lies beyond the domain of linguistics. It is
a serious mistake to try to use this meaning (or any meanings),
rather than formal features, as a starting-point for linguistic

Quite a few of the present-day Indo-European languages agree
with English in using an actor-action form as a favorite sentence-type.
Some, such as the other Germanic languages and French,
agree also in that the actor-action form is always a phrase, with
the actor and the action as separate words or phrases. In some
of these languages, however — for instance, in Italian and Spanish
and in the Slavic languages — the actor and the action are bound
forms which make up a single word: Italian canto [ˈkant-o] ‘I sing,’
canti [ˈkant-i] ‘thou singest,’ cant-a [ˈkant-a] ‘he (she, it) sings,’
and so on. A word which contains a favorite sentence-form of
its language is a sentence-word.

Some languages have different favorite sentence-types. Russian
has an actor-action type of sentence-word finite verbs, like
those of Italian: [poˈju] ‘I sing,’ [poˈjoš] ‘thou singest,’ [poˈjot]
‘he (she, it) sings,’ and so on. In addition to this, it has another
type of full sentence: [iˈvan duˈrak] ‘John (is) a fool,’ [solˈdat ˈxrabr]
‘the soldier (is) brave,’ [oˈtets ˈdoma] ‘Father (is) at home.’
In this second type, one component, which is spoken first, is
a substantive; the other form is a substantive to which the first
172is equated, or an adjective (adjectives have a special form for
this use), or an adverbial form.

When a language has more than one type of full sentence, these
types may agree in showing constructions of two parts. The common
name for such bipartite favorite sentence-forms is predications.
In a predication, the more object-like component is called
the subject, the other part the predicate. Of the two Russian types,
the former is called a narrative predication, the latter an equational
predication. For a language like English or Italian, which has
only one type of bipartite sentence, these terms are superfluous,
but often employed: John ran is said to be a predication, in
which the actor (John) is the subject and the action (ran) the

Latin had the same types of full sentence as Russian, but the
narrative type existed in two varieties: one with an actor-action
construction: cantat ‘he (she, it) sings,’ amat ‘he (she, it) loves,’
and one with a goal-action construction: cantātur ‘it is being sung,’
amātur ‘he (she, it) is loved.’ The equational type was less common
than in Russian: beātus ille ‘happy (is) he.’

Tagalog has five types of predication, with this common feature:
either the subject precedes and a particle [aj] (after vowels,
[j]) intervenes, or the reverse order is used without the particle.

There is, first, an equational type: [aŋ ˈba:ta j mabaˈit] ‘the
child is good,’ or, with inverse order, [mabaˈit aŋ ˈba:aʔ] ‘good
(is) the child.’ Then there are four narrative types, in which the
predicates are transient words, which denote things in four different
relations to an action. The four types of transient words

actor: [puˈmuːtul] ‘one who cut’

goal: [piˈnuːtul] ‘something cut’

instrument: [ipiˈnuːtul] ‘something cut with’

place: [pinuˈtuːlan] ‘something cut on or from.’

These transient words are by no means confined, like our verbs,
to predicative position; they can figure equally well, for instance,
in equational sentences, as: [aŋ puˈmuːtul aj si ˈhwan] ‘the one
who did the cutting was John,’ but in the predicate position they
produce four types of narrative predication:

actor-action: [sja j puˈmuːtul naŋ ˈka:huj] ‘he cut some wood’

goal-action: [piˈnuːtul nja aŋ ˈka:huj] ‘was-cut by-him the
wood,’ i.e. ‘he cut the wood’173

instrument-action: [ipiˈnuːtul nja aŋ ˈguːluk] ‘was-cut-with by-him
the bolo-knife,’ i.e. ‘he cut with the

place-action: [pinuˈtuːlan nja aŋ ˈka:huj] ‘was-eut-from by-him
the wood,’ i.e. ‘he cut (a piece) off the wood.’

Georgian distinguishes between an action-type, as [ˈv-tsʔter]
‘I-write’ and a sensation-type, as [ˈm-e-smi-s) ‘me-sound-is,’ i.e.
‘I hear.’ Such distinctions are never carried out with scientific
consistency; Georgian classifies sight in the action-type: [ˈv-naxav]

Not all favorite sentence-forms have bipartite structure: the
command in English consists of merely an infinitive form (come;
be good) and only occasionally contains an actor (you be good).
In German, beside a favorite sentence-type of actor-action which
closely resembles ours, there is an impersonal variety, which differs
by not containing any actor: mir ist kalt [miːr ist ˈkalt] ‘to-me
is cold,’ that is, ‘I feel cold;’ hier wird getanzt [ˈhiːr virt geˈtantst]
‘here gets danced,’ that is, ‘there is dancing here.’ In Russian,
there is an impersonal type which differs from the equational
predication by the absence of a subject: [ˈnužno] ‘it is necessary.’

11. 3. English has a sub-type of full sentences which we may
call the explicit-action type; in this type the action centers round
the verb do, does, did. This taxeme of selection appears in the
contrast between, say, I heard him and I did hear him. The explicitaction
type has several uses. When the verb is an emphatic element
(spoken with highest stress), the normal type emphasizes
the lexical content (the sememe) of the verb, as in “I heard him”
(but did not see him), or in “Run home!” (don't walk); the explicitaction
type emphasizes the occurrence (as opposed to nonoccurrence)
or the time (present or past) of the action, as in “I
did hear him,” or “Do run home!” Secondly, we use the explicitaction
type wherever the verb is modified by not, as in I didn't
hear him
or Don't run away; thus, English, by a taxeme of selection,
distinguishes a negative type of full sentence.

Further, within our explicit-action type, we distinguish a subtype
in which the verb do, does, did precedes the actor. This inverted
type occurs in formal yes-or-no questions, along with
question-pitch; Did John run away? Didn't John run away? in
contrast with the uninverted (informal) type: John ran away?
John didn't run away?174

The features just discussed are not so widely paralleled among
languages as the more general characteristics of English full
sentences. In German, for instance, the negative adverb is not
tied up with a special-sentence-type: Er kommt nicht [e:r ˈkomt
ˈnixt] ‘he comes not’ is like Er kommt bald [e:r ˈkomt ˈbalt] ‘he
comes soon.’ Other languages, however, resemble English in
using special sentence-types with negative value. In Finnish,
negative sentences have a special construction: the verb (which,
as in Italian, includes actor and action in one sentence-word)
is a special negative verb, which may be modified by an infinitive-like
form of another verb:

tableau luen | ‘I read’ | en lue | ‘I-don't read’ | luet | ‘thou readest’ | et lue | ‘thou-dost-not read’ | lukee | ‘he reads’ | ei lue | ‘he-doesn't read.’

In Menomini there are three main types of full sentence, equational,
narrative, and negative:

narrative: [piːw] ‘he-comes’

equational: [enuʔ pajiat] ‘he — the one who comes,’ that is,
‘It's he that's coming’

negative: [kan upianan] ‘not he-comes (negative),’ that is, ‘He
does not come.’

In the negative type the two parts are, on the one side, the negative
word [kan] in its various inflections and, on the other, the
rest of the sentence, marked by the use of special verb-forms.

Special types of full sentences for formal questions are more
widespread. German uses actor-action forms in which the verb
precedes the actor: Kommt er? [ˈkomt e:r?] ‘comes he?’ in contrast
with Er kommt [e:r ˈkomt] ‘he comes.’ French also uses special
interrogative constructions: ‘Is John coming?’ is either Jean
[žɑn vjɛnt I?] ‘John comes he?’ or Est-ce que Jean vient?
[ɛ s kə žɑn vjɛn?] ‘Is it that John comes?’ In Menomini the three
main types of full sentence have each an interrogative sub-type:

narrative: [piːʔ?] ‘Is he coming?’

equation: [enut pajiat?] ‘he (interrogative) the one who comes?’
that is, ‘Is it he that is coming?’

negative: [kanɛ:ʔ upianan?] ‘not (interrogative) he-comes (negative)?’
that is, ‘Isn't he coming?’

Other languages lack a special sentence-type for formal yes-or-no
questions, but some of them use special interrogative words, as
Latin venitne? [weˈnit ne?] ‘Is he coming?’ and num venit? ‘You
175don't mean to say he is coming?’ (expectation of negative reply),
in contrast with venit? ‘He is coming?’ This use of special little
words (particles) to mark a formal yes-or-no question, appears in
many languages, such as Russian, Chinese, Tagalog, Cree.

Most languages agree with English in marking supplement-questions
by the presence of special words, but the details differ:
in Tagalog and in Menomini, for instance, the supplement-question
is always an equational sentence, e.g., Menomini [awɛ:ʔ pajiat¿]
‘who the-one-who-comes?’ that is, ‘Who is coming?’

The English command is an example of a special sentence-type
used in exclamations. Other languages also have special types of
full sentence for some kinds of exclamations. In Menomini there
are two such, one of surprise, where the occurrence is new or
unforeseen, and one of disappointment at the non-occurrence of
something expected:

narrative: [piasah!] ‘and so he's coming!’
equational: [enusaʔ pajiat!] ‘and so it's he that's coming!’
negative: [kasaʔ upianan!] ‘and so he isn't coming!’

narrative: [piapah!] ‘but he was coming!’
equational: [enupaʔ pajiat!] ‘but he was the one who was
negative: [kapaʔ upianan!] ‘but he wasn't coming!’

11. 4. A sentence which does not consist of a favorite sentence-form
is a minor sentence. Some forms occur predominantly as
minor sentences, entering into few or no constructions other than
parataxis; such forms are interjections. Interjections are either
special words, such as ouch, oh, sh, gosh, hello, sir, ma'm, yes, or
else phrases (secondary interjections), often of peculiar construction,
such as dear me, goodness me, goodness gracious, goodness sakes alive,
oh dear, by golly, you angel, please, thank you, good-bye.

In general, minor sentences seem to be either completive or
exclamatory. The completive type consists of a form which merely
supplements a situation — that is, an earlier speech, a gesture, or
the mere presence of an object: This one. Tomorrow morning.
Gladly, if I can. Whenever you're ready. Here. When? With whom?
Mr. Brown: Mr. Smith (in introducing people). Drugs. State
. They occur especially as answers to questions; for this use
176we have the special completive interjections, yes and no. Even in
this regard languages differ: French says si ‘yes’ in answer to
negative questions, such as ‘Isn't he coming?’ but oui [wi] ‘yes’
in answer to others, such as ‘Is he coming?’ Some languages have
no such interjections. Polish answers with ordinary adverbs,
affirmatively with tak ‘thus, so’ and negatively with nie [ne]
‘not.’ Finnish answers affirmatively by an ordinary form, e.g.
Tulette-ko kawpungista?Tulernme. ‘Are you coming from town?’
— ‘We are coming,’ and negatively by its negative verb: Tunnetteko
herra Lehdon?
En (or En tunne) ‘Do you know Mr. Lehto?’
— ‘I don't’ (or ‘I don't know’).

Exclamatory minor sentences occur under a violent stimulus.
They consist of interjections or of normal forms that do not belong
to favorite sentence-types, and often show parataxis: Ouch, damn
This way, please! A substantive form naming a hearer is used
in English as a demand for his presence or attention: John! Little
You with the glasses! With parataxis: Hello, John! Come here,
little boy!
The interjections sir and ma'am are especially devoted to
this use; in the same way Russian uses an interjection [s], as
[da-s] ‘yes, sir; yes, ma'am,’ without distinction of sex. Many
languages have special vocative forms for this use, as Latin Balbus
(man's name), vocative Balbe, or Fox [iškwɛ:wa] ‘woman,’ vocative
[iškwe], and [iškwɛ:wak] ‘women,’ vocative [iškwɛ:tike]. In
Menomini the terms of relationship have special, highly irregular
vocative forms: [nɛʔnɛh] ‘my older brother,’ vocative [nanɛ:ʔ] or
[nekiːjah] ‘my mother,’ vocative [ne?ɛ:h]. Other words are spoken
as vocatives with short vowels instead of long: [mɛtɛ:rnuh] ‘woman,’
vocative [mɛtɛmuh]. In Sanskrit, vocative forms were

Occasionally we find minor sentences of aphoristic type (§ 9.9)
used with much the same value as full sentences; English examples
are The more you have, the more you want. The more, the merrier.
First come, first served. Old saint, young sinner.

11. 5. In most languages the sentence is characterized also by a
selective feature more general than all those we have been discussing:
some linguistic forms, which we call bound forms (§ 10.1),
are never used as sentences. English examples are the -ess [es] in
countess, lioness, duchess, etc., or the -ish [iš] in boyish, childish,
greenish, etc., or the -s [s] in hats, books, cups, etc. These are genuine
linguistic forms and convey a meaning, but they occur only in
177construction, as part of a larger form. Forms which occur as
sentences are free forms. Not every language uses bound forms:
modern Chinese, for instance, seems to have none.

A free form which consists entirely of two or more lesser free
forms, as, for instance, poor John or John ran away or yes, sir, is
a phrase. A free form which is not a phrase, is a word. A word, then,
is a free form which does not consist entirely of (two or more) lesser
free forms; in brief, a word is a minimum free form.

Since only free forms can be isolated in actual speech, the word,
as the minimum of free form, plays a very important part in our
attitude toward language. For the purposes of ordinary life, the
word is the smallest unit of speech. Our dictionaries list the words
of a language; for all purposes except the systematic study of
language, this procedure is doubtless more useful than would be a
list of morphemes. The analysis of linguistic forms into words is
familiar to us because we have the custom of leaving spaces between
words in our writing and printing. People who have not
learned to read and write, have some difficulty when, by any
chance, they are called upon to make word-divisions. This difficulty
is less in English than in some other languages, such as
French. The fact that the spacing of words has become part of our
tradition of writing, goes to show, however, that recognition of the
word as a unit of speech is not unnatural to speakers; indeed,
except for certain doubtful cases, people easily learn to make this

In our school tradition we sometimes speak of forms like book,
books, or do, does, did, done as “different forms of the same word.”
Of course, this is inaccurate, since there are differences of form
and meaning between the members of these sets: the forms just
cited are different linguistic forms and, accordingly, different

In other cases, inconsistencies in our habits of writing may make
us uncertain. We write John's in John's ready, where it is two
words (John and [z], an alternant of is) and in John's hat, where it
is one word (consisting of John and the bound form [-z], possessive).
We write the boy's as though it were two or three words, but,
strictly speaking, it is only one word, since the immediate constituents
are the boy and [-z] possessive, and the latter is a bound
form; this appears clearly in cases like the king of England's or
the man I saw yesterday's, where the meaning shows that the [-z]
178is in construction with the entire preceding phrase, so that the
two are united into a single long word.

11. 6. In the case of many languages, however, it is impossible to
distinguish consistently, on the one hand, between phrases and
words and, on the other hand, between words and bound forms.
The linguist cannot wait indefinitely for the chance of hearing a
given form used as a sentence — that is, spoken alone. Some forms
are rarely so used. Inquiry or experiment may call forth very
different responses from hearers. Are English forms like the, a, is,
and ever spoken alone? One can imagine a dialogue: Is?No;
was. The word because is said to be a woman's answer. An impatient
listener says And? We can imagine a hesitant speaker who
says The… and is understood by his hearers. Aside from such
far-fetched situations, the general structure of a language may
make one classification more convenient than another for our
purpose. The form the, though rarely spoken alone, plays much
the same part in our language as the forms this and that, which
freely occur as sentences; this parallelism leads us to class the
as a word:

this thing : that thing : the thing
: that : (the).

In other cases, the difficulty is due to features of phonetic modification.
The forms [z] in John's ready, [m] in I'm hungry, or [nt]
in Don't! are unpronounceable in English, but we have to class
them as words, for they are merely alternants of the pronounceable
forms is, am, not. In French we have even the case of a single
phoneme representing two words: au [o] in a phrase like au roi
[o rwa] ‘to the king,’ arises by phonetic modification of the two
words à [a] ‘to’ and le [lə] ‘the;’ this [o] is homonymous with the
words eau ‘water’ and haut ‘high.’

In other cases the doubtful forms are units of grammatical
selection rather than of modification, and yet, in view of the total
structure of their language, may be best classified as words. French,
again, has several forms of this sort. Absolute forms like moi
[mwa] ‘I, me’ and lui [Iɥi] ‘he, him’ are replaced in certain constructions
by shorter forms that do not ordinarily appear in
absolute use, such as je [žə] ‘I,’ me [mə] ‘me,’ il [il] ‘he,’ le [lə]
‘him;’ for instance: je le connais [žə l kɔnɛ] ‘I know him,’ il me
[i m kɔnɛ] ‘he knows me.’ The replacement of the absolute
forms by these conjunct forms is to be described as a feature of
179selection rather than of modification; nevertheless, the conjunct
forms, largely because of their parallelism with the absolute forms,
have the status of words.

A less important border-line case is the use of bound forms in
hypostasis (§ 9.7), as when we speak of a girl in her teens, taking
up all kinds of isms and ologies.

At the other extreme we find forms which lie on the border between
words and phrases. A form like blackbird resembles a two-word
phrase (black bird), but we shall find that a consistent
description of English is bound to class this form as a single (compound)
word. In this case there is a clear-cut difference, since in
blackbird the second word (bird), has a weaker stress instead of a
normal high stress, a difference which in English is phonemic, and
this formal difference correlates with the semantic difference between
blackbird and black bird. The distinction is not always so
clear: ice-cream [ˈajsˌkrijm], spoken with only one high stress, will
be classed as a (compound) word, but the variant pronunciation
ice cream [ˈajs ˈkrijm], with two high stresses, will be classed as a
two-word phrase. Similar variants exist in types like messenger
, lady friend.

This criterion of stress fails us in forms like devil-may-care (as
in a devil-may-care manner) or jack-in-the-pulpit (as the name of a
plant). If the former were devil-may-care-ish, we should not hesitate
to class it as a word, since here one of the immediate constituents
is the bound form -ish. The forms of the type devil-may-care
are classed as words (phrase-words) because of certain other
features which, within the system of the English language, place
them on a level with other words. One of these is their peculiar
function; as a phrase devil-may-care would be an actor-action form,
but as a phrase-word it fills the position of an adjective. Another is
their indivisibility: the plant-name jack-in-the-pulpit cannot be
modified by putting the word little in front of pulpit, but the corresponding
phrase permits of this and other expansions.

This latter principle, namely that a word cannot be interrupted
by other forms, holds good almost universally. Thus, one can
say black — I should say, bluish-black — birds, but one cannot
similarly interrupt the compound word blackbirds. The exceptions
to this principle are so rare as to seem almost pathological. Gothic
had a bound form [ga-] which was prefixed especially to verbs:
[ˈse:hwi] ‘he should see,’ [gaˈse:hwi] ‘he should be able to see.’
180Yet occasionally we find words included between this [ga-] and
the main body of the verb, as in the translation of Mark 8, 23:
[ˈfrah ina ga- u hwa ˈse:hwi] ‘he asked him whether [u] he saw
anything [hwa].’

None of these criteria can be strictly applied: many forms lie
on the border-line between bound forms and words, or between
words and phrases; it is impossible to make a rigid distinction
between forms that may and forms that may not be spoken in
absolute position.

11. 7. The word is not primarily a phonetic unit: we do not,
by pauses or other phonetic features, mark off those segments
of our speech which could be spoken alone. In various ways, however,
different languages give phonetic recognition to the word-unit:
some, like French, very little, and others, like English, very

As a free form, the word is capable of being spoken in absolute
position; accordingly, it is subject to the phonetic patterning of
its language. It is sure to contain at least one of the phonemes
which normally serve as syllabics; interjections, such as our sh
[š] and pst [pst], occasionally violate this principle. The initial
and final consonants and clusters in the word are necessarily such
as can occur at the beginning and at the end of speech; thus, no
English word begins with [ŋ] or [mb] and none ends with [h] or

Beyond this, many languages place further restrictions on the
phonetic structure of the word. We may find that some of the
permitted medial clusters do not occur within the body of a single
word; in English, permitted clusters like [šč, vt, tsv, ststr, as
in rash child, give ten, it's very cold, least strong, and double consonants,
like [nn, tt, bb], as in ten nights, that time, nab Bill, do
not occur within simple words. On the other hand, French, with
its insertion of [9], and languages like Fox or Samoan, which use
no final consonants, tolerate no more clusters within a phrase
than within a word.

Some languages have the peculiar restriction, known as vowelharmony,
of tolerating only certain combinations of vowels in the
successive syllables of a word. Thus, in Turkish, the vowels of a
word are either all front vowels [i, y, e, ø], as in [sevildirememek]
‘not to be able to cause to be loved,’ or all back vowels [ï, u, a, o],
as in [jazïldïramamak] ‘not to be able to cause to be written.’181

In Chinese we have the extreme of structural word-marking;
each word consists of one syllable and of two or three primary
phonemes: a non-syllabic simple or compound phoneme as initial,
a syllabic simple or compound phoneme as final; and one of the
pitch-schemes (§ 7.7); the initial non-syllabic may be lacking;
the language has no bound forms.

In English and many other languages, each word is marked by
containing one and only one high stress (forgiving; convict, verb;
convict, noun). In some of these languages the word-unit is even
more plainly marked, in that the position of a word-stress bears
a definite relation to the beginning or to the end of the word: in
Bohemian and in Icelandic the first syllable is stressed, in Cree
the third-last (the antepenult), in Polish the next-to-last (the
penult). In Latin the penult was stressed, as in amāmus [aˈma:mus]
‘we love,’ unless this syllable had a short vowel followed by
no more than one consonant, in which case the antepenult was
stressed, as in capimus [ˈkapimus] ‘we take.’ In languages like
these, the stress is a word-marker, which indicates the beginnings
or ends of words, but, since its position is fixed, it cannot distinguish
between different words. In Italian, Spanish, and modern
Greek, the stress comes always on one of the last three syllables
of a word. In ancient Greek a word had either a simple accent
on one of the last three syllables or a compound accent on one of
the last two, with some further restrictions based on the nature of
the primary phonemes in these syllables.

Among stress-using languages, some, like English, start the
stress at the beginning of a word whose stress comes on the first
syllable; witness contrasts like a name versus an aim or that scold
versus that's cold (§ 7.5); others, such as Dutch, Italian, Spanish,
and the Slavic languages, regulate the onset of stress by purely
phonetic habits, starting the stress on a consonant which precedes
a stressed vowel, even though this consonant belongs to another
word, as in Italian un altro [uˈn altro] ‘another.’ A language like
French, which uses no stress-phonemes, cannot in this way mark
its word-units.

Phonetic recognition of the word-unit, in cases like the above>
is disturbed chiefly by two factors. Words which contain, among
their ultimate constituents, two or more free forms, generally
have the phonetic character of phrases. In English, compound
words have the same medial clusters as phrases: stove-top [vt],
182chest-strap [ststr], pen-knife [nn], grab-bag [bb]; phrase-derivatives
may even have more than one high stress: old-maidish [ˈowld
ˈmejdiš], jack-in-the-pulpit [ˈǰɛk in ðe ˈpulpit].

On the other hand, words in included position are subject to
modulations and phonetic modifications which may remove the
phonetic characteristics of word-marking. Thus not in the phrase
don't [ˈdow nt] loses both its high stress and its syllabic; the phrase
can't is homonymous with the word cant; compare, similarly, lock it,
with locket, feed her [ˈfijd r̩] with feeder, and so on. In the normal
pronunciation at all [eˈt ɔl] the stress begins on the [t] of at. These
included variants, in which a word loses the phonetic features
that characterize words in absolute position, will concern us in
the next chapter. In the present connection it is worth noticing,
however, that in a small way these modified phrases may nevertheless
involve phonetic recognition of the word-unit, because they
contain phonetic sequences that do not occur in single words.
Thus, the final sequence [ownt] is permitted in English, but occurs
only in the phrases don't and won't, and not in any one word.
In South German dialects some initial clusters, such as [tn, tšt] occur
in phrases, thanks to phonetic modification of the first word,
as in [t nɑxt] ‘the night,’ [t štɑ:št] ‘thou standest,’ but not in any
one word. In North Chinese a phrase may end in syllabic plus
[r], as in [çjaw3 ˈma r3] ‘little horse,’ but only as a result of phonetic
modification of two words, — in our example, [ma3] ‘horse’ and
[r2] ‘son, child, small,’

In the few languages which use no bound forms, the word has
a double importance, since it is the smallest unit not only of free
form but also of linguistic form in general. In languages which
use bound forms, the word has great structural importance because
the constructions in which free forms appear in phrases
differ very decidedly from the constructions in which free or
bound forms appear in words. Accordingly, the grammar of these
languages consists of two parts, called syntax, and morphology.
However, the constructions of compound words and, to some extent,
of phrase-derivatives, occupy an intermediate position.183

Chapter 12

12. 1. Traditionally, the grammar of most languages is discussed
under two heads, syntax and morphology. The sentence-types,
which we surveyed in the last chapter, are placed under
the former heading, and so are the types of substitution (which
we shall consider in Chapter 15), but grammatical constructions,
which we shall now examine, are dealt with partly under the heading
of morphology. There has been considerable debate as to
the usefulness of this division, and as to the scope of the two headings.
In languages that have bound forms, the constructions in
which bound forms play a part differ radically from the constructions
in which all the immediate constituents are free forms. Accordingly,
we place the former under the separate heading of morphology.
The difficulty is this, that certain formal relations, such
as the relation between he and him, consist in the use of bound
forms, while the semantic difference between these forms can be
defined in terms of syntactic construction; he serves, for instance,
as an actor (he ran) and him as an undergoer (hit him). Nevertheless,
the traditional division is justified: it merely happens that in
these cases the meanings involved in the morphologic construction
are definable in terms of syntax instead of being definable merely
in terms of practical life. Syntactic constructions, then, are constructions
in which none of the immediate constituents is a bound
form. Border-line cases between morphology and syntax occur
chiefly in the sphere of compound words and phrase-words.

12. 2. The free forms (words and phrases) of a language appear
in larger free forms (phrases), arranged by taxemes of modulation,
phonetic modification, selection, and order. Any meaningful,
recurrent set of such taxemes is a syntactic construction. For
instance, the English actor-action construction appears in phrases
like these:

tableau John ran | Bill fell | John fell | Our horses ran away | Bill ran184

In these examples we see taxemes of selection. The one constituent
(John, Bill, our horses) is a form of a large class, which we
call nominative expressions; a form like ran or very good could not
be used in this way. The other constituent (ran, fell, ran away) is
a form of another large class, which we call finite verb expressions;
a form like John or very good could not be used in this way. Secondly,
we see a taxeme of order: the nominative expression precedes
the finite verb expression. We need not stop here to examine the
various other types and sub-types of this construction, which show
different or additional taxemes. The meaning of the construction
is roughly this, that whatever is named by the substantive expression
is an actor that performs the action named by the finite verb
expression. The two immediate constituents of the English actor-action
construction are not interchangeable: we say that the
construction has two positions, which we may call the positions of
actor and of action. Certain English words and phrases can appear
in the actor position, certain others in the action position. The
positions in which a form can appear are its functions or, collectively,
its function. All the forms which can fill a given position
thereby constitute a form-class. Thus, all the English words and
phrases which can fill the actor position in the actor-action construction,
constitute a great form-class, and we call them nominative
expressions; similarly, all the English words and phrases
which can fill the action position on the actor-action construction,
constitute a second great form-class, and we call them finite verb

12. 3. Since the constituents of phrases are free forms, the
speaker may separate them by means of pauses. Pauses are mostly
non-distinctive; they occur chiefly when the constituents are long
phrases; in English they are usually preceded by a pause-pitch.

We have seen (§ 11.1) that free forms which are united by no
other construction may be united by parataxis, the mere absence
of a phonetic sentence-final, as in It's ten o'clock [,] I have to go
[.] In ordinary English parataxis a pause-pitch appears between
the constituents, but we have also a variety of close parataxis
without a pause-pitch, as in please come or yes sir.

A special variety of parataxis is the use of semi-absolute forms,
which grammatically and in meaning duplicate some part of the
form with which they are joined in parataxis, as in John, he ran
. In French this type is regularly used in some kinds of
185questions, as Jean quand est-il venu? [žɑnnt ɛt i vny?] ‘John, when
did he come?’

Parenthesis is a variety of parataxis in which one form interrupts
the other; in English the parenthetic form is ordinarily preceded
and followed by a pause-pitch: I saw the boy [,] I mean Smith's boy
[,] running across the street [.] In a form like Won't you please come?
the please is a close parenthesis, without pause-pitch.

The term apposition is used when paratactically joined forms are
grammatically, but not in meaning, equivalent, e.g. John [,] the
poor boy
. When the appositional group appears in included position,
one of its members is equivalent to a parenthesis: John [,]
the poor boy [,] ran away [.] In English we have also close apposition,
without a pause-pitch, as in King John, John Brown, John the
, Mr. Brown, Mount Everest.

Often enough non-linguistic factors interfere with construction;
what the speaker has said is nevertheless meaningful, provided he
has already uttered a free form. In aposiopesis the speaker breaks
off or is interrupted: I thought he — . In anacolouthon he starts
over again: It's high time we — oh, well, I guess it won't matter.
When a speaker hesitates, English and some other languages offer
special parenthetic hesitation-forms, as [r] or [ɛ] in Mr. — ah —
or Mr. — what you may call him — Sniffen or that — thingamajig
— transmitter

12. 4. Features of modulation and of phonetic modification play
a great part in many syntactic constructions; they are known as
sandhi. 118 The form of a word or phrase as it is spoken alone is its
absolute form; the forms which appear in included positions are
its sandhi-forms. Thus, in English, the absolute form of the indefinite
article is a [ˈej]. This form appears in included position
only when the article is an emphatic element and the next word
begins with a consonant, as in “not a house, but the house.” If the
next word begins with a vowel, we have instead a sandhi-form, an
[ˈɛn], as in “not an uncle, but her uncle.”

A feature of modulation appears in the fact that when a, an
is not an emphatic element, it is spoken as an unstressed syllable,
as in a house [e ˈhaws], an arm [en ˈarm]. In English, a word in
absolute form has one high stress; hence we may say that in a
sandhi-form without high stress a word is spoken as if it were part
186of another word. Various languages use sandhi-forms of this sort;
they are known as atonic forms. This term is not altogether appropriate,
since the peculiarity is not always a lack of stress. In
the French phrase l'homme [l ɔm] ‘the man,’ the article le [lə] is
atonic, because its sandhi-form [l] could not be spoken alone on
account of the phonetic pattern (lack of a vowel). In the Polish
phrase [ˈdo nuk] ‘to the feet,’ the preposition do ‘to’ is atonic
precisely because it has the stress, for the stress in this language is
placed on the next-to-last syllable of each word, and falls on do
only because this word is treated as part of the following word.

An atonic form which is treated as part of the following word —
this is the case in our examples so far —- is a proclitic. An atonic
form which is treated as if it were part of the preceding word is an
enclitic; thus, in I saw him [aj ˈsɔ im], the [aj] is proclitic, but the
[im] enclitic.

The sandhi which substitutes an for a, and the sandhi by which
this and other words are unstressed in phrasal combinations, are
examples of compulsory sandhi. Other English sandhi habits are
optional, because paralleled by unaltered variants, which have
usually a formal or elevated connotation; for instance, the dropping
of [h] in him does not take place in the more elevated variant I
saw him
[aj ˈsɔ him]. Beside the sandhi-forms in did you? [ˈdiǰuw?],
won't you [ˈwownčuw?], at all [eˈtɔl] (in American English with the
voiced tongue-flip variant of [t]), we have the more elegant variants
[ˈdid juw? ˈwownt juw? et ˈɔl].

Sandhi-forms may be unpronounceable when taken by themselves;
this is the case in a number of English examples:

tableau absolute form | is | has | am | are | have | had | would | will | them | sandhi-form | John's ready | Dick's ready | John's got it | I'm ready | we're waiting | I've got it | he'd seen it | he'd see it | I'll go | that'll do. | watch 'em187

tableau absolute form | sandhi-form | not | it isn't | I won't | I can't | and | bread and butter

The French language has a great deal of sandhi. Thus, the
article la [la] ‘the’ (feminine) loses the [a] before a vowel or diphthong:
la femme [la fam] ‘the woman,’ but l'encre [l ɑnkr] ‘the ink,’
l'oie [l wa] ‘the goose.’ The adjective ce [sə] ‘this’ (masculine) adds
[t] before the same sounds: ce couteau [sə kuto] ‘this knife,’ but
cet homme [sət ɔm] ‘this man.’ A plural pronoun adds [z] before the
initial vowel of a verb: vous faites [vu fɛt] ‘you make,’ but vous êtes
[vuz ɛ:t] ‘you are.’ A plural noun-modifier behaves similarly:
les femmes [le fam] ‘the women,’ but les hommes [lez ɔm] ‘the men.’
A first-person or second-person verb adds [z], a third-person verb
[t], before certain initial vowels: va [va] ‘go thou,’ but vas-y [vaz i]
‘go thou there;’ elle est [ɛl ɛ] ‘she is,’ but est-elle? [ɛt ɛl?] ‘is she?’
A few masculine adjectives add sandhi-consonants before a vowel:
un grand garçonn grɑn garsɔn] ‘a big boy,’ but un grand homme
n grɑnt ɔm] ‘a great man.’

In languages with distinctions of pitch in the word, modifications
of pitch may play a part in sandhi. Thus, in Chinese, beside the
absolute form [i1] ‘one,’ there are the sandhi-forms in [ˌi4 phi2
ˈma3] ‘one horse’ and [i2 ko ˈžən2] ‘one man.’

Sandhi-modification of initial phonemes is less common than
that of the end of a word; it occurs in the Celtic languages, as, in
modern Irish:

tableau absolute form | sandhi-form | ‘cow’ | ‘the cow’ | ‘our cow’ | ‘egg’ | ‘the egg’ | ‘of the eggs’ | ‘her egg’ | ‘white’ | ‘white cow’ | ‘soft’ | ‘very soft’ | ‘break’ | ‘did break.’

12. 5. Our examples so far illustrate special or irregular cases
of sandhi, peculiar to certain forms and constructions. General
188or regular sandhi applies to any and all words in a short (close-knit)
phrase. In some forms of English, such as New England
and southern British, words which in absolute position have a
final vowel, add [r] before an initial vowel: water [ˈwɔtə] but the
water is
[ðe ˈwɔtər iz]; idea [ajˈdijə] but the idea is [ðij ajˈdijər
iz]. When three consonants come together in French, the word-final
adds [ə]; thus, porte [port] ‘carries’ and bien [bjɛn] ‘well’
appear in the phrase as porte bien [portə bjɛn] ‘carries well.’ A
word whose first syllable in absolute form contains [ə], either because
the word has no other syllabic or because otherwise it
would begin with an unpermitted cluster (§ 8.6), loses this [ə]
in the phrase whenever no unpermitted group would result: le
[lə] ‘the’ but l'homme [l ɔm] ‘the man;’ cheval [šəval] ‘horse,’
but un chevaln šval] ‘a horse;’ je [žə] ‘I,’ ne [nə] ‘not,’ le [lə]
‘it,’ demande [dəmɑnd] ‘ask,’ but je ne le demande pas [žə n lə
dmɑnd pa] ‘I don't ask it?’ and si je ne le demande pas [si ž nə l
dəmɑnd pa] ‘if I don't ask it.’

In Sanskrit there is a great deal of general sandhi; for instance,
final [ah] of the absolute form appears in the following sandhi-variants:
absolute [de:ˈvah] ‘a god,’ sandhi-forms: [de:ˈvas ˈtatra]
‘the god there,’ [derˈvaç čarati] ‘the god wanders,’ [de:ˈva e:ti]
‘the god goes,’ [de:ˈvo: dada:ti] ‘the god gives,’ and, with change
also of a following initial, before [ˈatra] ‘here,’ [de:ˈvo: tra] ‘the
god here.’ Certain words, however, behave differently; thus,
[ˈpunah] ‘again’ gives [ˈpunar dada:ti] ‘again he gives,’ [ˈpunar
ˈatra] ‘again here.’ The divergent words may be marked off by
some structural feature. Thus, in some Dutch pronunciations the
absolute forms heb [ˈhep] ‘have’ and stop [stop] ‘stop’ behave differently
in sandhi: heb ik? [ˈheb ek?] ‘have I?’ but stop ik? [ˈstop ek?]
‘do I stop?’ The forms which have the voiced consonant in sandhi
have it also whenever it is not at the end of the word, as hebben
[ˈhebe] ‘to have,’ in contrast with stoppen [ˈstope] ‘to stop.’
Sandhi-distinctions based on morphologic features like this, may
be called reminiscent sandhi.

Sandhi may go so far as to restrict the word-final in a phrase
beyond the ordinary medial restrictions of a language. Thus, the
sequence [ta] is permitted medially in Sanskrit, as in [ˈpatati] ‘he
falls,’ but [t] at the end of the word is in close-knit phrases replaced
by [d] before a vowel: absolute [ˈtat] ‘that,’ but [ˈtad asti]
‘that is.’189

12. 6. Taxemes of selection play a large part in the syntax of
most languages; syntax consists largely in defining them — in
stating, for instance, under what circumstances (with what accompanying
forms or, if the accompanying forms are the same,
with what difference of meaning) various form-classes (as, say,
indicative and subjunctive verbs, or dative and accusative nouns,
and so on) appear in syntactic constructions. We have seen that
the selective taxemes delimit form-classes. These classes are
most numerous in the languages that use most taxemes of
selection. The syntactic constructions of a language mark off
large classes of free forms, such as, in English, the nominative
expression or the finite verb expression. Since different languages
have different constructions, their form-classes also are
different. We shall see that the great form-classes of a language
are most easily described in terms of word-classes (such as the
traditional “parts of speech”), because the form-class of a phrase
is usually determined by one or more of the words which appear
in it.

In languages which make a wide use of selective taxemes, the
large form-classes are subdivided into smaller ones. For instance,
the English actor-action construction, in addition to the general
selective taxemes, shows some more specialized taxemes of the
same sort. With the nominative expressions John or that horse
we can join the finite verb expression runs fast, but not the finite
verb expression run fast; with the nominative expressions John
and Bill
or horses the reverse selection is made. Accordingly,
we recognize in each of these two form-classes a division into two
sub-classes, which we call singular and plural, such that a singular
nominative expression is joined only with a singular finite verb
expression, and a plural nominative expression only with a plural
finite verb expression. It would not do to define these sub-classes
by meaning — witness cases like wheat grows but oats grow. Further
examination shows us several varieties of selection: (1) many
finite verb expressions, such as can, had, went, appear with any
actor; (2) many, such as run : runs, show the twofold selection
just described; (3) one, was : were, shows a twofold selection that
does not agree with the preceding; (4) one, finally, am : is : are,
shows a threefold selection, with a special form that accompanies
the actor I, precisely the actor form as to which (2) and (3) disagree:190

tableau I can | I run | I was | I am | the boy can | the boy runs | the boy was | the boy is | the boys can | the boys run | the boys were | the boys are

Thus we find among nominative expressions and among finite
verb expressions a threefold subdivision, due to taxemes of selection;
among nominative expressions sub-class A contains only
the form I; sub-class B contains those which are joined with finite
verb expressions such as runs, was, is, and sub-class C contains
those which are joined with finite verb expressions such as run,
were, are. In fact, we can base our definition of the three subclasses
on the selection of the three finite verb forms am : is : are.
Conversely, we define the sub-classes of finite verb expressions by
telling with which nominative expressions (say, I : the boy : the
) they occur.

The narrower type of selection in cases like this one is in principle
no different from the more inclusive type by which our
language distinguishes great form-classes like nominative expressions
and finite verb expressions, but there are some differences
of detail. The narrower type of selection, by which great form-classes
are subdivided into selective types, is called agreement.
In a rough way, without real boundaries, we can distinguish three
general types of agreement.

12. 7. In our example, the agreement is of the simplest kind,
which is usually called concord or congruence: if the actor is a form
of sub-class A, the action must be a form of sub-class A, and so
on. Sometimes one of the subdivisions is otherwise also recognized
in the structure of the language; thus, in our example, classes
B and C of nominative expressions are otherwise also definable
in our language; namely, by the use of the modifiers this, that with
class B, but these, those with class C: we say this boy, this wheat,
but these boys, these oats. Accordingly, we view the subdivision of
nominative expressions into singulars and plurals as more fundamental
than that of finite verb expressions, and say that the latter
agree with or stand in congruence with the former. For the
same reason, we say that the forms this, that, these, those stand
in congruence with the accompanying substantive form. Congruence
plays a great part in many languages; witness for example
191the inflection of the adjectives in most Indo-European languages
in congruence with various sub-classes (number, gender, case)
of the noun: German der Knabe [der ˈkna:be] ‘the boy,’ ich sehe
den Knaben
[ix ˈze:e den ˈkna:ben] ‘I see the boy,’ die Knaben
[diː ˈkna:ben] ‘the boys,’ where the selection of der, den, die
agrees with the sub-classes of the noun (singular and plural,
nominative and accusative); in das Haus [das ˈhaws] ‘the house,’
the form das, as opposed to der, is selected in agreement with the
so-called gender-classes into which German nouns are divided.
These genders are arbitrary classes, each of which demands
different congruence-forms in certain kinds of accompanying
words. German has three gender-classes; for each of these I give
phrases showing the congruence of the definite article and of the
adjective halt ‘cold’:

“masculine gender”: der Hut [der ˈhuːt] ‘the hat,’ halter Wein
[ˌkalter ˈvajn] ‘cold wine’

“feminine gender”: die Uhr [diː ˈuːr] ‘the clock’ kalte Milch
[ˌkalte ˈmilx] ‘cold milk’

“neuter gender”: das Haus [das ˈhaws] ‘the house,’ kaltes Wasser
[ˌkaltes ˈvaser] ‘cold water.’

French has two genders, “masculine,” le couteau [lə kuto]
‘the knife,’ and “feminine,” la fourchette [la furšɛt] ‘the fork.’
Some languages of the Bantu family distinguish as many as twenty
gender-classes of nouns.

12. 8. In other cases the subsidiary taxeme of selection has to
do with the syntactic position of the form. For instance, we say
I know but watch me, beside me. The choice between the forms I
(he, she, they, we) and me (him, her, them, us) depends upon the
position of the form: the I-class appears in the position of actor,
the we-class in the position of goal in the action-goal construction
(watch me) and in the position of axis in the relation-axis construction
(beside me). This type of selection is called government; the
accompanying form (know, watch, beside) is said to govern (or to
demand or to take) the selected form (I or me). Government, like
congruence, plays a great part in many languages, including many
of the Indo-European family. Thus, in Latin, different verbs
govern different case-forms in the substantive goal: videt bovem
‘he sees the ox,’ nocet bovī ‘he harms the ox,’ ūtitur bove ‘he uses
the ox,’ meminit bovis ‘he remembers the ox.’ Similarly, different
main clauses may govern different forms of subordinate verbs,
192as in French je pense qu'il vient [žə pɑns k i vjɛn] ‘I think he is coming,’
but je ne pense pas qu-il vienne [žə n pɑns pa k i vjɛn] ‘I don't
think he is coming.’

Identity and non-identity of objects are in many languages
distinguished by selective features akin to government. In English
we say he washed him when actor and goal are not identical, but
he washed himself (a reflexive form) when they are the same person.
Swedish thus distinguishes between identical and non-identical
actor and possessor: han tog sin hatt [han ˈto:g siːn ˈhat] ‘he took
his (own) hat’ and han tog hans hatt [hans ˈhat] ‘his (someone else's)
hat.’ The Algonquian languages use different forms for non-identical
animate third persons in a context. In Cree, if we speak
of a man and then, secondarily, of another man, we mention the
first one as [ˈna:pe:w] ‘man,’ and the second one, in the so-called
obviative form, as [ˈna:pe:wa]. Thus, the language distinguishes
between the following cases, where we designate the principal
person as A and the other (the obviative) as B:

[ˈutinam uˈtastutin] ‘he (A) took his (A's) hat’
[ˈutinam utastuˈtinijiw] ‘he (A) took his (B's) hat’
[utinaˈmijiwa uˈtastutin] ‘he (B) took his (A's) hat’
[utinaˈmijiwa utastuˈtinijiw] ‘he (B) took his (B's) hat.’

12. 9. In the third type of agreement, cross-reference, the subclasses
contain an actual mention of the forms with which they are
joined. This mention is in the shape of a substitute-form, resembling
our pronouns. In non-standard English this occurs in such
forms as John his knife or John he ran away; here the form his knife
actually mentions a male possessor, who is more explicitly mentioned
in the accompanying semi-absolute form John; similarly,
the he in he ran away mentions the actor John — contrast Mary
her knife
and Mary she ran away. In French, cross-reference occurs
in the standard language especially in certain types of questions,
such as Jean où est-il? [žɑn u ɛt i?] ‘John where is he?’ that is,
‘Where is John?’ (§ 12.3). A Latin finite verb, such as cantat
‘he (she, it) sings,’ includes substitutive mention of an actor. It
is joined in cross-reference with a substantive expression that
makes specific mention of the actor, as in puella cantat ‘(the) girl
she-sings.’ In many languages verb-forms include substitutive
(pronominal) mention of both an actor and an undergoer, as, in
Cree [ˈwa:pame:w] ‘he saw him or her;’ accordingly, more specific
193mention of both actor and undergoer is in cross-reference [ˈwa:pame:w
ˈatimwa aˈwa na:pe:w] ‘he-saw-him (obviative) a-dog
(obviative) that man;’ that is, ‘the man saw a dog.’ Similarly, in
many languages, a possessed noun includes pronominal mention of
a possessor, as, in Cree, [ˈastutin] ‘hat,’ but [niˈtastutin] ‘my hat,’
[kiˈtastutin] ‘thy hat,’ [uˈtastutin] ‘his, her, its hat;’ hence, when
the possessor is mentioned in another word or phrase, we have
cross-reference, as in [ˈčan uˈtastutin] ‘John his-hat,’ i.e. ‘John's

12. 10. Every syntactic construction shows us two (or sometimes
more) free forms combined in a phrase, which we may call
the resultant phrase. The resultant phrase may belong to a form-class
other than that of any constituent. For instance, John ran
is neither a nominative expression (like John) nor a finite verb
expression (like ran). Therefore we say that the English actor-action
construction is exocentric: the resultant phrase belongs to the
form-class of no immediate constituent. On the other hand, the
resultant phrase may belong to the same form-class as one (or more)
of the constituents. For instance, poor John is a proper-noun
expression, and so is the constituent John; the forms John and
poor John have, on the whole, the same functions. Accordingly,
we say that the English character-substance construction (as in
poor John, fresh milk, and the like) is an endocentric construction.

The exocentric constructions in any language are few. In
English we have, beside the actor-action construction, also that of
relation-axis, as beside John, with me, in the house, by running
; the constituents are a prepositional expression and an accusative
expression, but the resultant phrase has a function different
from either of these, appearing in entirely different syntactic
positions (e.g. as a modifier of verbs: sit beside John, or of nouns:
the boy beside John). Another exocentric construction of English is
that of subordination. The constituents in one type (clause-subordination)
are a subordinating expression and an actor-action
phrase, as in if John ran away; the resultant phrase has the function
of neither constituent, but serves as a modifier (subordinate clause).
In the other type (phrase-subordination) the constituents are a
subordinating expression and any other form, especially a substantive:
as I, than John, and the resultant phrase has the function
of a modifier (as big as I, bigger than John). Although the resultant
phrase in an exocentric construction has a function different from
194the function of any constituent, yet one of these constituents is
usually peculiar to the construction and serves to characterize the
resultant phrase; thus, in English, finite verbs, prepositions, and
subordinating conjunctions regularly appear in the exocentric
constructions just illustrated, and suffice to characterize them.

Endocentric constructions are of two kinds, co-ordinative (or
serial) and subordinative (or attributive). In the former type the
resultant phrase belongs to the same form-class as two or more of
the constituents. Thus, the phrase boys and girls belongs to the
same form-class as the constituents, boys, girls; these constituents
are the members of the co-ordination, and the other constituent is
the co-ordinator. Sometimes there is no co-ordinator: books, papers,
pens, pencils, blotters (were all lying…); sometimes there is one
for each member, as in both Bill and John, either Bill or John.
There may be minor differences of form-class between the resultant
phrase and the members; thus Bill and John is plural, while the
members are each singular.

In subordinative endocentric constructions, the resultant phrase
belongs to the same form-class as one of the constituents, which
we call the head: thus, poor John belongs to the same form-class as
John, which we accordingly call the head; the other member, in
our example poor, is the attribute. The attribute may in turn be a
subordinative phrase: in very fresh milk the immediate constituents
are the head milk, and the attribute very fresh, and this phrase, in
turn, consists of the head fresh and the attribute very. In this way
there can be several ranks of subordinative position; in very fresh
there are three: (1) milk, (2) fresh, (3) very. In the same way,
the head also may show an attributive construction: the phrase
this fresh milk consists of the attribute this and the head fresh milk,
and this, in turn, of the attribute fresh and the head milk.

12. 11. If all the syntactic constructions which go to make up a
phrase are endocentric, then the phrase will contain among its
ultimate constituents some word (or several words, members of a
co-ordination) whose form-class is the same as that of the phrase.
This word is the center of the phrase. In the phrase all this fresh
, the word milk is the center, and in the phrase all this fresh
bread and sweet butter
, the words bread and butter are the centers.
Since most of the constructions in any language are endocentric,
most phrases have a center: the form-class of a phrase is usually
the same as that of some word that is contained in the phrase.
195The exceptions are phrases of exocentric construction, and these,
too, we have seen, are definable in terms of word-classes. The
syntactic form-classes of phrases, therefore, can be derived from
the syntactic form-classes of words: the form-classes of syntax are
most easily described in terms of word-classes. Thus, in English, a
substantive expression is either a word (such as John) which
belongs to this form-class (a substantive), or else a phrase (such as
poor John) whose center is a substantive; and an English finite
verb expression is either a word (such as ran) which belongs to this
form-class (a finite verb), or else a phrase (such as ran away) whose
center is a finite verb. An English actor-action phrase (such as
John ran or poor John ran away) does not share the form-class of
any word, since its construction is exocentric, but the form-class of
actor-action phrases is defined by their construction: they consist
of a nominative expression and a finite verb expression (arranged
in a certain way), and this, in the end, again reduces the matter to
terms of word-classes.

The term parts of speech is traditionally applied to the most
inclusive and fundamental word-classes of a language, and then,
in accordance with the principle just stated, the syntactic form-classes
are described in terms of the parts of speech that appear in
them. However, it is impossible to set up a fully consistent scheme
of parts of speech, because the word-classes overlap and cross each

In speaking of form-classes we use the term expression to include
both words and phrases: thus John is a substantive, poor John a
substantive phrase, and both forms are substantive expressions.

Within the great form-classes which contain both words and
(thanks to endocentric constructions) a vast number of phrasal
combinations, there may be sub-classes due to small differences of
phrasal construction. For instance, when an attribute like fresh,
good, or sweet is joined to the head milk, as in fresh milk, this
resultant phrase is still capable of joining with other attributes, as
in good, sweet, fresh milk: the phrase has entirely the same functions
as its center (and head), namely the word milk. If, however, we
join a form like milk or fresh milk with the attribute this, the
resultant phrase, this milk or this fresh milk has not quite the same
function as the head or center, since the resultant phrase cannot be
joined with attributes like good, sweet: the construction in this
, this fresh milk is partially closed. The possibilities in this
196direction, in fact, are limited to adding the attribute all, as in all
this milk
or all this fresh milk. When the attribute all has been
added, the construction is closed: no more attributes of this type
(adjectives) can be added.

12. 12. An example of a taxeme of order is the arrangment by
which the actor form precedes the action form in the normal type
of the English actor-action construction: John ran. In languages
which use highly complex taxemes of selection, order is largely
non-distinctive and connotative; in a Latin phrase such as pater
amat fīlium
‘the father loves the son,’ the syntactic relations are all
selective (cross-reference and government) and the words appear
in all possible orders (pater fīlium amat, fīlium pater amat, and so
on), with differences only of emphasis and liveliness. In English,
taxemes of order appear in the difference between actor-action and
action-goal, as in John ran and catch John; the difference between
John hit Bill and Bill hit John rests entirely upon order. In general,
however, taxemes of order in English occur along with taxemes of
selection. Languages which in this respect and in the general
configuration of their syntax resemble English, may still show great
differences as to taxemes of order. Thus, standard German differs
from English in allowing only one attribute (word or phrase) of the
verb to precede a finite verb: heute spielen wir Ball [ˈhojte ˈšpiːlen
viːr ˈbal] ‘today play we ball.’ Further, it places several elements
last in the sentence: certain adverbs, as ich stehe um sieben Uhr
[ix ˈšte:e um ˈziːben ˈuːr ˈawf] ‘I get at seven o'clock up;’
participles, as ich habe ihn heute gesehen [ix ˌha:be iːn ˈhojte
geˈze:n] ‘I have him today seen;’ infinitives, as ich werde ihn heute
[ix ˌverde iːn ˈhojte ˈze:n] ‘I shall him today see;’ the verb
of a dependent clause: wenn ich ihn heute sehe [ven ix iːn ˈhojte
ˈze:e] ‘if I him today see.’

French has a complicated and rigid system of ordering certain
substitute (“conjunct”) accompaniments of its verbs. In the
ordinary (non-interrogative) sentence-type, it distinguishes seven
positions of these elements, which precede the finite verb:

(1) actors, such as je [žə] ‘I,’ il [il] ‘he, it,’ ils [il] ‘they,’ on
[on] ‘one,’ ce [sə] ‘it, that’

(2) the negative adverb ne [nə] ‘not’

(3) farther goals of first and second persons, such as me [mə]
‘to me,’ vous [vu] ‘to you,’ and of the reflexive se [sə] ‘to himself,
herself, themselves’197

(4) nearer goals, such as me [mə] ‘me,’ vous [vu] ‘you,’ se [sə]
‘himself, herself, themselves,’ le [lə] ‘him, it,’ les [le] ‘them’

(5) farther goals of the third person: lui [lɥi] ‘to him, to her,’
leur [lœ:r] ‘to them’

(6) the adverb y [i] ‘there, thither, to it, to them’

(7) the adverb enn] ‘from there, of it, of them.’

For example: (1-2-3-4) il ne me le donne pas [i n mə l dɔn pɑ]
‘he does not give it to me’

(1-3-6-7) il m'y en donne [i m j ɑn dɔn] ‘he gives me some of it

(1-4-5) on le lui donne [on lə lɥi dɔn] ‘one gives it to him’

(1-2-6-7) il n'y en a pas [i n j ɑnn a pɑ] ‘there aren't any,’
literally ‘it has not of them there.’

Occasionally order serves finer distinctions. In French most
adjectives follow their nouns: une maison blanche [yn mezon blɑnš]
‘a white house;’ a certain few precede: une belle maison [yn bel
mezon] ‘a pretty house;’ others precede only with transferred
meanings or with emphatic or intense connotations: une barbe
[yn barbə nwa:r] ‘a black beard’: une noire trahison [yn
nwa:r traizon] ‘a black betrayal;’ un livre excellent n liːvr
eksɛlɑn] ‘an excellent book’: un excellent livre ‘a splendid book!’
A few show greater differences of meaning: un livre cher [œn liːvrə
šɛ:r] ‘a costly book’: un cher ami [œn šɛ:r ami] ‘a dear friend,’
sa propre main [sa proprəmɛn] ‘his own hand’: une main propre
[yn mɛn propr] ‘a clean hand.’

Viewed from the standpoint of economy, taxemes of order are
a gain, since the forms are bound to be spoken in some succession;
nevertheless, few languages allow features of order to work alone:
almost always they merely supplement taxemes of selection.

12. 13. The languages of the Indo-European family are peculiar
in having many parts of speech; no matter upon what constructions
we base our scheme, a language like English will show at
least half a dozen parts of speech, such as substantive, verb, adjective,
adverb, preposition, co-ordinating conjunction, and subordinating
conjunction, in addition to interjections. Most languages
show a smaller number. A distribution into three types is quite
frequent (Semitic, Algonquian); usually one resembles our substantives
and one our verbs. It is a mistake to suppose that our part-of-speech
system represents universal features of human expression.
If such classes as objects, actions, and qualities exist apart from
198our language, as realities either of physics or of human psychology,
then, of course, they exist all over the world, but it would still
be true that many languages lack corresponding parts of speech.

In languages with few parts of speech, the syntactic form-classes
appear rather in phrases. Often the class of a phrase is indicated
by some special word, a marker; strictly speaking, the marker and
the form which it accompanies are joined in an exocentric construction
which determines the class of the phrase. Aside from
this selective feature, the constructions are likely to be distinguished
by word-order.

The classical instance is Chinese. The parts of speech are full
and particles (that is, markers). The principal constructions
are three.

(1) The favorite sentence-construction is one of subject and
predicate, much like the English actor-action construction; the
subject precedes the predicate: [tha1 ˈxaw3] ‘he is good,’ [tha1
ˈlaj2] ‘he came.’ In certain cases, depending on differences of form-class,
the predicate is marked by the particle [šə4] at its beginning:
[tha1 šə4 ˈxaw3 ˌžən2] ‘he (p.) good man,’ that is, ‘he is a good

(2) There is an endocentric construction in which the attribute
precedes the head; in meaning this resembles the similar English
constructions: [ˈxaw3 ˌžən2] ‘good man,’ [ˈman4 ˌčhy4] ‘slowly go,’
that is, ‘go slowly.’ The attribute is in certain cases marked by
the particle [ti1] at its end: [ˈtiŋ3 ˌxaw3 ti2 ˈžən2] ‘very good man;’
[ˌwo3 ti2 ˈfu4 čhin1] ‘I (p.) father,’ that is, ‘my father;’ [ˈtso4 čo2
ti1 fan2] ‘sit (p.) person,’ that is, ‘a sitting person;’ [ˈwo3 ˈçje3
ˌtsə4 ti ˈpi3] ‘I write (p.) brush,’ that is, ‘the brush I write with’
— in this example the attribute is a phrase of subject-predicate
construction; [ˈmaj3 ti ˈšu1] ‘buy (p.) book,’ that is, ‘the purchased

(3) A second endocentric construction, in which the attribute
follows the head, resembles rather the English action-goal and
relation-axis constructions: [ˌkwan1 ˈman2] ‘shut the door,’ [ˌtsaj4
ˈčuŋ1 kwo] ‘in China.’ We may call this, somewhat inexactly,
the action-goal construction, to distinguish it from (2).

Taxemes of selection consist largely in the marking off of a
form-class which serves as subject in (1), as head in (2), and as
goal in (3), resembling the English substantive expression. To
this form-class (we may call it the object expression) only a few
199words may be said to belong in their own right; these are substitute-words
of the type [tha1] ‘he, she’ or [wo3] ‘I.’ The other object
expressions are phrases with various markers. The commonest
of these markers are certain particles which precede as attributes
of type (2), such as [čə4] ‘this,’ [na4] ‘that,’ [na3] ‘which?’ Thus,
[ˈčə4 ko4] ‘this piece,’ that is, ‘this (thing).’ In most instances
these markers do not immediately join with a full word; but only
with certain ones, like the [ko4] ‘piece’ in the last example, which
hereby constitute a form-class of numeratives; the phrase of marker
plus numerative joins the ordinary full word in construction
(2), as: [čə ko ˈžən2] ‘this (individual) man;’ [ˌwu3 ˌljaŋ4 ˈčhə1]
‘five (individual) cart,’ that is, ‘five carts.’ Another kind of object
expression is characterized by the particle [ti1] at its end: [ˌmaj4
ˈšu1 ti] ‘sell book (p.),’ that is ‘bookseller.’

In this way complex phrases are built up: [tha1 ˈtaw4 ˈthjen2
li3 ˈčhy4] ‘he enter field interior go,’ that is, ‘he goes into the field;’
here the first word is the subject, the rest of the phrase the predicate;
in this predicate the last word is the head and the other three
are an attribute; this attribute consists of the action [taw4] ‘enter’
and the goal [ˈthjen2 li3] ‘field interior,’ in which the first word is an
attribute of the second. In the sentence [ni3 ˈmej2 pa3 ˈmaj3
ˈmej2 ti ˌčhjen3 ˈkej3 wo3] ‘you not take buy coal (p.) money give I,’
the first word is the subject, the rest the predicate; this predicate
consists of an attribute, [mej2] ‘not’ and a head; within this head,
the first five words are again an attribute and the last two [ˈkej3
wo3] ‘give I’ a head, whose construction is action and goal. In the
five-word attribute [pa3 ˌmaj3 ˈmej2 ti ˌčhjen3] ‘take buy coal
(p.) money,’ the first word is an action and the rest a goal; this
goal consists of the head [chjen3] ‘money’ and the attribute [ˌmaj3
ˈmej2 ti] which is marked as such by the particle [ti1] appended
to the phrase [ˌmaj3 ˈmej2] ‘buy coal,’ whose construction is action-goal.
Thus the sentence means ‘you not taking buy-coal-money
give me,’ that is ‘you haven't given me money to buy coal.’

In Tagalog, the parts of speech are, again, full word and particle,
but here the full words are subdivided into two classes which we
may call static and transient. The latter resemble our verbs in
forming a special kind of predicate (the narrative type, with four
sub-types, § 11.2) and in showing morphologic distinctions of
tense and mode, but they differ from our verbs because, on the one
hand, they are not restricted to the function of predicate and, on
200the other hand, there exist non-narrative predicates. The chief
constructions are subject and predicate, marked optionally by
order (predicate precedes subject) or by the particle [aj] and order
(subject precedes predicate marked by initial [aj]), as illustrated
in § 11.2. The subject and the equational predicate are selectively
marked: the class of forms which fill these positions resembles the
English substantive expression and, even more, the Chinese object
expression. A few substitute-words, such as [aˈku] ‘I’ and [siˈja]
‘he, she,’ belong to this class by their own right; all other object
expressions are phrases, characterized by the presence of certain
attributes, as [isa ŋ ˈba:taʔ] ‘one child,’ or by certain particles,
chiefly [si] before names, as [si ˈhwan] ‘John,’ and [aŋ] before other
forms, as [aŋ ˈba:taʔ] ‘the child, a child,’ [aŋ puˈla] ‘the red,’ that
is, ‘the redness,’ [aŋ ˈpuːtul] ‘the cut,’ or, to illustrate transient
forms, [aŋ puˈmuːtul] ‘the one who cut,’ [aŋ piˈnuːtul] ‘that which
was cut,’ [aŋ ipiˈnuːtul] ‘that which was cut with,’ [aŋ pinuˈtuːlan]
‘that which was cut from.’ There are four attributive constructions.
In one, a particle [na], after vowels [ŋ], intervenes between
head and attribute, in either order, as [aŋ ˈba:ta ŋ sumuːˈsuːlat]
or [aŋ sumuːˈsuːlat na ˈba:taʔ] ‘the writing child;’ [aŋ puˈla ŋ
panˈju] ‘the red handkerchief,’ [aŋ panˈju ŋ iˈtu] ‘this handkerchief.’
Another, more restricted attributive construction lacks the particle,
as [hinˈdiː aˈku] ‘not I,’ [hinˈdiː mabaˈit] ‘not good.’ In the third
attributive construction the attribute is an object expression in a
special form: thus, [aˈku] ‘I’ is replaced by [ku], and [siˈja] ‘he,
she’ by [niˈja], and the particle [si] by [ni], the particle [aŋ] by
[nan]: [aŋ puˈla naŋ panˈju aj matiŋˈkad] ‘the red of the handkerchief
is bright;’ [aŋ ˈba:ta j kuˈma:in naŋ ˈka:nin] ‘the child ate
(some) rice,’ (actor-action); [kiˈna:in nan ˈba:taʔ aŋ ˈka:nin]
‘the rice was eaten by the child’ (goal-action); see also the examples
in § 11.2. In the fourth attributive construction, too, the attribute
is an object expression: [si] is replaced by [kaj] and [aŋ] by
[sa]; the attribute tells of a place: [aŋ ˈba:ta j naˈna:ug sa ˈba:haj]
‘the child came out of the house, out of a house.’

12. 14. The details of syntax are often complicated and hard to
describe. On this point, any fairly complete grammar of a language
like English, German, Latin, or French, will prove more enlightening
than would an abstract discussion. Syntax is obscured, however,
in most treatises, by the use of philosophical instead of
formal definitions of constructions and form-classes. As a single
201illustration of the more complex syntactic habits, we shall survey
the main features of one construction in present-day (colloquial
standard) English — the construction which we may call character-substance,
as in fresh milk.

This construction is attributive, and the head is always a noun-expression
— that is, a noun or an endocentric phrase with a noun
as center. The noun is a word-class; like all form-classes, it is to be
defined in terms of grammatical features, some of which, in fact,
appear in what follows. When it has been defined, it shows a class-meaning
which can be roughly stated as ‘object of such and such
a species;’ examples are boy, stone, water, kindness. The attribute
in our construction is always an adjective expression — that is, an
adjective or an endocentric phrase with an adjective as center. The
adjective is in English a word-class (part of speech), definable
precisely by its function in the character-substance construction
which we are now to discuss; its class-meaning will emerge from
our discussion as something like ‘character of specimens of a species
of objects;’ examples are big, red, this, some. Beside these features
of selection, the character-substance construction contains a feature
of order: the adjective expression precedes the noun expression
: poor John, fresh milk.

The adjectives are divided into two classes, descriptive and
limiting, by the circumstance that when adjectives of both these
classes occur in a phrase, the limiting adjective precedes and
modifies the group of descriptive adjective plus noun. Thus, in a
form like this fresh milk, the immediate constituents are the limiting
adjective this, and the noun phrase fresh milk, which consists, in
turn, of the descriptive adjective fresh and the noun milk. This
difference subdivides our character-substance construction into
two sub-types, the quality-substance construction, where the attribute
is a descriptive adjective expression, and the limitation-substance
construction, where the attribute is a limiting adjective.

The quality-substance construction and the form-class of descriptive
adjectives are both divided into several types by features
of order. For instance, we say big black sheep and never *black big
, kind old man and never *old kind man, and so on. We shall
not stop to examine these sub-types. The meaning of the form-class
of descriptive adjectives is roughly ‘qualitative character of

The form-class of limiting adjectives is much smaller than that
202of descriptive adjectives, and constitutes, in fact, what we shall
later define as an irregular form-class — that is, a form-class which
has to be described in the shape of a list of the forms; however, the
boundary between limiting and descriptive adjectives is not completely
definable. The class-meaning of limiting adjectives will
appear from the following discussion as something like ‘variable
character of specimens.’

Our limiting adjectives fall into two sub-classes of determiners
and numeratives. These two classes have several subdivisions and
are crossed, moreover, by several other lines of classification.

The determiners are defined by the fact that certain types of
noun expressions (such as house or big house) are always accompanied
by a determiner (as, this house, a big house). The class-meaning
is, roughly, ‘identificational character of specimens.’ This
habit of using certain noun expressions always with a determiner,
is peculiar to some languages, such as the modern Germanic and
Romance. Many languages have not this habit; in Latin, for
instance, domus ‘house’ requires no attribute and is used indifferently
where we say the house or a house.

A number of features subdivides the determiners into two classes,
definite and indefinite. Of these features, we shall mention only
one: a definite determiner can be preceded by the numerative all
(as in all the water) but an indefinite determiner (as, some in some
) cannot.

The definite determiners are: any possessive adjective (John's
, my house) and the words this (these), that (those), the. The
class of possessive adjectives is definable in terms of morphology.
It is worth observing that Italian, which has a character-substance
construction much like ours, does not use possessive adjectives as
determiners: il mio amico [il mio aˈmiko] ‘the my friend’ (that is,
‘my friend’) contrasts with un [un] mio amico ‘a my friend’ (that
is, ‘a friend of mine’). The class-meaning of definite determiners
is ‘identified specimens.’ A precise statement of how the specimens
are identified, is a practical matter outside the linguist's control;
the identification consists in possession by some person (John's
), spatial relation to the speaker (this house), description by
some accompanying linguistic form (the house I saw), or purely
situational features (the sky, the chairman), among which earlier
mention by speech is to be reckoned (“I saw a man, but the man
did not see me”). Among the definite determiners, this : these
203and that : those are peculiar in showing congruence with the number-class
of the noun (this house : these houses).

The indefinite determiners are a (an), any, each, either, every,
neither, no, one, some, what, whatever, which, whichever, and the
phrasal combinations many a, such a, what a. The class-meaning
is ‘unidentified specimens.’

The word a is peculiar in its sandhi-form an, used before vowels.
The word one occurs not only as an indefinite determiner (one
), but also in some entirely different functions (as in a big one,
if one only knew); this phenomenon may be designated as class-cleavage.
The meanings of the various indefinite determiners are
in part linguistically definable in terms of grammatical features
of wider bearing than our present subject. For instance, what and
which are interrogative, introducing supplement-questions, which
prompt the hearer to supply a speech-form (what man? which man?)
Whatever and whichever are relative, marking their noun as part
of a subordinate clause (whatever book you take, …). No and
neither are negative, ruling out all specimens. Each, which, and
whichever imply a limited field of selection: that is, the specimens
concerned belong to an identified part (or to the identified whole)
of the species (which book? which parent?); either and neither go
farther in limiting the field to two specimens.

Some of the determiners are atonic (barring, of course, the
case where they are emphatic elements): my, our, your, his, her,
its, their, the, a; others are sometimes atonic or spoken with secondary

The types of noun expressions which always have a determiner,
are preceded, when no more specific determiner is present, by
the articles, definite the and indefinite a, whose meaning is merely
the class-meaning of their respective form-classes. A grammatical
classification, such as definite and indefinite, which always accompanies
some grammatical feature (here the types of noun expression,
which demand a determiner), is said to be categoric. The
definite and indefinite categories may be said, in fact, to embrace
the entire class of English noun expressions, because even those
types of noun expression which do not always take a determiner,
can be classed as definite or indefinite: John, for instance, as definite,
kindness as indefinite.

According to the use and non-use of determiners, English noun
expressions fall into a number of interesting sub-classes:204

I. Names (proper nouns) occur only in the singular number,
take no determiner, and are always definite: John, Chicago.
The class meaning is ‘species of object containing only one
.’ Here and in what follows, space forbids our entering
into details, such as the class-cleavage by which a
name occurs also as a common noun, in cases like homonymy
(two Johns, this John); nor can we take up sub-classes, such
as that of river-names, Which are always preceded by the
(the Mississippi).

II. Common nouns occur in both categories, definite and indefinite.
The class-meaning is ‘species of object occurring in
more than one specimen
.’ In the plural number they require
a determiner for the definite category (the houses), but not
for the indefinite (houses, corresponding to the singular form
a house).

A. Bounded nouns in the singular number require a determiner
(the house, a house). The class meaning is ‘species of object
occurring in more than one specimen, such that the specimens
cannot be subdivided or merged

B. Unbounded nouns require a determiner for the definite category
only (the milk : milk). The class-meaning is ‘species
of object occurring in more than one specimen, such that
the specimens can be subdivided or merged

1. Mass nouns never take a and have no plural (the milk : milk).
The class-meaning is that of B with the added proviso that
the specimens ‘exist independently.’

2.Abstract nouns in the indefinite singular without a determiner
include all the specimens (life is short); with a determiner and
in the plural, the specimens are separate (a useful life; nine
). The class-meaning is that of B with the proviso
that the specimens ‘exist only as the demeanor (quality, action,
relation) of other objects.’

Among the subdivisions of II, class-cleavage is frequent and interesting,
as, an egg, eggs (A), but “he got egg on his necktie”
(B1); coffee (B1), but an expensive coffee (A).

The limiting adjectives of the other class, numeratives, fall
into various sub-classes, of which we shall merely mention a few.
Two of them, all and both precede a determiner (all the apples); the
rest follow (the other apples). Two, however, precede a in phrases
which are determiners: many a, such a. The numeratives few,
205hundred, thousand, and those formed with the suffix -ion (million
and so on), are preceded by a in phrases which serve as numeratives
with plural nouns (a hundred years). The numeratives same,
very, one — this last differs by class-cleavage from the determiner
one — are used only with definite nouns (this same book, the very
, my one hope); the numeratives much, more, less are used only
with indefinite nouns (much water); the numerative all is used with
both kinds of nouns but only with definite determiners (all the
milk; all milk
). Some, such as both, few, many, and the higher numbers,
are used only with plural nouns; others, such as one, much,
little, only with singular nouns. Some numeratives are used also
in other syntactic positions, as, many and few as predicate adjectives
(they were many), and all, both as semi-predicative attributes
(the boys were both there). Some other interesting lines of classification
among the English numeratives will appear when we take
up the substitutive replacement of noun expressions in Chapter 15.206

Chapter 13

13. 1. By the morphology of a language we mean the constructions
in which bound forms appear among the constituents. By
definition, the resultant forms are either bound forms or words,
but never phrases. Accordingly, we may say that morphology
includes the constructions of words and parts of words, while
syntax includes the constructions of phrases. As a border region
we have phrase-words (jack-in-the-pulpit) and some compound
words (blackbird), which contain no bound forms among their
immediate constituents, and yet in some ways exhibit morphologic
rather than syntactic types of construction.

In general, morphologic constructions are more elaborate than
those of syntax. The features of modification and modulation
are more numerous and often irregular — that is, confined to
particular constituents or combinations. The order of the constituents
is almost always rigidly fixed, permitting of no such
connotative variants as John ran away : Away ran John. Features
of selection minutely and often whimsically limit the constituents
that may be united into a complex form.

Accordingly, languages differ more in morphology than in syntax.
The variety is so great that no simple scheme will classify
languages as to their morphology. One such scheme distinguishes
analytic languages, which use few bound forms, from synthetic,
which use many. At one extreme is a completely analytic language,
like modern Chinese, where each word is a one-syllable
morpheme or a compound word or phrase-word; at the other, a
highly synthetic language like Eskimo, which unites long strings
of bound forms into single words, such as [a:wlisa-ut-issʔar-si-niarpu-ŋa]
‘I am looking for something suitable for a fish-line.’
This distinction, however, except for cases at the former extreme,
is relative; any one language may be in some respects more analytic,
but in other respects more synthetic, than some other language.
Another scheme of this sort divided languages into four
morphologic types, isolating, agglutinative, polysynthetic, and inflecting.
207Isolating languages were those which, like Chinese, used
no bound forms; in agglutinative languages the bound forms were
supposed merely to follow one another, Turkish being the stock
example; polysynthetic languages expressed semantically important
elements, such as verbal goals, by means of bound forms, as
does Eskimo; inflectional languages showed a merging of semantically
distinct features either in a single bound form or in closely
united bound forms, as when the suffix in a Latin form like
amō ‘I love’ expresses the meanings ‘speaker as actor,’ ‘only one
actor,’ ‘action in present time,’ ‘real (not merely possible or hypothetical)
action.’ These distinctions are not co-ordinate, and
the last three classes were never clearly defined.

13. 2. Since the speaker cannot isolate bound forms by speaking
them alone, he is usually unable to describe the structure of
words. The statement of morphology requires systematic study.
The ancient Greeks made some progress in this direction, but,
in the main, our technique was developed by the Hindu grammarians.
No matter how refined our method, the elusive nature of
meanings will always cause difficulty, especially when doubtful
relations of meaning are accompanied by formal irregularities.
In the series goose, gosling, gooseberry, gander, we shall probably
agree that the first two forms are morphologically related, in the
sense that [gɑz-] in gosling is a phonetic modification of goose,
but the [guwz-] in gooseberry does not fit the meaning, and, on the
other hand, the formal resemblance [g-] of goose and gander is
so slight that one may question whether it really puts the practical
relation of meaning into linguistic form. This last difficulty appears
also in the pair duck : drake, with their common [d… k]. One
soon learns that one cannot look to the speakers for an answer,
since they do not practise morphologic analysis; if one bothers
them with such questions, they give inconsistent or silly answers.
If the history of a language is known, one often finds that the
ambiguity was absent in some older state of the language — it
appears, for instance, that some centuries ago ‘gooseberry’ was
*grose-berry and had nothing to do with a goose — but facts of this
sort evidently do not tell us how things work in the present state
of the language.

In describing the modulations and modifications which occur in
syntax, we naturally take the absolute form of a word or phrase
as our starting-point, but a bound form which occurs in several
208shapes will lead to several entirely different forms of description,
according to our choice of a basic alternant. For instance, the
plural-suffix of English nouns appears ordinarily in three shapes:
[-ez] glasses, [-z] cards, [-s] books; by taking each of these three, in
turn, as one's starting-point, one can arrive at three entirely
different statements of the facts.

Very often there are further difficulties. Sometimes a grammatical
feature, such as a phonetic modification, appears to express a
meaning which is usually expressed by a linguistic form, as in
man : men, where modification of the vowel takes the place of the
plural-suffix. In other cases there is not even a grammatical
feature: a single phonetic form, in the manner of homonymy, represents
two meanings which are usually distinguished by means
of a linguistic form, as, singular and plural noun in the sheep
(grazes) : the sheep (graze). Here the Hindus hit upon the apparently
artificial but in practice eminently serviceable device of speaking
of a zero element: in sheep : sheep the plural-suffix is replaced by
zero — that is, by nothing at all.

13. 3. What with these and other difficulties, any inconsistency
of procedure is likely to create confusion in a descriptive statement
of morphology. One must observe, above all, the principle of
immediate constituents (§ 10.2). This principle leads us, at the
outset, to distinguish certain classes of words, according to the
immediate constituents:

A. Secondary words, containing free forms:

1. Compound words, containing more than one free form:
door-knob, wild-animal-tamer. The included free forms are
the members of the compound word: in our examples, the
members are the words door, knob, tamer, and the phrase
wild animal.

2. Derived secondary words, containing one free form: boyish,
old-maidish. The included free form is called the underlying
; in our examples the underlying forms are the
word boy and the phrase old maid.

B. Primary words, not containing a free form:

1. Derived primary words, containing more than one bound
form: re-ceive, de-ceive, con-ceive, re-tain, de-tain, contain.

2. Morpheme-words, consisting of a single (free) morpheme:
man, boy, cut, run, red, big.209

The principle of immediate constituents will lead us, for example,
to class a form like gentlemanly not as a compound word, but as a
derived secondary word, since the immediate constituents are the
bound form -ly and the underlying word gentleman; the word
gentlemanly is a secondary derivative (a so-called de-compound)
whose underlying form happens to be a compound word. Similarly,
door-knobs is not a compound word, but a de-compound, consisting
of the bound form [-z] and the underlying word door-knob.

The principle of immediate constituents leads us to observe the
structural order of the constituents, which may differ from their
actual sequence; thus, ungentlemanly consists of un- and gentlemanly,
with the bound form added at the beginning, but gentlemanly
consists of gentleman and -ly with the bound form added at
the end.

13. 4. As examples of relatively simple morphologic arrangements,
we may take the constructions of secondary derivation that
appear in English plural nouns (glass-es) and past-tense verbs

As to selection, the bound forms are in both cases unique, but
the underlying forms belong to two great form-classes: the plural
nouns are derived from singular nouns (as, glasses from glass) and
the past-tense verbs from infinitive verbs (as, landed from land).
Other, subsidiary taxemes of selection will concern us later.

As to order, the bound form, in both cases, is spoken after the
underlying form.

By a feature of modulation common to nearly all constructions of
English morphology, the underlying form keeps its stress, and the
bound form is unstressed.

The taxemes of phonetic modification are more elaborate, and
will show us some peculiarities that appear in the morphology of
many languages.

To begin with, the bound form appears in several alternants,
different shapes which imply, in this case, features of phonetic

glass : glasses [-ez]
pen : pens [-z]
book : books [-s].

If we collect examples, we soon find that the shape of the bound
form is determined by the last phoneme of the accompanying form:210

[-ez] appears after sibilants and affricates (glasses, roses, dishes,
garages, churches, bridges); [-z] appears after all other voiced
phonemes (saws, boys, ribs, sleeves, pens, hills, cars); and [-s] after
all other unvoiced phonemes (books, cliffs). Since the differences
between the three alternants [-ez, -z, -s] can be described in terms
of phonetic modification, we say that they are phonetic alternants.
Since the distribution of the three alternants is regulated according
to a linguistically recognizable characteristic of the accompanying
forms, we say that the alternation is regular. Finally, since the
deciding characteristic of the accompanying forms is phonemic
(namely, the identity of the last phoneme), we say that the alternation
is automatic.

Regular alternations play a great part in the morphology of most
languages. Not all regular alternations are phonetic or automatic.
In German, for instance, the singular nouns are divided, by certain
syntactic features, into three form-classes which are known as
genders (§ 12.7); now, German plural nouns are derived from
singulars by the addition of bound forms which differ according to
the gender of the underlying singular:

masculine nouns add [-e], with certain vowel-changes: der Hut
[huːt] ‘hat’: Hilte [ˈhy:te] ‘hats;’ der Sohn [zo:n] ‘son’: Sohne
[ˈzø:ne] ‘sons;’ der Baum [bawm] ‘tree’: Bäume [ˈbojme] ‘trees’

neuter nouns add [-e] without vowel-change: das Jahr [ja:r]
‘year’: Jahre [ˈja:re] ‘years;’ das Boot [bo:t] ‘boat’: Boote [ˈbo:te]
‘boats;’ das Tier [tiːr] ‘animal’: Tiere [ˈtiːre] ‘animals’

feminine nouns add [-en]: die Uhr [uːr] ‘clock, watch’: Uhren
[ˈuːren] ‘clocks, watches;’ die Last [last] ‘burden’: Lasten [ˈlasten]
‘burdens;’ die Frau [fraw] ‘woman’: Frauen [ˈfrawen] ‘women.’

This alternation (aside from special features which we need not
consider) is regular, but it is not phonetic, since, of the three
alternants, [-e] with vowel change, [-e], and [-en], the last is not,
in the system of the language, phonetically akin to the first two;
and the alternation is not automatic, but grammatical, since it
depends not upon phonetic, but upon grammatical (in this instance,
syntactic) peculiarities of the underlying forms.

13. 5. We have not yet described in terms of phonetic modification,
the kinship of the three alternants [-ez, -z, -s] of the bound
form that appears in English plural nouns. It is evident that three
entirely different statements are possible, according to our choice
of one or another of the three forms as our starting-point. Our
211aim is to get, in the long run, the simplest possible set of statements
that will describe the facts of the English language. To try out the
different possible formulae with this aim in view, often involves
great labor. In the present instance our trouble is small, because
our alternation has an exact parallel in English syntax: the enclitic
word whose absolute form is is [ˈiz], alternates quite like our plural

Bess's ready [iz, ez] 119
John's ready [z]
Dick's ready [s].

Since in this case the absolute form is necessarily serves as the
starting-point of description, we reach the simplest formula if we
take [-ez] as the basic alternant also of the bound form. We can
say, then, that in English any morpheme of the form [iz, ez], unstressed,
loses its vowel after all phonemes except sibilants and
affricates, and then replaces [z] by [s] after unvoiced sounds. This
covers also the alternation of the third-person present-tense verb
suffix in misses : runs : breaks and of the possessive-adjective
suffix in Bess's, John's, Dick's. Moreover, it leads us to use a
parallel formula in the case of the past-tense suffix of verbs. This
suffix appears in three similar alternants:

land : landed [-ed]
live : lived [-d]
dance : danced [-t],

and we need not hesitate, now, to take [-ed] as the basic form for
our description and to say that this form loses its vowel after all
phonemes except dental stops, and then replaces [d] by [t] after all
unvoiced sounds.

13. 6. A survey of English plural nouns will soon show that the
statement we have made holds good for an indefinitely large
number of forms, but not for a certain limited number of exceptions.

In some instances the constituent form in the plural differs
phonetically from the underlying singular noun:

knife [najf] : knives [najv-z]
mouth [mawθ] : mouths [mawðz]
house [haws] : houses [ˈhawz-ez].212

We can describe the peculiarity of these plurals by saying that
the final [f, θ, s] of the underlying singular is replaced by [v, ð, z]
before the bound form is added. The word “before” in this
statement means that the alternant of the bound form is the one
appropriate to the substituted sound; thus, the plural of knife
adds not [-s], but [-z]: “first” the [-f] is replaced by [-v], and “then”
the appropriate alternant [-z] is added. The terms “before, after,
first, then,” and so on, in such statements, tell the descriptive
. The actual sequence of constituents, and their structural
order (§ 13.3) are a part of the language, but the descriptive order
of grammatical features is a fiction and results simply from our
method of describing the forms; it goes without saying, for instance,
that the speaker who says knives, does not “first” replace [f] by
[v] and “then” add [-z], but merely utters a form (knives) which
in certain features resembles and in certain features differs from
a certain other form (namely, knife).

If the English plural nouns which exhibit this voicing of a final
spirant in the underlying form, showed any common phonetic
or grammatical feature that distinguished them from other nouns,
we could describe this peculiarity as a regular alternant. This,
however, seems not to be the case; we have also plurals like cliffs,
myths, creases, where [f, θ, s] of the underlying form appears unchanged.
We can make our general statement cover one group,
but will then have to furnish a list of the cases that do not fall
under the general statement. A set of forms that is not covered
by a general statement, but has to be presented in the shape of
a list, is said to be irregular. We try, of course, to arrange our
description so that as many forms as possible will be included
in general statements. The choice is often decided for us by the
circumstance that one group of forms is of indefinite extent and
therefore amenable to a general statement, but not to a list. In
the case of English nouns in [-s], we obviously face this condition,
for house : houses is the only instance where [-s] is replaced by [z]
in the plural, while an indefinite number of plural nouns retains
the [-s] of the underlying form (glasses, creases, curses, dances,
and so on). Our list, in this case, includes only one form, houses,
a unique irregularity. The list of plurals which substitute [-ð]
for the [-θ] of the underlying form is not large, embracing only the
forms baths, paths, cloths, mouths (and for some speakers also
laths, oaths, truths, youths); on the other side we find a number
213of current forms, such as months, widths, drouths, myths, hearths,
and, what is more decisive, the habit of keeping [-θ] in the formation
of plurals that are not traditional and may be formed by a
speaker who has not heard them: the McGraths, napropaths, monoliths.
In the case of [-f] the list is larger: knives, wives, lives, calves,
halves, thieves, leaves, sheaves, beeves, loaves, elves, shelves (and for
some speakers also hooves, rooves, scarves, dwarves, wharves); we
decide to call these irregular on the strength not only of counter-instances,
such as cliffs, toughs, reefs, oafs, but also of less common
or occasional forms, such as (some good) laughs, (general) staffs,

Where the two treatments occur side by side, as in laths [Iɛθs]
or [lɛðz], roofs or rooves, there is usually some slight difference of
connotation between the variants. The noun beef, as a mass-noun
(§ 12.14), has no ordinary plural by its side; the plural beeves is
a specialized derivative, since it deviates in its meaning of ‘oxen,
cattle,’ with archaic-poetic connotation.

We may note in passing that the grammatical features we have
discussed, determine features of the phonetic pattern (§ 8.5),
by defining groups like sibilant-affricate, dental stop, voiced, unvoiced,
and establishing the relation [f, θ, s] versus [v, ð, z], and
[t] versus [d].

We may describe “voicing of final spirant plus suffix [-ez, -z,
(-s)]” as an irregular alternant of the regular plural-suffix [-ez, -z, -s];
the irregularity consists in a phonetic modification of the underlying
form. The same modification is accompanied by modification
of the syllabic in the uniquely irregular staff: staves. In
cloth [klɔθ] : clothes [klowz] we have a uniquely irregular plural
with specialized meaning (‘garments, clothing’), beside the irregular
plural cloths [klɔðz] with normal meaning.

The homonymous third-person present-tense suffix of verbs is
accompanied by phonetic modification of the underlying form in
do [duw] : does [doz], say [sej] : says [sez], have [hɛv] : has [hɛz].

The past-tense suffix [-ed,-d,-t] is accompanied by phonetic modification
in the irregular forms say : said, flee : fled, hear [hijr] : heard
[hr̩d], keep : kept (and, similarly, crept, slept, swept, wept; leaped
and leapt are variants), do : did, sell : sold (and, similarly, told),
make : made, have : had.

13. 7. In some cases the bound form appears in an unusual
shape. In die : dice the alternant [-s] appears against the general
214habit; in penny : pence the same feature is accompanied by modification
(loss of [-ij]) in the underlying form, together with specialization
of meaning, in contrast with the normal variant pennies.
In the past tense, we find [-t] instead of [-d] in the archaic-flavored
variants burnt, learnt. If we say that in English the unpermitted
final cluster [-dt] is replaced by [-t], we can class here, with [-t]
instead of [-ed], the forms bent, lent, sent, spent, built.

Both constituents show irregular phonetic modification in
feel : felt and similarly in dealt, knelt, dreamt, meant. If we say
that the unpermitted final clusters [-vt, -zt] are replaced by [-ft,-st],
we can class here also leave : left and lose : lost. The bound form appears
in the alternant [-t] instead of [-d], and the underlying form
replaces the syllabic and all that follows by [ɔ] in seek [sijk] : sought
[sɔt] and, similarly, in bought, brought, caught, taught, thought.

In the extreme case, an alternant bears no resemblance to the
other alternants. In ox : oxen the bound form added in the plural
is [-n̩] instead of [-ez, -z, -s]. If the language does not show parallel
cases which warrant our describing the deviant form in terms of
phonetic modification, an alternant of this sort is said to be suppletive;
thus, [-n̩] in oxen is a suppletive alternant of [-ez, -z, -s],
because English grammar shows no phonetic modification of [-ez]
to [-n̩]. In other instances it is the underlying form which suffers
suppletion. Beside the ordinary derivation of kind : kinder,
warm : warmer, and so on, we have good : better, where the underlying
word good is replaced by an entirely different form bet-,
which we describe, accordingly, as a suppletive alternant of good.
In the same way, the infinitive be suffers suppletion, by [i-], in
the third-person present-tense form is [iz]. In child : children, a
suppletive alternant [-rn̩] of the bound form is accompanied by
phonetic modification of the underlying word.

Another extreme case is that of zero-alternants (§ 13.2), in which
a constituent is entirely lacking, as in the plurals sheep, deer, moose,
fish, and so on. These plurals are irregular, for although some of
them (for instance, species of fish, like perch, bass, pickerel, large
enough to be eaten in separate specimens, and not named after
other objects) can be classified by purely practical features of
meaning, they have no formal characteristic by which we could
define them. The past-tense suffix of verbs shows a zero-alternant
in bet, let, set, wet, hit, slit, split, cut, shut, put, beat, cast, cost, burst,
shed, spread, wed. The third-person present-tense suffix has a
215zero-alternant in can, shall, will, must, may, and, in certain constructions
(for instance, with the modifier not), in need, dare; this
is a regular grammatical alternation, since these verbs are definable
by their syntactic function of taking an infinitive modifier
without the preposition to. Our possessive-adjective suffix [-ez, -z,
-s] has a zero-alternant in one instance, namely, after an underlying
form which ends in the plural-suffix [-ez, -z, -s,] as the-boys'.

A zero-alternant may go with modification of the accompanying
form. Thus, the plural nouns geese, teeth, feet, mice, lice, men,
women [ˈwimn̩] add no bound form to the singular, but contain a
different syllabic. In these plurals a grammatical feature, phonetic
modification, expresses a meaning (namely, the sememe ‘more
than one
object’) which is normally expressed by a linguistic form
(namely, the morpheme [-ez, -z, -s]). We may say that “substitution
of [ij]” (for the stressed syllabic of the underlying form) in
geese, teeth, feet, “substitution of [aj]” in mice, lice, “substitution
of [e]” in men, and “substitution of [i]” in women, are alternants
of the normal plural-suffix — substitution-alternants or substitution-forms.
In our past-tense verbs we find substitution of various
syllables taking the place of [-ed, -d, -t], as:

[ɑ] got, shot, trod

[ɛ] drank, sank, shrank, rang, sang, sprang, began, ran, swam,
sat, spat

[e] bled, fed, led, read, met, held, fell

[i] bit, lit, hid, slid

[ɔ] saw, fought

[o] clung, flung, hung, slung, swung, spun, won, dug, stuck,

[u] shook, took

[ej] ate, gave, came, lay

[aw] bound, found, ground, wound

[ow] clove, drove, wove, bore, swore, tore, wore, broke, spoke,
woke, chose, froze, rose, smote, wrote, rode, stole, shone; with dove
as a variant beside regular dived

[(j)uw] knew, blew, flew, slew, drew, grew, threw.

In stand : stood we have a more complex case with an alternant
describable as “substitution of [u] and loss of [n].”

A zero-alternant replaces the bound form, and a suppletive alternant
the underlying form, in cases like be : was, go : went,
I : my, we : our, she : her, bad : worse.216

In cases like have [hɛv] : had [hɛ-d] or make [mejk] : made [rnej-d],
one of the constituents is modified by the loss of a phoneme. This
loss may be described as a minus-feature; like zero-features or
substitution-features, minus-features may occur independently.
For instance, in a French adjective, the regular type has only
one form, regardless of whether the adjective accompanies a
masculine or a feminine noun, e.g. rouge [ruːž] ‘red’ : un livre rouge
n liːvrə ruːž] ‘a red book,’ masculine, and une plume rouge
[yn plym ruːž] ‘a red feather or pen,’ feminine. In a fairly large
irregular type, however, the masculine and feminine forms differ:
un livre vert [vɛ:r] ‘a green book,’ but une plume verte [vɛrt] ‘a
green feather or pen.’ Thus:

tableau masculine | plat | laid | distinct | long | bas | gris | frais | gentil | léger | soul | plein | feminine | laide | distincte | longue | basse | grise | fraîche | gentille | légère | soule | pleine.

It is evident that two forms of description are here possible.
We could take the masculine forms as a basis and tell what consonant
is added in each case in the feminine form, and this would,
of course, result in a fairly complicated statement. On the other
hand, if we take the feminine form as our basis, we can describe
this irregular type by the simple statement that the masculine
form is derived from the feminine by means of a minus-feature,
namely, loss of the final consonant and of the cluster [-kt]. If we
take the latter course, we find, moreover, that all the other differences
between the two forms, as to vowel quantity and as to
nasalization (as in our last example), re-appear in other phases of
French morphology and can in large part be attributed to the
phonetic pattern.

The last part of our discussion has shown us that a word may
have the character of a secondary derivative and yet consist of
217only one morpheme, accompanied by a zero-feature (sheep, as
a plural; cut as a past), by a substitution-feature (men, sang),
by suppletion (went, worse), or by a minus-feature (French vert,
masculine). We class these words as secondary derivatives and recognize
their peculiarity by calling them secondary morpheme-words.

13. 8. The bound forms which in secondary derivation are added
to the underlying form, are called affixes. Affixes which precede
the underlying form are prefixes, as be- in be-head; those which
follow the underlying form are called suffixes, as [-ez] in glasses
or -ish in boyish. Affixes added within the underlying form are
called infixes; thus, Tagalog uses several infixes which are added
before the first vowel of the underlying form: from [ˈsuːlat] ‘a
writing’ are derived [suˈmuːlat] ‘one who wrote,’ with the infix
[-um-], and [siˈnuːlat] ‘that which was written,’ with infix [-in-].
Reduplication is an affix that consists of repeating part of the
underlying form, as Tagalog [suː-ˈsuːlat] ‘one who will write,’
[ˈga:mit] ‘thing of use’: [ga:-ˈga:mit] ‘one who will use.’ Reduplication
may be of various extent: Fox [wa:pamɛ:wa] ‘he looks at
him’: [wa:-wa:pamɛ:wa] ‘he examines him,’ [wa:pa-wa:pamɛ:wa]
‘he keeps looking at him.’ It may differ phonetically in some conventional
way from the underlying word: ancient Greek [ˈphajnej]
‘it shines, it appears’: [pam-ˈphajnej] ‘it shines brightly;’ Sanskrit
[ˈbharti] ‘he bears’: [ˈbi-bharti] ‘he bears up,’ [ˈbhari-bharti]
‘he bears off violently.’

13. 9. We have seen that when forms are partially similar, there
may be a question as to which one we had better take as the underlying
form, and that the structure of the language may decide this
question for us, since, taking it one way, we get an unduly complicated
description, and, taking it the other way, a relatively
simple one. This same consideration often leads us to set up an
artificial underlying form. For instance, in German the voiced
mutes [b, d, g, v, z] are not permitted finals, and are in final
position replaced by the corresponding unvoiced phonemes. Accordingly
we get sets like the following:

tableau Underlying word | Derived word | Gras | grasen | Haus | hausen | Spasz | spaszen | aus | auszen218

It is evident that if we took the underlying words in their actual
shape as our basic forms, we should have to give a long list to tell
which ones appeared in derivatives with [z] instead of [s]. On the
other hand, if we start from an artificial underlying form with
[-z], as [gra:z-, hawz-], in contrast with [spa:s, aws], we need give
no list and can account for the uniform final [-s] which actually
appears in the independent forms, by the rule of permitted finals.
Similarly for the other voiced mutes, as in

tableau rund | runde | bunt | bunte

where we set up a theoretical basic form [rund-] in contrast with
[bunt]. We have seen that in some languages these theoretical
forms appear also in the phrase, by reminiscent sandhi (§ 12.5).

Similarly, some languages permit no final clusters and yet show
included free forms with clusters. Compare the following noun-forms
in Menominiː

tableau Singular (suffix zero) | Plural (suffix [-an]) | ‘my hand’ | ‘my hands’ | ‘a heart’ | ‘hearts’ | ‘birch-bark’ | ‘pieces of birchbark’ | ‘my thumb’ | ‘my thumbs’ | ‘medicine-bundle’ | ‘medicine-bundles.’

It is evident that a description which took the singular forms as
a basis would have to show by elaborate lists what consonants, as,
[k, j, s, čj, tj], are added before a suffix; the simple and natural
description is to take as a starting-point the free forms not in their
absolute shape, but in the form which appears before suffixes, as
[wiːkiːhs-] and the like.

Another example is furnished by Samoan, which permits no
final consonants at all, and therefore has sets like the following:

tableau Without suffix | With suffix [-ia] | ‘weep’ | ‘wept’ | ‘drink’ | ‘drunk’ | ‘enter’ | ‘entered.’

It is clear that a useful description will here set up the basic forms
in theoretical shape, as [tanis-, inum-, uluf-].219

13. 10. Modulation of secondary phonemes often plays a part
in morphologic constructions. In English, affixes are normally
unstressed, as in be-wail-ing, friend-li-ness and the like. In our
foreign-learned vocabulary, shift of stress to an affix is a taxeme
in many secondary derivatives. Thus, some suffixes have pre-suffixal
: the accent is on the syllable before the suffix, regardless
of the nature of this syllable; thus, -ity in able : ability, formal :
formality, major : majority; [-jn̩] in music : musician, audit : audition,
educate : education; [-ik] in demon : demonic, anarchist : anarchistic,
angel : angelic. In the derivation of some of our foreign-learned
nouns and adjectives from verbs, the stress is put on the
prefix: from the verb insert [inˈsr̩t] we derive the noun insert
[ˈinsr̩t]; similarly, contract, convict, convert, converse, discourse, protest,
project, rebel, transfer. In other cases this modulation appears
along with a suffix: conceive : concept, perceive : percept, portend : portent;
in some, the underlying verb has to be theoretically set up, as
in precept.

In some languages modulation has greater scope. In Sanskrit,
with some suffixes the derivative form keeps the accent of the
underlying form:

[ˈke:ça-] ‘hair’ : [ke:ça-vant-] ‘having long hair’
[puˈtra-] ‘son’ : [puˈtra-vant-] ‘having a son.’

Others are accompanied by shift of accent to the first syllable:

[ˈpuruša-] ‘man’ : [ˈpa:wruš-e:]a-j ‘coming from man’
[vaˈsti-] ‘bladder’ : [ˈva:st-e:ja-] ‘of the bladder.’

Others have presuffixal accent:

[ˈpuruša-] ‘man’ : [puruˈša-ta:-] ‘human nature’
[de:ˈva-] ‘god’ : [de:ˈva-ta:-] ‘divinity.’

Other affixes are themselves accented:

[ˈṛši-] ‘sage’ : [a:rš-e:ˈja-] ‘descendant of a sage’
[saˈrama:-] (proper noun) : [sa:ram-e:ˈja-] ‘descended from

Others require an accentuation opposite to that of the underlying

[ˈatithi-] ‘guest’ : [a:tiˈth-ja-] ‘hospitality’
[paliˈta-] ‘gray’ : [ˈpa:lit-ja-] ‘grayness.’220

Tagalog uses both stress and vowel-lengthening as auxiliary
phonemes; three suffixes of the form [-an] differ in the treatment of
these modulations.

Suffix [-an]1 is characterized by presuffixal stress and by long
vowel in the first syllable of the underlying form:

[ˈiːbig] ‘love’ : [iːˈbiːgan] ‘love-affair’
[iˈnum] ‘drink’ : [iːˈnuːman] ‘drinking-party.’

The meaning is ‘action (often reciprocal or collective) by more
than one actor.’

Suffix [-an]2 is stressed when the underlying word has stress on
the first syllable; otherwise it is treated like [-an]1:

[ˈtuːlug] ‘sleep’ : [tuluˈgan] ‘sleeping-place’

[kuˈluŋ] ‘enclose’ : [kuːˈluːŋan] ‘place of imprisonment.’

The meaning is ‘place of action, usually by more than one actor,
or repeated.’

Suffix [-an]3 has presuffixal stress when the underlying word is
stressed on the first syllable; it is stressed when the underlying
word is stressed on the last syllable; there is no vowel-lengthening
beyond what is demanded by the phonetic pattern:

(a) [ˈsa:giŋ] ‘banana’ : [saˈgiːŋan] ‘banana-grove’
[kuˈluŋ] ‘enclose’ : [kuluˈŋan] ‘cage, crate’

(b) [ˈpuːtul] ‘cut’ : [puˈtuːlan] ‘that which may be cut from’
[laˈkas] ‘strength’ : [lakaˈsan] ‘that upon which strength may
be expended.’

The meaning is (a) ‘an object which serves as locality of the
underlying object, action, etc.,’ and (b) ‘that which may be acted

In languages with auxiliary phonemes of pitch, these may play
a part in morphology. Thus, in Swedish, the suffix -er of agent-nouns
shows the normal compound word-pitch of polysyllables
(§ 7.7) in the resultant form: the verb-stem [le:s-] ‘read’ forms
läser [˅le:ser] ‘reader;’ but the -er of the present tense demands
simple word-pitch in the resultant form: (han) läser [ˈle:ser] ‘(he)

13. 11. In all observation of word-structure it is very important
to observe the principle of immediate constituents. In Tagalog, the
underlying form [ˈta:wa] ‘a laugh’ appears reduplicated in the
derivative [ta:ˈta:wa] ‘one who will laugh;’ this form, in turn,
221underlies a derivative with the infix [-um-], namely [tuma:ˈta:wa]
‘one who is laughing.’ On the other hand, the form [ˈpiːlit] ‘effort’
first takes the infix [-um-], giving [puˈmiːlit] ‘one who compelled,’
and is then reduplicated, giving [-puːpuˈmiːlit], which underlies
[nag-puːpuˈmiːlit] ‘one who makes an extreme effort.’ Close
observation of this principle is all the more necessary because now
and then we meet forms which compromise as to immediate constituents.
Tagalog has a prefix [paŋ-], as in [aˈtip] ‘roofing’ :
[paŋ-aˈtip] ‘that used for roofing; shingle.’ The [ŋ] of this prefix
and certain initial consonants of an accompanying form are subject
to a phonetic modification — we may call it morphologic sandhi
by which, for instance, our prefix joins with [ˈpuːtul] ‘a cut’ in the
derivative [pa-ˈmuːtul] ‘that used for cutting,’ with substitution
of [m] for the combination of [-ŋ] plus [p-]. In some forms, however,
we find an inconsistency as to the structural order; thus, the form
[pa-mu-ˈmurtul] ‘a cutting in quantity’ implies, by the actual
sequence of the parts, that the reduplication is made “before” the
prefix is added, but at the same time implies, by the presence of
[m-] for [p-] in both reduplication and main form, that the prefix is
added “before” the reduplication is made. A carelessly ordered description
would fail to bring out the peculiarity of a form like this.

13. 12. In languages of complex morphology we can thus observe
a ranking of constructions: a complex word can be described
only as though the various compoundings, affixations, modifications,
and so on, were added in a certain order to the basic form.
Thus, in English, the word actresses consists, in the first place,
of actress and [-ez], just as lasses consists of lass and [-ez]; actress,
in turn consists of actor and -ess, just as countess consists of count
and -ess; actor, finally, consists of act and [-r̩]. There would be
no parallel for a division of actresses, say into actor and -esses. In
languages of this type, then, we can distinguish several ranks
of morphologic structure.

In many languages these ranks fall into classes: the structure
of a complex word reveals first, as to the more immediate constituents,
an outer layer of inflectional constructions, and then
an inner layer of constructions of word-formation. In our last
example, the outer, inflectional layer is represented by the construction
of actress with [-ez], and the inner, word-formational
layer by the remaining constructions, of actor with -ess and of
act with [-r̩].222

This distinction cannot always be carried out. It is based on
several features. The constructions of inflection usually cause
closure or partial closure (§ 12.11), so that a word which contains
an inflectional construction (an inflected word) can figure as a
constituent in no morphologic constructions or else only in certain
inflectional constructions. The English form actresses, for instance,
can enter into only one morphologic construction, namely the
derivation of the possessive adjective actresses' (with the zero-alternant
of [-ez, -z, -s], § 13.7). This latter form, in turn, cannot
enter into any morphologic construction; it has complete closure.

Another peculiarity of inflection, in contrast with word-formation,
is the rigid parallelism of underlying and resultant
forms. Thus, nearly all English singular nouns underlie a derived
plural noun, and, vice versa, nearly all English plural nouns are
derived from a singular noun. Accordingly, English nouns occur,
for the most part in parallel sets of two: a singular noun (hat) and
a plural noun derived from the former (hats). Given one of these,
the speaker is usually capable of producing the other. Each such
set of forms is called a paradigmatic set or paradigm, and each
form in the set is called an inflected form or inflection. Some languages
have large paradigms, which contain many inflections.
In Latin, for instance, the verb appears in some 125 inflectional
forms, such as amāre ‘to love,’ amō ‘I love,’ amās ‘thou lovest,’
amat ‘he loves,’ amāmus ‘we love,’ amem ‘I may love,’ amor
‘I am loved,’ and so on; the occurrence of one form usually guarantees
the occurrence of all the others. It is this parallelism of
the inflections which forces us to treat a single phonetic form,
like sheep as a set of homonyms, a singular noun sheep (corresponding
to lamb) and a plural noun sheep (corresponding to lambs).
It is this parallelism also, which leads us to view entirely different
phonetic forms, like go : went, as morphologically related
(by suppletion): go as an infinitive (parallel, say, with show) and
went as a past-tense form (parallel, then, with showed).

The parallelism, to be sure, is sometimes imperfect. Defective
paradigms lack some of the inflections; thus, can, may, shall, will,
must have no infinitive, must has no past tense, scissors no singular.
If, as in these cases, the lacking form happens to underlie the actually
existing ones, we do best to set up a theoretical underlying
form, such as a non-existent infinitive *can or singular *scissor-.
On the other hand, some irregular paradigms are over-differentiated.
223Thus, corresponding to a single form of an ordinary paradigm
like play (to play, I play, we play), the paradigm of be has
three forms (to be, I am, we are), and, corresponding to the single
form played, it has the forms (I) was, (we) were, been. The existence
of even a single over-differentiated paradigm implies homonymy
in the regular paradigms.

The parallelism of inflected forms goes hand in hand with a
further characteristic: the different inflections differ in syntactic
function. If we say the boys chauffe, our syntactic habit of congruence
(§ 12.7) requires us, when the boy is the actor, to supply also
the form chauffes. In the case of the present and past inflections
of the English verb this is not true: the parallelism of plays : played
is not required by any habits of our syntax, but is carried out
none the less rigidly.

If there are several ranks of inflection, we get compound paradigms;
the inflections of the English noun, for instance, consist of
an outer construction, the derivation of the possessive adjective,
and an inner one, the derivation of the plural:

tableau singular | plural. | nominative-accusative | possessive adjective

In the Latin verb we find a very complicated compound paradigm:
an outer layer for different actors or undergoers, distinguished
as to person (speaker, hearer, third person), number (singular,
plural), and voice (actor, undergoer), an inner layer for
differences of tense (present, past, future) and mode (real, hypothetical,
unreal), and an innermost layer for a difference as to
completion of the act (imperfectic, perfectic).

13. 13. We come, finally, to an important characteristic of
inflection, akin to those we have mentioned, the derivational unity
of paradigms The inflectional forms of a paradigm do not each
enter into composition and derivation, but the paradigm as a
whole is represented by some one form. In English, the forms of
a noun-paradigm are represented by the singular, as in man-slaughter,
mannish, and those of the verb-paradigm by the infinitive,
as in playground, player. An English paradigm consists
of an underlying word (itself a member of the paradigm) and some
secondary derivatives containing this underlying word; as a constituent
in further derivation and composition, the paradigm, as
224a whole, is represented by the underlying form; the English language,
accordingly, may be said to have word-inflection, word-derivation,
and word-composition.

In many languages, especially in those which have a more complex
morphology, none of the forms in a paradigm can conveniently
be viewed as underlying the others. Thus, the regular paradigms
of the German verb contain a common element which is not equal
to any of the inflectional forms. For instance, the paradigm
represented by the forms lachen [ˈlax-en] ‘(to) laugh,’ (ich)
[lax-e] ‘(I) laugh,’ (er) lacht [lax-t] ‘(he) laughs,’ (er) lachte
[lax-te] ‘(he) laughed,’ gelacht [ge-ˈlax-t] ‘laughed’ (participle),
and so on, shows a common element lach- [lax-] in all the inflectional
forms, but none of these inflectional forms consists simply
of the element lach- without an affix. In secondary derivation and
composition the paradigm is represented by this same form, as in
Lacher [ˈlax-er] ‘laugher’ and Lachkrampf [ˈlax-ˌkrampf] ‘laughing-spasm.’
This lach-, strictly speaking, is a bound form; it is called
the kernel or stem of the paradigm. The German verb is an example
of stem-inflection, stem-derivation, and stem-composition. In our
description, we usually treat the stem as if it were a free form.

In some languages of this type, the common element of the
paradigm differs from the stem which represents the paradigm
in derivatives and compounds. Thus, an ancient Greek noun-paradigm
has stem-inflection. It contains a common element, a
kernel, much like the German verb-stem, e.g. [hipp-] ‘horse’:

tableau Singular | Plural | nominative | vocative | accusative | dative | genitive

In secondary derivation, however, this paradigm is represented
not by the common element [hipp-], but by a special deriving-form
[hipp-o-] as in [hipˈpo-te:s] ‘horseman,’ or with loss of the [o]
by phonetic modification, in [hipp-iˈkos] ‘pertaining to horses.’
Similarly, as a compound-member, the paradigm is represented
by a special compounding-form, homonymous with the preceding:
[hippo-ˈkantharos] ‘horse-beetle.’ Thus, we distinguish between
the kernel [hipp-], which actually (subject, however, in principle,
225to phonetic modification) appears in all the forms, and the stem
[hipp-o-], which underlies the further derivatives.

Some exceptions to the principle of paradigmatic unity are only
apparent. The possessive-adjective form in the English compounds
like bull's-eye or the plural form in longlegs are due, as we shall see,
to the phrasal structure of these compounds. Real exceptions do,
however, occur. German has a suffix -chen [-xen] ‘small,’ which
forms secondary derivatives from nouns, as: Tisch [tiš] ‘table’ :
Tischchen [ˈtiš-xen] ‘little table.’ In the system of German morphology,
this is a construction of word-formation, but in a certain
few instances the suffix [-xen] is added to nouns which already have
plural inflection: beside Kind [kint] ‘child’: Kindchen [ˈkint-xen]
‘little child,’ the plural inflection Kinder [ˈkinder] ‘children’ underlies
the derivative Kinderchen [ˈkinder-xen] ‘little children.’ If a
language contained too many cases of this sort, we should simply
say that it did not distinguish such morphologic layers as are
denoted by the terms inflection and word-formation.226

Chapter 14
Morphologic Types

14. 1. Of the three types of morphologic constructions which can
be distinguished according to the nature of the constituents —
namely, composition, secondary derivation, and primary derivation
(§ 13.3) — the constructions of compound words are most
similar to the constructions of syntax.

Compound words have two (or more) free forms among their
immediate constituents (door-knob). Under the principle of immediate
constituents, languages usually distinguish compound
words from phrase-derivatives (as, old-maidish, a secondary derivative
with the underlying phrase old maid), and from de-compounds
(as, gentlemanly, a secondary derivative with the underlying compound
word gentleman). Within the sphere of compound words,
the same principle usually involves a definite structural order; thus,
the compound wild-animal-house does not consist, say, of three
members wild, animal, and house, and not of the members wild
and animal-house, but of the members wild animal (a phrase) and
house; and, similarly, the compound doorknob-wiper consists, unmistakably,
of the members door-knob and wiper, and not, for instance,
of door and knob-wiper.

The grammatical features which lead us to recognize compound
words, differ in different languages, and some languages, doubtless,
have no such class of forms. The gradations between a word and a
phrase may be many; often enough no rigid distinction can be
made. The forms which we class as compound words exhibit some
feature which, in their language, characterizes single words in
contradistinction to phrases.

In meaning, compound words are usually more specialized than
phrases; for instance, blackbird, denoting a bird of a particular
species, is more specialized than the phrase black bird, which denotes
any bird of this color. It is a very common mistake to try
to use this difference as a criterion. We cannot gauge meanings
accurately enough; moreover, many a phrase is as specialized in
meaning as any compound: in the phrases a queer bird and meat
227and drink, the words bird, meat are fully as specialized as they are
in the compounds jailbird and sweetmeats.

14. 2. In languages which use a single high stress on each
word, this feature distinguishes compound words from phrases.
In English the high stress is usually on the first member; on the
other member there is a lesser stress, as in door-knob [ˈdowr-ˌnɑb],
upkeep [ˈop-ˌkijp]. Certain compounds have the irregularity of
leaving the second member unstressed, as in gentleman [ˈǰentl̩mn̩],
Frenchman [ˈfrenčmn̩]; contrast milkman [ˈmilk-ˌmɛn]. Certain
types of compounds, chiefly some whose members are adverbs and
prepositions, stress the second member: without, upon. Accordingly,
wherever we hear lesser or least stress upon a word which
would always show high stress in a phrase, we describe it as a
compound-member: ice-cream [ˈajs-ˌkrijm] is a compound, but
ice cream [ˈajs ˈkrijm] is a phrase, although there is no denotative
difference of meaning. However, a phrase as prior member in a
compound keeps all its high stresses: in wild-animal-house [ˈwajldˈ-ˈɛniml̩-ˌhaws]
the stress assures us only that house is a compound-member;
the rest of the structure is shown by other criteria.

As to the phonetic pattern, compound words are generally
treated like phrases: in English, clusters like [vt] in shrovetide or
[nn] in pen-knife do not occur within simple words. Sandhi-like
phonetic modifications mark a compound as a single word only
when they differ from the sandhi of syntax in the same language.
Thus gooseberry [ˈguwzbr̩ij] is marked as a compound because the
substitution of [z] for [s] is not made in English syntax, but only in
morphology, as in gosling [ˈgɑzliŋ]. Similarly, in French, pied-à-terre
[pjet-a-tɛ:r] ‘temporary lodging’ (literally ‘foot-on-ground’) beside
pied [pje] ‘foot,’ or pot-au-feu [pɔt-o-fø] ‘broth’ (literally
‘pot-on-the-fire’) beside pot [po] ‘pot,’ or vinaigre [vin-ɛgr] ‘vinegar’
(literally ‘sour-wine’) beside vin [vɛn] ‘wine,’ are marked as compounds,
because French nouns do not exhibit these types of sandhi
in the phrase, but only in word-constructions, such as pieter [pjete]
‘toe the mark,’ potage [pɔta:ž] ‘thick soup,’ vinaire [vinɛ:r] ‘pertaining
to wine;’ contrast, for instance, the phrase vin aigre [vɛn ɛgr]
‘sour wine.’

More striking phonetic modifications may mark a compound;
thus, in the following examples the prior member suffers greater
modification than it does in any phrase of its language: holy
[ˈhowlij] : holiday [ˈhɑlidej], moon : Monday, two [tuw] : twopence
228[topn̩s]; Old English [ˈfe:ower] ‘four’ : [ˈfitðer-ˌfe:te] ‘four-footed;’
the second member, in Sanskrit [na:wh] ‘ship’ : [ati-ˈnuh] ‘gone
from the ship;’ ancient Greek [paˈte:r] ‘father’ : [ew-ˈpato:r] ‘well-fathered;’
Gothic dags ‘day’ : fidur-dogs ‘four days old;’ both
members, in English breakfast [ˈbrekfest], blackguard [ˈblɛgr̩d],
boatswain [ˈbowsn̩], forecastle [ˈfowksl̩]; in some cases there is also
a variant form without modification, as in forehead [ˈfɑred], waistcoat
[ˈwesket]. In extreme cases, of course, the form may be so
unlike the independent word that we may hesitate between calling
it a compound-member or an affix: a form like fortnight [ˈfort-ˌnajt,
ˈfortnet] lies on the border between compound and simple word.

The order of the members in a compound word may be fixed,
while that of the phrase is free, as in bread-and-butter [ˈbred-n̩-ˌbotr̩]
‘slices of bread spread with butter,’ contrasting with the phrase,
as in she bought bread and butter, she bought butter and bread. This
criterion is likely to break down, however, because the order in a
phrase, too, may be fixed: we have also a specialized phrase [ˈbred
n̩ ˈbotr] with the same order and the same meaning as the compound.
Contrasting order is a surer mark: French blanc-bec
[blɑn-bɛk] ‘callow young person’ (literally ‘white-beak’) is characterized
as a compound, because adjectives like blanc in the phrase
always follow their noun: bee blanc ‘white beak.’ English examples
are to housekeep, to backslide, to undergo, since in a phrase a noun
goal like house and adverbs of the type back, under would follow
the verb (keep house, slide back).

14. 3. The commonest, but also the most varied and most
difficult to observe, of the features which lead us to distinguish
compound words from phrases, are grammatical features of selection.

The plainest contrast appears in languages with stem-composition
(§ 13.13). A stem like German lach-, which represents a whole verb
paradigm in a German compound like Lachkrampf [ˈlax-ˌkrampf]
‘laughing-spasm,’ but does not actually occur as an independent
word, makes the compound unmistakably different from any
phrase. Even more plainly, a compounding-stem, such as ancient
Greek [hippo-] ‘horse,’ may differ formally from all the inflections
of its paradigm, and, in any case, characterizes a compound by its
invariability; thus, [hippo-] joins some other stem, such as [ˈkantharo-]
‘beetle,’ to form a compound stem, [hippo-ˈkantharo-]
‘horse-beetle,’ but remains unchanged in all the inflectional forms
229of this compound: nominative [hippoˈkantharo-s], accusative
[hippoˈkantharo-n], and so on.

Even when the compound-member is formally equal to some
word, it may characterize the compound. In ancient Greek a
noun-stem is inflected by means of suffixes. Accordingly, the
first member of a compound noun-stem will remain the same in all
forms of the paradigm. Thus, the phrase ‘new city’ will show
various inflectional forms of two paradigms:

nominative [neˈa: ˈpolis]
accusative [neˈa:n ˈpolin]
genitive [ne˅a:s ˈpoleo:s],

and so on, but the compound stem [neˈa:-poli-] ‘Naples,’ whose
first member is in nominative singular form, will show this first
member unchanged in all the inflections:

nominative [neˈa:polis]
accusative [neˈa:polin]
genitive [nea:ˈpoleo:s].

In German, the adjective has word-inflection; the underlying
form is used as a complement of verbs: Das ist rot [das ist ˈro:t]
‘that is red,’ and the derived inflections appear as modifiers of
nouns: roter Wein [ˈro:ter ˈvajn] ‘red wine.’ The absence of inflectional
suffixes therefore characterizes the compound-member in
a form like Rotwein [ˈro:t-ˌvajn] ‘red-wine.’

The use of prefixes and suffixes may decide for us what is the
beginning and what the end of a word or stem. In German, the
past participle of verbs is formed by the addition to the stem of a
prefix [ge-] and a suffix [-t], as in gelacht [ge-ˈlax-t] ‘laughed.’ The
position of these affixes, accordingly, shows us that a form like
geliebkost [ge-ˈliːpˌko:s-t] ‘caressed’ is one word, derived from a
compound stem, but that a form like liebgehabt [ˈliːp ge-ˌhap-t]
‘liked’ is a two-word phrase. This gives us a standard for the
classification of other inflectional forms, such as the infinitives
liebkosen [ˈliːp-ˌko:zen] ‘to caress’ and liebhaben [ˈliːp ˌha:ben]
‘to like.’

Sometimes the compound-member resembles an inflectional
form, but one which would be impossible in the phrase. The
[-z, -s] on the prior members of bondsman, kinsman, landsman,
marksman resembles the possessive-adjective suffix, but possessive
adjectives like bond's, land's and so on, would not be so used in the
230phrase. In French, the adjective grande [grɑnd] ‘great,’ as in une
grande maison
[yn grɑnd mezon] ‘a big house,’ drops the final consonant
(§ 13.7) to make the inflectional form used with masculine
nouns: un grand gargonn grɑn garsɔn] ‘a big boy;’ but, as a
compound-member, the latter form appears also with certain
feminine nouns: grand'mere [grɑn-mɛ:r] ‘grandmother,’ grand'porte
[grɑn-pɔrt] ‘main entry.’ Compound-members of this type are
especially common in German: Sonnenschein [ˈzonen-ˌšajn] ‘sunshine’
has the prior member Sonne in a form which, as a separate
word in a phrase, could only be plural; in Geburtstag [geˈburts-ˌta:k]
‘birthday,’ the [-s] is a genitive-case ending, but would not be
added, in an independent word, to a feminine noun like die Geburt

A compound-member may be characterized by some feature of
word-formation which differs from what would appear in an
independent word. In ancient Greek there was a highly irregular
verb-paradigm, containing such forms as [daˈmao:] ‘I tame,’
[eˈdme:the:] ‘he was tamed,’ and so on, which grammarians conveniently
describe on the basis of a stem-form [dame:-]. From this
paradigm there is derived, on the one hand, the independent agent-noun
[dme:ˈte:r] ‘tamer,’ and, on the other hand, with a different
suffix, an agent-noun [-damo-], which is used only as a second
member of compound words, as in [hipˈpo-damo-s] ‘horse-tamer.’
Compounds with special features of word-formation are known as
synthetic compounds. Synthetic compounds occurred especially in
the older stages of the Indo-European languages, but the habit is
by no means extinct. In English, the verb to black underlies the
independent agent-noun blacker (as in a blacker of boots), but forms
also, with a zero-element, the agent-noun -black which appears in
the compound boot-black; similarly, to sweep forms sweeper and the
second member of chimney-sweep. Even forms like long-tailed or
red-bearded are not aptly described as containing the words tailed,
bearded (as in tailed monkeys, bearded lady); the natural starting-point
is rather a phrase like long tail or red beard, from which they
differ by the presence of the suffix -ed. This is the same thing as
saying that we use compounds of the type long-tailed, red-bearded
regardless of the existence of words like tailed, bearded: witness
forms like blue-eyed, four-footed, snub-nosed. Another modern
English synthetic type is that of three-master, thousand-legger.

In English, we freely form compounds like meat-eater and meat-eating,
231but not verb-compounds like *to meat-eat; these exist only
in a few irregular cases, such as to housekeep, to bootlick. Now, to
be sure, words like eater and eating exist alongside the compounds;
the synthetic feature consists merely in the restriction that a
phrase like eat meat is paralleled by compounds only when -er or
-ing is at the same time added. We may designate the types meat-eating
and meat-eater as semi-synthetic compounds.

14. 4. Among the word-like features of the forms which we
class as compound words, indivisibility (§ 11.6) is fairly frequent:
we can say black — I should say, bluish-black — birds, but we
do not use the compound word blackbird with a similar interruption.
In some instances, however, other features may lead us to
class a form as a compound word, even though it is subject to
interruption. In Fox, a form like [ne-pjɛ:či-wa:pam-a:-pena] ‘we
have come to see him (her, them)’ has to be classed as a compound
word, because the inflectional prefix [ne-] ‘I (but not thou)’ and
the inflectional suffixes [-a:-] ‘him, her, them’ and [-pena] ‘plural
of first person’ unmistakably mark the beginning and end of a
word (§ 14.3). The members of the compound are the particle
[pjɛ:či] ‘hither’ and the verb-stem [wa:pam-] ‘see (an animate
object).’ Nevertheless, the Fox language sometimes inserts words
and even short phrases between the members of such compounds,
as in [ne-pjɛ:čci-keta:nesa-wa:pam-a:-pena] ‘we have come to see
her, thy daughter.’ In German, compound-members can be combined
serially; Singvögel [ˈziŋ-ˌfø:ɡel] ‘songbirds,’ Raubvogel
[ˈrawp-ˌfø:ɡel] ‘birds of prey,’ Sing- oder Raubvogel [ˈziŋ-o:der-ˈrawp-ˌfø:gel]
‘songbirds or birds of prey.’

Generally, a compound-member cannot, like a word in a
phrase, serve as a constituent in a syntactic construction. The
word black in the phrase black birds can be modified by very (very
black birds
), but not so the compound-member black in blackbirds.
This feature serves to class certain French forms as compound
words: thus, sage-femme [sa:ž-fam] ‘midwife’ is to be classed as a
compound, in contrast with a homonymous phrase meaning ‘wise
woman,’ because only in the latter can the constituent sage ‘wise’
be accompanied by a modifier: très sage femme [trɛ sa:ž fam]
‘very wise woman.’ This restriction, like the preceding, is occasionally
absent in forms which by other features are marked as
compound words. In Sanskrit, where stem-composition plainly
marks the prior member of compound words, this member is
232nevertheless occasionally accompanied by a modifying word, as
in [čitta-pramaˈthiniː de:ˈva:na:m ˈapi] ‘mind-disturbing of-gods
even,’ that is ‘disturbing to the minds even of gods,’ where the
genitive plural noun (‘of gods’) is a syntactic modifier of the
compound-member [čitta-] ‘mind.’

14. 5. The description and classification of the forms which
the structure of a language leads us to describe as compound words,
will depend upon the characteristic features of this language.
Linguists often make the mistake of taking for granted the universal
existence of whatever types of compound words are current
in their own language. It is true that the main types of compound
words in various languages are somewhat similar, but this similarity
is worthy of notice; moreover, the details, and especially
the restrictions, vary in different languages. The differences are
great enough to prevent our setting up any scheme of classification
that would fit all languages, but two lines of classification are
often useful.

One of these two lines of classification concerns the relation of
the members
. On the one hand, we have syntactic compounds,
whose members stand to each other in the same grammatical
relation as words in a phrase; thus, in English, the members of
the compounds blackbird and whitecap (the difference between
these two examples will concern us later) show the same construction
of adjective plus noun as do the words in the phrases black
and white cap. On the other hand, we have asyntactic compounds
like door-knob, whose members stand to each other in a
construction that is not paralleled in the syntax of their language
— for English has no such phrasal type as *door knob.

The syntactic compound differs from a phrase only in the essential
features which (in its language) distinguish compound
words from phrases — in English, then, chiefly by the use of
only one high stress. It may differ lexically from the corresponding
phrase, as does dreadnaught; the corresponding phrase, dread
, has an archaic connotation, and the normal phrase would
be fear nothing. We can set up sub-classes of syntactic compounds
according to the syntactic constructions which are paralleled by
the members, as, in English, adjective with noun (blackbird, whitecap,
bull's-eye), verb with goal noun (lickspittle, dreadnaught),
verb with adverb (gadabout), past participle with adverb (castaway),
and so on.233

Many compounds are intermediate between the syntactic and
asyntactic extremes: the relation of the members parallels some
syntactic construction, but the compound shows more than the
minimum deviation from the phrase. For instance, the compound
verb to housekeep differs from the phrase keep house by the simple
feature of word-order. In such cases we may speak of various
kinds of semi-syntactic compounds. The difference of order appears
also in upkeep versus keep up, and in the French blanc-bec versus
bec blanc (§ 14.2). In turnkey versus turn the key or turn keys, the
difference lies in the use of the article or of the number-category.
Even types like blue-eyed, three-master, meat-eater, viewed as synthetic
compounds, can be said to correspond to blue eyes, three
, eat meat, and to differ from these phrases by simple formal
characteristics, including the addition of the bound forms -ed, -er
to the second member. In French, boîte-à-lettres [bwɑ:t-a-lɛtr],
literally ‘box-for-letters,’ and boîte-aux-lettres [bwɑ:t-o-lɛtr], literally
‘box-for-the-letters,’ both meaning ‘mail-box, post-box,’
differ in the choice of preposition and in the use of the article
from the normal phrasal type, which would give boîte pour des
[bwɑt puːr de lɛtr] ‘box for letters;’ the use of d and certain
other prepositions in place of more specific ones, and differences
of article (especially of zero in place of the phrasal article represented
by the form des), are in French well-marked features which
enable us to set up a class of semi-syntactic compounds.

Where semi-syntactic compounds are definable, they can be
further classified in the same manner as syntactic compounds:
thus, in the semi-syntactic blue-eyed the members have the same
construction as in the syntactic blackbird, in three-master the same
as in three-day, in housekeep, turnkey the same as in lickspittle, in
upkeep the same as in gadabout.

Asyntactic compounds have members which do not combine
in syntactic constructions of their language. Thus, in door-knob,
horsefly, bedroom, salt-cellar, tomcat we see two nouns in a construction
that does not occur in English syntax. Other asyntactic types
of English compounds are illustrated by fly-blown, frost-bitten
crestfallen, footsore, fireproof, foolhardyby-law, by-path, ever-glade
dining-room, swimming-holebindweed, cry-baby, driveway,
playground, blowpipebroadcast, dry-clean, foretellsomewhere,
everywhere, nowhere. Compounds with obscure members,
such as smokestack, mushroom, or with unique members, such as
234cranberry, huckleberry, zigzag, choo-choo, are, of course, to be classed
as asyntactic.

Although the relation between the members of asyntactic compounds
is necessarily vague, yet we can sometimes extend the
main divisions of syntactic and semi-syntactic compounds to
cover also the asyntactic class. In English, for instance, the coordinative
or copulative relation which we see in a semi-syntactic
compound like bittersweet (compare the phrase bitter and sweet), can
be discerned also in asyntactic compounds like zigzag, fuzzy-wuzzy,
choo-choo. Most asyntactic compounds seem to have a
kind of attribute-and-head construction: door-knob, bulldog, cranberry.
To the extent that one can carry out this comparison, one
can therefore distinguish between copulative compounds (Sanskrit
dvandva) and determinative (attributive or subordinative) compounds
(Sanskrit tatpurusha); these divisions will cross those
of syntactic, semi-syntactic, and asyntactic compounds. One
may even be able to mark off smaller divisions. The Hindu
grammarians distinguished among copulative compounds a special
sub-group of repetitive (amredita) compounds, with identical members,
as in choo-choo, bye-bye, goody-goody. In English, we can
mark off also a class in which the members show only some elementary
phonetic difference, as zigzag, flimflam, pell-mell, fuzzy-wuzzy.
The Hindus found it convenient to set off, among the
determinatives, a special class of syntactic attribute-and-head compounds
(karmadharaya), such as blackbird.

14. 6. The other frequently usable line of classification concerns
the relation of the compound as a whole to its members.
One can often apply to compounds the distinction between endocentric
and exocentric constructions which we met in syntax
(§ 12.10). Since a blackbird is a kind of a bird, and a door-knob
a kind of a knob, we may say that these compounds have the
same function as their head members; they are endocentric. On
the other hand, in gadabout and turnkey the head member is an
infinitive verb, but the compound is a noun; these compounds
are exocentric (Sanskrit bahuvrihi). To take a copulative type as
an example, the adjective bittersweet (‘bitter and sweet at the
same time’) is endocentric, since the compound, like its co-ordinated
members, bitter and sweet, has the function of an adjective,
but the plant-name bittersweet is exocentric, since, as a noun, it
differs in grammatical function from the two adjective members.
235Another type of English exocentric compounds consists of adjectives
with noun head: two-pound, five-cent, half-mile, (in) applepie

The difference of form-class may be less radical, but still recognizable
in the system of the language. In English, the nouns
longlegs, bright-eyes, butterfingers are exocentric, because they occur
both as singulars, and, with a zero-affix, as plurals (that longlegs,
those longlegs). In French, the noun rouge-gorge [ruːž-gɔrž] ‘robin’
(literally ‘red-throat’) is exocentric, because it belongs to the
masculine gender-class (le rouge-gorge ‘the robin’), while the head
member belongs to the feminine gender (la gorge ‘the throat’).
In the English type sure-footed, blue-eyed, straight-backed the synthetic
suffix [-ed, -d, -t] goes hand in hand with the exocentric value
(adjective with noun head); however, one might perhaps hesitate
as to the classification, since -footed, -eyed, -backed might be viewed
as adjectives (compare horned, bearded). Types like clambake, upkeep
are better described as endocentric, in English grammar, because
the head members -bake and -keep can be viewed as nouns of
action derived, with a zero-feature, from the verbs; if English did
not use many zero-features in derivation and did not form many
types of action nouns, we should have to class these compounds as
exocentric. Similarly, our description will probably work out best
if we class bootblack, chimney-sweep as endocentric, with -black and
-sweep as agent-nouns.

On the other hand, the large class of English compounds that is
exemplified by whitecap, longnose, swallow-tail, blue-coat, blue-stocking,
red-head, short-horn has noun function and a noun as head
member, and yet is to be classed as exocentric, because the construction
implies precisely that the object does not belong to the
same species as the head member: these compounds mean ‘object
possessing such-and-such an object (second member) of such-and-such
quality (first member).’ This appears in the fact that the
number-categories (longlegs) and the personal-impersonal categories
(nose … it; longnose … he, she) do not always agree.
In three-master, thousand-legger the synthetic suffix goes hand in
hand with this exocentric relation. Nevertheless, there are borderline
cases which may prevent a clear-cut distinction. The compound
blue-bottle is endocentric if we view the insect as ‘like a
bottle,’ but exocentric if we insist that the ‘bottle’ is only part of
the insect.236

The Hindus distinguished two special sub-classes among exocentric
compounds, namely numeratives (dvigu), nouns with a
number as prior member, such as, in English, sixpence, twelvemonth,
fortnight, and adverbials (avyayibhava), adverbs with noun head,
such as bareback, barefoot, hotfoot, or with noun subordinate, such as
uphill, downstream, indoors, overseas.

14. 7. In secondary derivative words we find one free form, a
phrase (as in old-maidish) or a word (as in mannish), as an immediate
constituent; in the latter case, the underlying word may
be a compound word (as in gentlemanly) or, in its own turn, a
derived word (as in actresses, where the underlying word actress
is itself a secondary derivative from the underlying word actor).
We have seen, however, that for the description of some languages,
we do well to set up theoretical underlying forms, namely stems,
which enable us to class certain forms as secondary derivatives
although, strictly speaking, they do not contain a free form
(§ 13.13). A similar device is called for in the description of forms
like English scissors, oats, where we set up a theoretical scissor-,
oat- as underlying forms, just as we class cranberry, oatmeal,
scissor-bill as compound words.

The underlying free form, actual or theoretical, is accompanied
either by an affix, or, as we saw, in Chapter 13, by a grammatical

In many languages, secondary derivatives are divided, first of
all, into inflectional forms and word-formational forms (§ 13.12),
but we may do well to recall that languages of this sort nevertheless
often contain border-line forms, such as, in English, beeves or
clothes, which predominantly resemble inflectional types, but show
a formal-semantic deviation. In the same way, learned [ˈlr̩ned],
drunken, laden, sodden, molten, and the slang broke ‘out of funds’
deviate from the strictly inflectional past participles learned
[lr̩nd], drunk, loaded, seethed, melted, broken.

The inflectional forms are relatively easy to describe, since they
occur in parallel paradigmatic sets; the traditional grammar of
familiar languages gives us a picture of their inflectional systems.
It may be worth noticing, however, that our traditional grammars
fall short of scientific compactness by dealing with an identical
feature over and over again as it occurs in different paradigmatic
types. Thus, in a Latin grammar, we find the nominative-singular
sign -s noted separately for each of the types amīcus ‘friend,’ lapis
237‘stone,’ dux ‘leader,’ tussis ‘cough,’ manus ‘hand,’ faciēs ‘face,’
when, of course, it should be noted only once, with a full statement
as to where it is and where it is not used.

Word-formation offers far more difficulty, and is largely neglected
in our traditional grammars. The chief difficulty lies in
determining which combinations exist. In very many cases we
have to resign ourselves to calling a construction irregular and
making a list of the forms. Only a list, for instance, can tell us
from which English male nouns we derive a female noun by means
of the suffix -ess, as in countess, lioness, and it will probably require
a subsidiary list to tell in which of these derivatives a final [r̩] is
replaced by non-syllabic [r], as in waiter : waitress, tiger : tigress
for the type without this change, as in author : authoress is probably
regular. Special cases, such as duke : duchess, master : mistress,
thief : thievess demand separate mention.

Once we have established a construction of this kind, we may be
able to set up a typical meaning and then, as in the case of inflection,
to look for parallels. Our suffix -ess, for instance, has a
definable linguistic meaning, not only because of the parallel
character of all the sets like count : countess, lion : lioness, but
also because English grammar, by the distinction of he : she,
recognizes the meaning of the -ess derivatives. Accordingly, we
are able to decide, much as we are in the case of inflection, whether
a given pair of forms, such as man : woman, does or does not show
the same relation. This enables us to draw up supplementary
statements, resembling our descriptions of paradigms, which show
the various formal aspects of some grammatically determined
semantic unit. Thus, we find the sememe ‘female of such-and-such
male’ expressed not only by the suffix -ess, but also by composition,
as in elephant-cow, she-elephant, nanny-goat, and by suppletion, as
in ram : ewe, boar : sow; some such pairs show inverse derivation,
the male derived from the female, as goose : gander, duck : drake.

Similarly, we should probably need a complete list to tell which
English adjectives underlie comparative forms in -er of the type
kinder, shorter, longer, and, having this list, we could recognize
semantically equivalent pairs, such as good : better, much : more,
little : less, bad : worse.

In other groups the semantic relations are not grammatically
definable. Thus, we derive a great many verbs from nouns by
means of various changes., including a zero-element, but the meanings
238of these derived verbs in relation to the underlying noun are
manifold: to man, to dog, to heard, to nose, to milk, to tree, to table> to
, to bottle, to father, to fish, to clown, and so on. Or, again, we
derive verbs from adjectives in several varieties of the meanings
‘to become so-and-so’ and ‘to make (a goal) so-and-so,’ with
various formal devices:

zero: to smoothe

zero, from comparative: to lower

zero, from quality-noun: old : to age

modification of vowel: full: to fill

suppletion (?) : dead : to kill.

prefixes: enable, embitter, refresh, assure, insure, belittle

suffix -en: brighten

suffix -en, from quality-noun: long : lengthen.

To this list we must add a large number of foreign-learned types,
such as equal : equalize, archaic : archaize, English : anglicize, simple
: simplify, vile : vilify, liquid : liquefy, valid : validate, long :
elongate, different : differentiate, debile : debilitate, public : publish.

When derivation is made by means of grammatical features,
such as phonetic modification (man : men ; mouth : to mouthe) or
modulation (convict verb : convict noun) or suppletion (go : went)
or zero-elements (cut infinitive : cut past tense; sheep singular :
sheep plural; man noun : to man verb), we may have a hard time
deciding which form of a set we had better describe as the underlying
form. In English, we get a simpler description if we take
irregular paradigms (such as man : men or run : ran) as underlying,
and regular paradigms (such as to man or a run) as derived.
In most cases this criterion is lacking; thus, we shall find it hard to
decide, in cases like play, push, jump, dance, whether to take the
noun or the verb as the underlying form. Whatever our decision,
the derivative word (e.g. to man derived from the noun man, or
a run derived from the verb to run) will often contain no affixes, and
will be described (for reasons that will shortly appear) as a secondary

In the same way, phrase-derivatives, such as old-maidish, derived
from the phrase old maid, offer no special difficulty so long as they
contain a derivational affix, such as -ish, but when the phrase is
accompanied only by a zero-feature, as in jack-in-the-pulpit or
devil-may-care, we have the difficult type of phrase-words. These
239differ from phrases in their uninterrupted and syntactically inexpansible
character, and often in their exocentric value.

14. 8. Primary words contain no free forms among their immediate
constituents. They may be complex, consisting of two
or more bound forms, as per-ceive, per-iain, de-ceive, de-tain, or they
may be simple, as boy, run, red, and, in, ouch.

The bound forms which make up complex primary words,
are determined, of course, by features of partial resemblance, as
in the examples just cited. In many languages the primary words
show a structural resemblance to secondary words. Thus, in
English, the primary words hammer, rudder, spider resemble secondary
words like dance-r, lead-er, ride-r. The part of the primary
word which resembles the derivational affix of the secondary word
(in our examples, -er) can be described as a primary affix. Thus,
the primary words hammer, rudder, spider are said to contain
a primary suffix -er. The remaining part of the primary word —
in our examples, the syllable [hɛm-] in hammer, [rod-] in rudder,
[spajd-] in spider — is called the root. The root plays the same part
in primary words as the underlying form (e.g. dance, lead, ride)
in secondary words (dancer, leader, rider).

This distinction between primary affixes and roots is justified
by the fact that the primary affixes are relatively few and vague
in meaning, while the roots are very numerous and therefore relatively
clear-cut as to denotation. 120

In accordance with this terminology, primary words that do
not contain any affix-like constituents (e.g. boy, run, red) are
classed as primary root-words. The roots which occur in primary
root-words are free roots, in contrast with bound roots which
occur only with a primary affix, such as the root [spajd-] in spider.

Primary affixes may be extremely vague in meaning and act
merely as an obligatory accompaniment (a determinative) of the
root. In English, the commonest primary suffixes do not even
tell the part of speech; thus, we have, with -er, spider, bitter,
linger, ever, under; with -le, bottle, little, hustle; with -ow, furrow,
240yellow, borrow. In other cases the meaning is more palpable; thus,
-ock, in hummock, mattock, hassock, and so on, forms nouns denoting
a lumpy object of moderate size, and this is confirmed by
its use as a secondary suffix (class-cleavage) in words like hillock,
bullock. Our foreign-learned prefixes get a vague but recognizable
meaning from contrasts like con-tain, de-tain, per-tain, re-tain.
In some languages, however, primary affixes bear relatively concrete
meanings. The Algonquian languages use primary suffixes
that denote states of matter (wood-like solid, stone-like solid,
liquid, string-like thing, round thing), tools, parts of the body,
animals, woman, child (but not, apparently, adult males). Thus,
in Menomini, the verb-form [kepa:hkwaham] ‘he puts a cover on
it,’ has a stem [kepa:hkwah-], which consists of the root [kep-]
‘obstruction of opening,’ and the primary suffixes [-a:hkw-] ‘wood
or other solid of similar consistency,’ and [-ah-] ‘act on inanimate
object by tool.’ Similarly, in Menomini, [akuapiːnam] ‘he takes
it from the water,’ the verb-stem consists of the root [akua-]
‘removal from a medium,’ and the suffixes [-epiː-] ‘liquid’ and
[-en-] ‘act on object by hand;’ [niːsunak] ‘two canoes’ is a particle
consisting of the root [niːsw-] ‘two’ and the primary suffix
[-unak] ‘canoe.’ These affixes are used also in secondary derivation.
Some of them are derived from independent words or stems;
thus, in Fox, [pjɛ:tehkwɛ:wɛ:wa] ‘he brings a woman or women’
is an intransitive verb (that is, cannot be used with a goal-object,
— much as if we could say *he woman-brings) containing the
primary suffix [-ehkwɛ:wɛ:-] ‘woman,’ which is derived from the
noun [ihkwɛ:wa] ‘woman.’ In Menomini, the cognate [-ehkiwɛ:-],
as in [piːtehkiwɛ:w] (same meaning), does not stand in this relation
to any noun, because the old noun for ‘woman’ is here obsolete,
and the actual word is [metɛ:muh] ‘woman,’ In some languages
the use of primary affixes derived from nouns covers much
the same semantic ground as our syntactic construction of verb
with goal-object. This habit is known as incorporation; the classical
instance is Nahuatl, the language of the Aztecs, where a
noun like [naka-tl] ‘meat’ is represented by a prefix in a verb-form
like [ni-naka-kwa] ‘I-meat-eat,’ that is, ‘I eat meat.’

A root may appear in only one primary word, as is the case
with most ordinary English roots, such as man, boy, cut, red, nast (in
nasty), ham- (in hammer), or it may appear in a whole series
of primary words, as is the case with many of our foreign-learned
241roots, like [-sijv] in deceive, conceive, perceive, receive. In either
case, the primary word may underlie a whole series of secondary
derivatives; thus, man underlies men, man's, men's, mannish, manly,
(to) man (mans, manned, manning); deceive underlies deceiver,
deceit, deception, deceptive; conceive underlies conceivable, conceit,
concept, conception, conceptual; perceive underlies perceiver, percept,
perceptive, perception, perceptible, perceptual; and receive
underlies receiver, receipt, reception, receptive, receptacle. Moreover,
secondary derivatives like these may exist where the primary
word is lacking; thus, we have no such primary word as *preceive,
but we have the words precept, preceptor, which are best
described as secondary derivatives of a theoretical underlying
form *pre-ceive.

The roots of a language make up its most numerous class of
morphological forms and accordingly bear its most varied and
specific meanings. This is clearest in languages which have roots
as free forms, as, in English, boy, man, cut, run, red, blue, green,
brown, white, black. The clear-cut meaning will be found also
in bound roots, such as yell- in yellow, purp- in purple, nast- in
nasty, and so on. In most languages, however, there are also
roots of very vague meaning, such as, in English, the foreign-learned
roots of the type -ceive, -tain, -fer (conceive, contain, confer,
and so on). This is particularly the case in languages whose primary
affixes are relatively varied and specific in meaning.

Once we have set up a root, we face the possibility of its modification.
This possibility is obvious when the root occurs as an
ultimate constituent in a secondary derivative: thus, in the secondary
derivative duchess the modification of the underlying word
duke- is at the same time a modification of the root duke, and in
the secondary derivatives sang, sung, song, the modifications of
the underlying sing, are necessarily modifications of the root sing.
The alternant shapes of roots are in some languages so varied
that the describer may well hesitate as to the choice of a basic
form. In ancient Greek we find the alternants [dame:-, dme:-,
dmo:-, dama-, dam-] in the forms [e-ˈdame:] ‘he tamed,’ [e-ˈdme:the:]
‘he was tamed,’ [ˈdmo:-s] ‘slave,’ [daˈma-o:] ‘I tame,’ [hipˈpodam-o-s]
‘horse-tamer.’ Our whole description of Greek morphology,
including even the distribution of derivatives into primary
and secondary types, will depend upon our initial choice of a
basic form for roots of this sort. In the Germanic languages,
242modification of the root, with or without affix-like determinatives,
occurs in words of symbolic connotation, as flap, flip, flop. If
we take flap as the basic form of this root, we shall describe flip,
flop as derivatives, formed by substitution of [i] ‘smaller, neater’
and by substitution of [ɑ] ‘larger, duller.’ Similar cases are, with
substitution of [i]: snap : snip, snatch : snitch, snuff : sniff,
bang : bing, yap : yip; of [ij]: squall : squeal, squawk : squeak,
crack : creak, gloom : gleam, tiny : teeny, of [o]: mash : mush,
flash : flush, crash : crush. At first glance, we should describe
these forms as secondary derivatives, since the word flap can be
said to underlie the words flip, flop, but it is possible that a detailed
description of English morphology would work out better
if we viewed words like flip, flop as primary modifications of
“the root flap-,” instead of deriving them from the actual
word flap.

The roots of a language are usually quite uniform in structure.
In English they are one-syllable elements, such as man, cut, red;
many of them are free forms, occurring as root-words, but many,
such as [spajd-] in spider, [hɛm-] in hammer, and, especially,
foreign-learned roots like [-sijv] in conceive, perceive, are bound
forms. Some of these bound roots end in clusters that do not occur
in word-final, as [lomb-] in lumber or [liŋg-] in linger. In Russian,
the roots are monosyllabic, with the exception of some that
have [l] or [r] between vowels of the set [e, o], as in [ˈgolod-] ‘hunger,’
[ˈgorod-] ‘city.’ We have seen an example of the variability
of a root in ancient Greek; for this language, as well as, apparently,
for Primitive Indo-European, we probably have to set up roots
of several different shapes, monosyllabic, such as [do:-] ‘give,’
and disyllabic, such as [dame:-] ‘tame.’ In North Chinese, all
the roots are monosyllabic free forms consisting, phonetically,
of an initial consonant or cluster (which may be lacking), a final
syllabic (including diphthongal types with non-syllabic [j, w, n, ŋ]),
and a pitch-scheme. The Malayan languages have two-syllable
roots, with stress on one or the other syllable, as in the Tagalog
root-words [ˈba:haj] ‘house’ and [kaˈmaj] ‘hand.’ In the Semitic
languages the roots consist of an unpronounceable skeleton of
three consonants; accordingly, every primary word adds to the
root a morphologic element which consists of a vowel-scheme.
Thus, in modern Egyptian Arabic, a root like [k-t-b] ‘write’ appears
in words like [katab] ‘he wrote,’ [ka:tib] ‘writing (person),’
243[kita:b] ‘book,’ and, with prefixes, [ma-ka:tib] ‘places for writing,
studies,’ [ma-ktab] ‘place for writing, study,’ [je-ktub] ‘he is
writing;’ similarly, the root [g-l-s] ‘sit’ appears in [galas] ‘he sat,’
[ga:lis] ‘sitting person,’ [ma-ga:lis] ‘councils,’ [ma-glas] ‘council.’

In a few languages, such as Chinese, the structure of the roots is
absolutely uniform; in others, we find some roots that are shorter
than the normal type. It is a remarkable fact that these shorter
roots belong almost always to a grammatical or a semantic sphere
which can be described, in terms of English grammar, as the sphere
of pronoun, conjunction, and preposition. In German, which has
much the same root structure as English, the definite article contains
a root [d-], for in the forms der, dem, den, and so on, the rest
of the word (-er, -em, -en, and so on) is in each case a normal inflectional
ending, appearing also in the inflectional forms of an
adjective like ‘red’: rot-er, rot-em, rot-en. The same applies to the
interrogative pronoun ‘who?’ with forms like wer, wem, wen. In
Malayan and in Semitic, many words in this semantic sphere have
only one syllable, as, in Tagalog, [at] ‘and,’ or the syntactic particles
[aŋ] ‘sign of object-expression,’ [aj] ‘sign of predication,’
[na] ‘sign of attribution.’ This semantic sphere is roughly the same
as that in which English uses atonic words.

14. 9. Perhaps in most languages, most of the roots are morphemes.
Even in cases like English sing : sang : sung : song or
flap : flip : flop, a relevant description will view one of the forms
as basic and the others as secondary derivatives or as primary
derivatives with phonetic modification of the root. In other cases,
however, we find clearly-marked phonetic-semantic resemblances
between elements which we view as different roots. The pronominal
words of English are probably best described as containing monosyllabic
roots that resemble each other, especially as to the initial

[ð-]: the, this, that, then, there, thith-er, thus.

[hw-]: what, when, where, whith-er, which, why; modified to [h]
in who, how.

[s-]: so, such.

[n-]: no, not, none, nor, nev-er, neith-er.

Complex morphologic structure of the root is much plainer in
the case of English symbolic words; in these we can distinguish,
with varying degrees of clearness, and with doubtful cases on the
244border-line, a system of initial and final root-forming morphemes,
of vague signification. It is plain that the intense, symbolic connotation
is associated with this structure. Thus, we find recurrent

[fl-] ‘moving light’: flash, flare, flame, flick-er, flimm-er.

[fl-] ‘movement in air’: fly, flap, flit (flutt-er).

[gl-] ‘unmoving light’: glow, glare, gloat, gloom (gleam, gloam-ing,
glimm-er), glint.

[sl-] ‘smoothly wet’: slime, slush, slop, slobb-er, slip, slide.

[kr-] ‘noisy impact’: crash, crack (creak), crunch.

[skr-] ‘grating impact or sound’: scratch, scrape, scream.

[sn-] ‘breath-noise’: sniff (snuff), snore, snort, snot.

[sn-] ‘quick separation or movement’: snap (snip), snatch

[sn-] ‘creep’: snake, snail, sneak, snoop.

[ǰ] ‘up-and-down movement’: jump, jounce, jig (jog, jugg-le),
jangle (jingle).

[b-] ‘dull impact’: bang, bash, bounce, biff, bump, bat.

In the same vague way, we can distinguish finals:

[-ɛš] ‘violent movement’: bash, clash, crash, dash, flash, gash,
mash, gnash, slash, splash.

[-ejr] ‘big light or noise’: blare, flare, glare, stare.

[-awns] ‘quick movement’: bounce, jounce, pounce, trounce.

[-im], mostly with determinative [-r], ‘small light or noise’:
dim, flimmer, glimmer, simmer, shimmer.

[-omp] ‘clumsy’: bump, clump, chump, dump, frump, hump,
lump, rump, stump, slump, thump.

[-ɛt], with determinative [-r], ‘particled movement’: batter, clatter,
chatter, spatter, shatter, scatter, rattle, prattle.

In this last instance we see a formal peculiarity which confirms
our classification. In English morphology there is no general
restriction to the occurrence of [-r̩] or [-l̩] as suffixes, and, in particular,
they are not ruled out by the presence of [r, l] in the body
of the word: forms like brother, rather, river, reader, reaper or little,
ladle, label are common enough. The symbolic roots, however, that
contain an [r], are never followed by the determinative suffix [-r̩],
but take an [-l̩] instead, and, conversely, a symbolic root containing
[l] is never followed by [-l̩], but only by [-r̩]: brabble and blabber
are possible as English symbolic types, but not *brabber or *blabble.245

The analysis of minute features, such as the root-forming morphemes,
is bound to be uncertain and incomplete, because a
phonetic similarity, such as, say, the [b-] in box, beat, bang, represents
a linguistic form only when it is accompanied by a semantic
similarity, and for this last, which belongs to the practical world,
we have no standard of measurement.246

Chapter 15

15. 1. Having surveyed sentence-types (Chapter 11) and constructions
(Chapters 12, 13, 14), we turn now to the third type of
meaningful grammatical arrangement, substitution (§ 10.7).

A substitute is a linguistic form or grammatical feature which,
under certain conventional circumstances, replaces any one of a
class of linguistic forms. Thus, in English, the substitute I replaces
any singular-number substantive expression, provided that this
substantive expression denotes the speaker of the utterance in
which the substitute is used.

The grammatical peculiarity of substitution consists in selective
features: the substitute replaces only forms of a certain class, which
we may call the domain of the substitute; thus, the domain of the
substitute I is the English form-class of substantive expressions.
The substitute differs from an ordinary linguistic form, such as
thing, person, object, by the fact that its domain is grammatically
definable. Whether an ordinary form, even of the most inclusive
meaning, such as thing, can be used of this or that practical situation,
is a practical question of meaning; the equivalence of a substitute,
on the other hand, is grammatically determined. For
instance, no matter whom or what we address, we may mention
this real or pretended hearer in the form of a substantive expression
by means of the substitute you — and for this we need no practical
knowledge of the person, animal, thing, or abstraction that we are
treating as a hearer.

In very many cases, substitutes are marked also by other peculiarities:
they are often short words and in many languages atonic;
they often have irregular inflection and derivation (I : me : my) and
special syntactic constructions. In many languages they appear
as bound forms and may then be characterized by morphologic
features, such as their position in structural order.

15. 2. One element in the meaning of every substitute is the
class-meaning of the form-class which serves as the domain of the
substitute. The class-meaning of the substitute you, for example,
247is the class-meaning of English substantive expressions; the class-meaning
of I is that of singular substantive expressions, and the
class-meaning of the substitutes they and we is that of plural substantive

Some substitutes add a more specific meaning which does not
appear in the form-class, but even in these cases a set of several
substitutes systematically represents the whole domain. Thus,
who and what together cover the class-meaning of English substantive
expressions. In the same way, he, she, and it together
cover the class-meaning of singular substantive expressions; within
the set, he and she cover the same sub-domain as who, and it the
same sub-domain as what, but the distinction between he and she
implies a further and independent subdivision. Our selection of
substitutes, then, divides English substantive expressions into the
sub-classes of personal (replaced by who and he-she) and non-personal
(replaced by what and it), and it subdivides the personal
singulars into the sub-classes of male (replaced by he) and female
(replaced by she).

In addition to the class-meaning, every substitute has another
element of meaning, the substitution-type, which consists of the
conventional circumstances under which the substitution is made.
Thus, I replaces any singular substantive expression (this domain
gives us the class-meaning of I), provided that this substantive
expression denotes the speaker of the very utterance in which the
I is produced: this is the substitution-type of I. The circumstances
under which a substitution is made are practical circumstances,
which the linguist, for his part, cannot accurately define. In detail,
they differ greatly in different languages; in speaking a foreign
language, we have great difficulty in using the proper substitute-forms.

15. 3. Nevertheless, it will be worth our while to leave, for a
moment, the ground of linguistics, and to examine the problems
which here confront the student of sociology or psychology. We
find, at once, that the various types of substitution represent
elementary circumstances of the act of speech-utterance. The
substitution-types in I, we, and you are based upon the speaker-hearer
relation. The types of this, here, now and that, there, then
represent relations of distance from the speaker or from the speaker
and the hearer. The interrogative type of who, what, where, when
stimulates the hearer to supply a speech-form. The negative type
248of nobody, nothing, nowhere, never excludes the possibility of a
speech-form. These types are remarkably widespread and uniform
(except for details) in the languages of the world; among them we
find the practical relations to which human beings respond more
uniformly than to any others — numerative and identificational
relations, such as positive-negative, all, some, any, same, other,
and, above all, the numbers, one, two, three, and so on. These
are the relations upon which the language of science is based;
the speech-forms which express them make up the vocabulary of
mathematics. Many of these substitution-types have to do with
species and individuals: they select or identify individuals (all,
some, any, each, every, none, and so on) out of a species. Perhaps
every language has a form-class of object-expressions, with a
class-meaning of the type ‘species occurring in individual specimens.’
Accordingly, the substitutes for object-expressions, pronominals,
will usually show the most varied substitution-types.
In English, where object-expressions are a special part of speech,
the noun, the substitutes for the noun make up a part of speech,
the pronoun; together, these two constitute a greater part of
speech, the substantive. The pronouns differ from nouns, for one
thing, in not being accompanied by adjective modifiers (§ 12.14).

To a large extent, some substitution-types are characterized,
further, by the circumstance that the form for which substitution
is made, has occurred in recent speech. Thus, when we say
Ask that policeman, and he will tell you, the substitute he means,
among other things, that the singular male substantive expression
which is replaced by he, has been recently uttered. A substitute
which implies this, is an anaphoric or dependent substitute, and
the recently-uttered replaced form is the antecedent. This distinction,
however, seems nowhere to be fully carried out: we
usually find some independent uses of substitutes that are ordinarily
dependent, as, for instance, the independent use of it in it's raining.
Independent substitutes have no antecedent: they tell the
form-class, and they may even have an elaborate identificational
or numerative substitution-type — as, for instance, somebody, nobody
— but they do not tell which form of the class (for instance,
which particular noun) has been replaced.

On the whole, then, substitution-types consist of elementary
features of the situation in which speech is uttered. These features
are so simple that, for the most part, they could be indicated
249by gestures: I, you, this, that, none, one, two, all, and so on. Especially
the substitutes of the ‘this’ and ‘that’ types resemble
interjections in their semantic closeness to non-linguistic forms
of response; like interjections, they occasionally deviate from the
phonetic pattern of their language (§ 9.7). Since, aside from the
class-meaning, the substitution-type represents the whole meaning
of a substitute, we can safely say that the meanings of substitutes
are, on the one hand, more inclusive and abstract, and,
on the other hand, simpler and more constant, than the meanings
of ordinary linguistic forms. In their class-meaning, substitutes
are one step farther removed than ordinary forms from practical
reality, since they designate not real objects but grammatical
form-classes; substitutes are, so to speak, linguistic forms of the
second degree. In their substitution-type, on the other hand, substitutes
are more primitive than ordinary linguistic forms, for
they designate simple features of the immediate situation in
which the speech is being uttered.

The practical usefulness of substitution is easy to see. The substitute
is used more often than any one of the forms in its domain;
consequently, it is easier to speak and to recognize. Moreover,
substitutes are often short forms and often, as in English, atonic,
or, as in French, otherwise adapted to quick and easy utterance.
In spite of this economy, substitutes often work more safely and
accurately than specific forms. In answer to the question Would
you like some fine
, fresh cantaloupes? The answer How much are
is perhaps more likely to be followed by a delay or
aberration of response (“misunderstanding”) than the answer
How much are they? This is especially true of certain substitutes,
such as I, whose meaning is unmistakable, while the actual mention
of the speaker's name would mean nothing to many a

15. 4. Returning to the ground of linguistics, we may be somewhat
bolder, in view of what we have seen in our practical excursion,
about stating the meanings of substitutes. We observe,
also, that in many languages, the meanings of substitutes recur
in other forms, such as the English limiting adjectives (§ 12.14).

The meaning of the substitute you may be stated thus:

A. Class-meaning: the same as that of the form-class of substantive
expressions, say ‘object or objects;’

B. Substitution-type: ‘the hearer.’250

The meaning of the substitute he may be stated thus:

A. Class-meanings:

1. Definable in terms of form-classes:

(a) the same as that of the form-class of singular substantive
expressions, say ‘one object;’

(b) the same as that of the form-class defined by the
substitutes who, someone, say ‘personal;’

2. Creating an otherwise unestablished form-class: he is used
only of certain singular personal objects (the rest are replaced,
instead, by she), which, accordingly, constitute
a sub-class with a class-meaning, say ‘male;’

B. Substitution-types:

1. Anaphora: he implies, in nearly all its uses, that a
substantive designating a species of male personal objects
has recently been uttered and that he means one
individual of this species; say ‘recently mentioned;’

2. Limitation: he implies that the individual is identifiable
from among all the individuals of the species mentioned;
this element of meaning is the same as that of the syntactic
category of definite nouns (§ 12.14) and can be
stated, say, as ‘identified.’

15. 5. Substitutes whose substitution-type consists of nothing
but anaphora, are (simple) anaphoric substitutes: apart from
their class-meanings (which differ, of course, according to the
grammatical form-classes of different languages), they say only
that the particular form which is being replaced (the antecedent)
has just been mentioned. In English, finite verb expressions are
anaphorically replaced by forms of do, does, did, as in Bill will
misbehave just as John did
. The antecedent here is misbehave; accordingly,
the replaced form is misbehaved. A few English verb-paradigms,
such as be, have, will, shall, can, may, must, lie outside
the domain of this substitution: Bill will be bad just as John was
(not did). Nouns, in English are anaphorically replaced by one,
plural ones, provided they are accompanied by an adjective attribute:
I prefer a hard pencil to a soft one, hard pencils to soft ones.
This use of one as an anaphoric pronoun differs by class-cleavage
from the several attributive uses of the word one (§ 12.14), especially
in forming a plural, ones. The details of this anaphoric substitution
will concern us later (§ 15.8-10).

In subordinate clauses introduced by as or than, we have in English
251a second kind of anaphora for a finite verb expression: we say
not only Mary dances better than Jane does, but also Mary dances
better than Jane
. We can describe this latter type by saying that
(after as and than) an actor (Jane) serves as an anaphoric substitute
for an actor-action expression (Jane dances), or we can say
that (after as and than) a zero-feature serves as an anaphoric substitute
for a finite verb expression accompanying an actor expression.
Another case of an anaphoric zero-feature in English is the replacement
of infinitive expressions after the preposition to (as in I
haven't seen it, but hope to
) and after the finite verbs which take
an infinitive attribute without to (as in I'll come if I can). Similarly,
we have zero-anaphora for participles after forms of be
and have, as in You were running faster than I was; I haven't seen it,
but Bill has
. Zero-anaphora for nouns with an accompanying
adjective occurs freely in English only for mass nouns, as in I like
sour milk better than fresh
. For other nouns we use the anaphoric
one, ones, except after certain limiting adjectives.

While some forms of simple anaphoric substitution seem to
occur in every language, there are great differences of detail. The
use of one, ones, is peculiar to English; related languages of similar
structure use zero-anaphora quite freely for nouns after adjectives,
as, German grosze Hunde und kleine [ˈgro:se ˈhunde unt ˈklajne]
‘big dogs and little ones;’ French des grandes pommes et des petites
[de grand pɔm e de ptit] ‘big apples and small ones.’ In some languages
the subject in the full sentence-types can be replaced by
zero-anaphora; thus, in Chinese, to a statement like [wo3 ˈjuŋ4 i2
khwaj ˈpu4] ‘I need one piece (of) cloth,’ the response may be
[ˈjuŋ4 i4 ˈphi1 mo?] ‘Need one roll (interrogative particle)?’ In
Tagalog this happens in subordinate clauses, as in the sentence
[aŋ ˈpuˈnuʔ aj tuˈmuːˈbuʔ haŋˈgaŋ sa magˈbuːŋa] ‘the tree (predicative
particle) grew until (attributive particle) bore-fruit.’

15. 6. Perhaps all languages use pronominal substitutes which
combine anaphora with definite identification: the replaced form
is an identified specimen of the species named by the antecedent.
This, we have seen, is the value of the English pronoun he, as in
Ask a policeman, and he will tell you. Substitutes of this kind are
often, but misleadingly, called “anaphoric;” a better name would
be definite. In most languages, including English, the definite
substitutes are not used when the antecedent is the speaker or the
hearer or includes these persons; for this reason, the definite
252substitutes are often spoken of as third-person substitutes. They
usually share various peculiarities with the substitutes that refer
to the hearer and to the speaker.

The English definite or third-person pronouns, he, she, it, they,
differ for singular and plural replaced forms, and, in the singular,
for personal arid non-personal antecedents: personal he, she, versus
non-personal it. We have seen that the difference of singular and
plural is otherwise also recognized by the language (as, for instance,
in the inflection of nouns: boy, boys), and we shall see that the same
is true of the difference of personal and non-personal. Within the
personal class, however, the distinction between he used with a
male antecedent, and she, with a female antecedent, is otherwise
imperfectly recognized in our language (as, in the use of the suffix
-ess, § 14.7). The distinction, then, between the pronoun-forms
he and she, creates a classification of our personal nouns into male
(defined as those for which the definite substitute is he) and female
(similarly defined by the use of the substitute she). Semantically,
this classification agrees fairly well with the zoological division
into sexes.

In languages with noun-genders (§ 12.7), the third-person pronouns
usually differ according to the gender of the antecedent.
Thus, in German, masculine nouns, such as der Mann [der ˈman]
‘the man,’ der Hut [huːt] ‘the hat,’ have the third-person substitute
er [e:r], as when er ist grosz [e:r ist ˈgro:s] ‘he, it is big,’ is
said of either a man or a hat, or of any other antecedent that
belongs to the “masculine” congruence-class;

feminine nouns, such as die Frau [diː ˈfraw] ‘the woman,’ die
[uːr] ‘the clock,’ have the third-person substitute sie [ziː],
as in sie ist grosz, ‘she, it is big;’

neuter nouns, such as das Haus [das ˈhaws] ‘the house,’ or das
Weib [vajp] ‘the woman,’ have the third-person substitute es
[es], as in es ist grosz.

This distinction, unlike that of he and she in English, accords
with a distinction in the form of noun-modifiers (such as der : die :
das ‘the’).

The meaning of definite identification — that is, the way in
which the individual specimen is identified from among the species
named by the antecedent — varies for different languages and
would probably be very hard to define. It is important to notice,
however, that in languages which have a category of “definite”
253noun-modifiers (such as, in English the, this, that, my, John's,
etc., § 12.14), the definite pronoun identifies the individual in the
same fashion as a definite modifier identifies its head noun; thus,
a he after the antecedent policeman is equivalent in denotation,
except for the peculiar value that lies in the use of a substitute, to
the phrase the policeman. We need mention only a few widespread
peculiarities, such as the case, not very common in English, that
the definite pronoun is spoken before its antecedent: He is foolish
who says so
. If the antecedent is a predicate complement after a
form of the verb to be, the definite pronoun is normally it, regardless
of number, personality, or sex: it was a two-storey house; it's he;
it's me (I), it's the boys. Instead of an infinitive phrase as an actor
(to scold the boys was foolish), we more commonly use it, with the
infinitive phrase following in close parataxis (§ 12.2): it was foolish
to scold the boys
. An actor-action phrase, such as you can't come,
does not serve as an actor; but does appear in close parataxis with
it as an actor: it's too bad you can't come. This anticipatory use of
the definite pronoun extends, in German, to almost any actor, with
the restriction that the pronoun comes first; thus, beside ein Mann
kam in den Garten
[ajn ˈman ˈka:m in den ˈgarten] ‘a man came
into the garden,’ there is the form es kam ein Mann in den Garten,
where the use of es resembles the English use of the adverb there.
If the noun in parataxis is plural, this German es accompanies a
plural verb: beside zwei Manner kamen in den Garten [tsvaj ˈmener
ˈka:men] ‘two men came into the garden,’ there is the form es
kamen zwei Manner in den Garten.

In French, the definite pronoun replaces an adjective: êtes-vous
je le suis [ɛ:t vu œrø? — žə l sɥi.] ‘Are you happy?
— I am.’ A step beyond this, we find definite pronouns in marginal
uses without any antecedent, as in English slang beat it ‘run away,’
cheese it ‘look out,’ he hot-footed it home ‘he ran home,’ let 'er go.
We use they as an actor for people in general: they say Smith is
doing very well
. The commonest use of this sort is the pseudo-impersonal
use of a definite pronoun as a merely formal actor, in
languages that have a favorite actor-action construction: it's raining;
it's a shame. This may occur alongside a genuine impersonal
construction (§ 11.2). Thus, in German, beside the genuine impersonal
mir war kalt [miːr va:r ˈkalt] ‘to-me was cold; I felt
cold,’ hier wird getanzt [ˈhiːr virt geˈtantst] ‘here gets danced;
there is dancing here,’ the definite pronoun es may appear as an
254actor, provided it comes first in the phrase: es war mir halt; es wird
hier getanzt
. In Finnish, the impersonal and the pseudo-impersonal
are used for different meanings: puhutaan ‘there is talking’ is a
genuine impersonal, but sadaa ‘it's raining’ contains a definite substitute
actor ‘he, she, it,’ just as does puhuu ‘he, she, it is talking.’

15. 7. The definite substitutes in most languages are not used
when the replaced form designates the speaker or the hearer or
groups that include these persons; in this case a different type, the
personal substitute is used. The first-person substitute I replaces
mention of the speaker, and the second-person substitute thou,
of the hearer. These are independent substitutes, requiring no
antecedent utterance of the replaced form.

In addition to the I and thou substitutes, most languages use
also forms for groups of people that include the speaker or the
hearer or both. Thus, in English, for a group of people which
includes the speaker, the substitute is we; if the speaker is not
included, but the hearer is, the substitute is ye. Many languages
distinguish all three of these possibilities, as, Tagalog, which, beside
[aˈku] ‘I’ and [iˈkaw] ‘thou,’ has the plural-like forms:

speaker only included (exclusive first person plural): [kaˈmi] ‘we’

speaker and hearer included (inclusive first person plural):
[ˈta:ju] ‘we’

hearer only included (second person plural): [kaˈju] ‘ye.’

Similarly, languages which distinguish a dual number, allow of
five combinations, as in Samoan: ‘I-and-he,’ ‘I-and-thou,’ ‘ye-two,’
‘I-and-they,’ ‘I-and-thou-and-he (-or-they),’ ‘thou-and-they.’ A
few languages distinguish also a trial number (‘three persons’) in
their personal pronouns.

The English forms thou, ye are, of course, archaic; modern
English is peculiar in using the same form, you, both for the hearer
and for a group of persons that includes the hearer.

Many languages use different second-person substitutes according
to different social relations between speaker and hearer.
Thus, French uses vous [vu] ‘you’ much like English, for both
singular and plural, but if the hearer is a near relative, an intimate
friend, a young child, or a non-human being (such as a god), there
is a special intimate singular-form toi [twa]. German uses the third-person
plural pronoun ‘they’ for both singular and plural second
person: Sie spaszen [ziː ˈšpa:sen] is both ‘they are jesting’ and ‘you
(singular or plural) are jesting,’ but the intimate forms, used much
255like those of French, distinguish singular and plural: du spaszest
[duː špa:sest] ‘thou art jesting,’ ihr spaszt [iːr špa:st] ‘ye are jesting.’

The meaning of second-person substitutes is limited in some
languages by the circumstance that they are not used in deferential
speech; instead, the hearer is designated by some honorific
term (your Honor, your Excellency, your Majesty). In Swedish or
in Polish, one says, for instance, ‘How is Mother feeling?’ or ‘Will
the gentleman come to-morrow?’ where the terms here italicized
denote the hearer. Some languages, such as Japanese and Malay,
distinguish several substitutes for both first and second persons,
according to deferential relations between speaker and hearer.

The personal substitutes and the definite (“third-person”) substitutes
in many languages group themselves, by virtue of common
features, into a kind of closed system of personal-definite substitutes.
In English, both sets he, she, it, they and I, we, you (thou, ye), are
atonic in the phrase; most of them have a special accusative case-form
(me, us, him, her, them, thee); most of them derive their possessive
adjectives irregularly (my, our, your, his, her, their, thy),
and some of these adjectives have a special form for zero anaphora
(mine, etc., § 15.5). In French, the personal-definite pronouns
have special (conjunct) forms when they serve as actors or
goals of verbs (§ 12.12); these have case-inflection for different
positions, which is otherwise foreign to French substantives; moreover,
they underlie possessive adjectives, as moi [mwa] ‘I,’ mon
[mɔn šapo] ‘my hat,’ while other substantives do not:
le chapeau de Jean [lə šapo d žɑn] ‘the hat of John; John's hat.’
Very commonly the personal-definite substitutes have special
syntactic constructions. Thus, in English, German, and French,
the finite verb has special congruence-forms for different persons
as actors: I am : thou art: he is; French nous savons [nu savɔn] ‘we
know,’ vous savez [vu save] ‘you know,’ elles savent [ɛl sa:v] ‘they
(feminine) know,’ ils savent [i sa:v] ‘they know.’

The personal-definite pronouns may even have a fairly systematic
structure. Thus, in the Algonquian languages, an initial
element [ke-] appears in the forms that include the hearer; if
the hearer is not included, [ne-] denotes the speaker; if neither
is included, the initial is [we-], as, in Menominiː

tableau ‘thou’ | ‘we’ (inclusive) | ‘ye’ | ‘I’ | ‘we’ (exclusive) | ‘he’ | ‘they’256

Samoan, with a distinction of dual and plural numbers, has:

tableau ‘I’ | ‘we two’ (excl.) | ‘we’ (excl.) | ‘we two’ (incl.) | ‘we’ (incl.) | ‘thou’ | ‘ye two’ | ‘ye’ | ‘he’ | ‘they two’ | ‘they’

The dual-trial-plural distinction appears in the language of
Annatom Island (Melanesian):

[ainjak] ‘I,’ [aijumrau] ‘we two’ (excl.), [aijumtai] ‘we three’
(excl.), [aijama] ‘we’ (excl.),

[akaijau] ‘we two’ (incl.), [akataij] ‘we three’ (incl.), [akaija]
‘we’ (incl.),

[aiek] ‘thou,’ [aijaurau] ‘ye two,’ [aijautaij] ‘ye three,’ [aijaua]

[aien] ‘he,’ [arau] ‘they two,’ [ahtaij] ‘they three,’ [ara] ‘they.’

In many languages, personal-definite substitutes appear as
bound forms. Thus, Latin had definite-personal actors or goals
in the finite verb-forms:

amō ‘I love,’ amās ‘thou lovest,’ amat ‘he (she, it) loves,’
amāmus ‘we love,’ amātis ‘ye love,’ amant ‘they love,’

amor ‘I am loved,’ amāris ‘thou art loved,’ amātur ‘he (she, it)
is loved,’ amāmur ‘we are loved,’ amāminī ‘ye are loved,’ amantur
‘they are loved.’

Some languages, in the same way, include both actor and
goal, as Cree: [nisa:kiha:w] ‘I love him,’ [nisa:kiha:wak] ‘I love
them,’ [kisa:kiha:w] ‘thou lovest him,’ [nisa:kihik] ‘he loves me,’
[nisa:kihikuna:n] ‘he loves us (excl.),’ [kisa:kihitina:n] ‘we love
thee,’ [kisa:kihitin] ‘I love thee,’ and so on, through a large

Likewise, in Cree, the possessor of an object appears in a bound
form: [nitastutin] ‘my hat,’ [kitastutin] ‘thy hat,’ [utastutin]
‘his hat,’ and so on. In all these cases, the third-person bound form
may stand in cross-reference with a noun antecedent: Latin pater
‘father he-loves; the father loves’ (§ 12.9).

The personal-definite system may be elaborated by distinctions
of identity and non-identity, such as the difference of me and myself,
where the latter form implies identity with the actor (I
washed myself
, § 12.8), or the Scandinavian hans ‘his’ and sin
‘his (own).’ These differences appear also in bound forms, as in
the obviative forms of Algonquian (§ 12.8); similarly, ancient
Greek, beside an ordinary bound actor, as in [ˈelowse] ‘he washed,’
257had a middle-voice form, where the actor is at the same time affected
by the action: [eˈlowsato] ‘he washed himself’ or ‘he washed for

Other specializations are less common; thus, Cree, beside a
verb with actor and goal, such as [ninituma:w] ‘I ask for him,
call him,’ [ninitute:n] ‘I ask for it,’ and a form with actor and two
goals, [ninitutamawa:w] ‘I ask him for it,’ has also a form with
actor, goal, and interested person [ninitutamwa:n] ‘I ask for it
with reference to him,’ that is, ‘for his use’ or ‘at his behest.’

15. 8. Demonstrative or deictic substitution-types are based on
relative nearness to the speaker or hearer. In English we have two
such types, for nearer and for farther away; they coincide with
the values of the limiting adjectives this and that (§ 12.14). Demonstrative
substitutes may be dependent (that is, they may refer
anaphorically to an antecedent speech-form that names the
species), or independent. In either case, however, they identify
the individual object within the (named or unnamed) species.
Demonstrative pronoun substitution, in English, is made by the
pronouns this, (these), that (those), which differ, by class-cleavage,
from the limiting adjectives, or by phrases consisting of these
limiting adjectives plus the anaphoric one (§ 15.5). These forms
are not ordinarily used to replace personal nouns — for the anticipatory
use in This is my brother; these are my brothers cannot
be viewed as personal. The dependent substitutes in the singular
are this one, that one, and the independent this, that; hence we have
the distinction between, say, of these books, I like this one better
than that one
, but, of unnamed objects, I like this better than that.
In the plural, however, these and those are in either case used without
the anaphoric ones.

In French we can see a more differentiated system. There are
three types of demonstrative limitation and substitution: a general
type from which two special types are differentiated by the
addition of the adverbs ci [si] for nearer position and [la] for
farther away. The forms of the limiting adjective, the dependent
pronoun, and the independent pronoun, are distinct:

tableau adjective | dependent pronoun | independent pronoun | singular | masculine | feminine258

tableau adjective | dependent pronoun | independent pronoun | plural | masculine | feminine

Thus: cette plume-ci [sɛt plym si] ‘this pen,’ de ces deux plumes,
je préfère celle-ci à celle-là
[də se dø plym, žə prefɛ:r sɛl si a sɛl la]
‘of these two pens, I prefer this one to that one;’ but, of unnamed
things, je préfère ceci à cela [sə si a sə la] ‘I prefer this to that.’
The pronouns without ci and are confined to certain constructions:
de ces deux plumes, je préfère celle que vous avez [sɛl kə vuz ave]
‘of these two pens, I prefer the one you have;’ independent: c'est
[s ɛt ase] ‘that's enough.’

Demonstrative substitution-types are not always fully distinct
from definite, and, similarly, demonstrative limiting modifiers
may merge with mere definite markers of the type ‘the.’ In
German, more than one dialect has only a single paradigm whose
forms are used proclitically as a definite article, der Mann [der
ˈman] ‘the man,’ and with accent as a demonstrative limiting
adjective, der Mann [ˈde:r ˈman] ‘that man,’ and as a pronoun,
der [ˈde:r] ‘that one.’ This last use, in German, is but slightly
distinguished from that of the definite pronoun er [e:r] ‘he;’ the
chief difference, perhaps, is the use of der (not er) in the second of
two paratactic full sentences: es war einmal sin Mann, der hatte
drei Sohne
[es ˈva:r ajnˌma:l ajn ˈman, de:r ˌhate ˌdraj ˈzø:ne]
‘there was once a man, he (literally, ‘that-one’) had three sons.’

Many languages distinguish more types of demonstrative substitution;
thus, some English dialects add yon, for things farthest
away, to the distinction of this and that. Latin had hic for things
nearest the speaker, iste for those nearest the hearer; and ille
for those farthest away. The Kwakiutl language makes the same
distinctions, but doubles the number by distinguishing also between
‘in sight’ and ‘out of sight.’ Cree has [awa] ‘this,’ [ana] ‘that,’
and [o:ja] ‘that recently present but now out of sight.’ Eskimo
has a whole series: [manna] ‘this one,’ [anna] ‘that one in the north,’
[qanna] ‘that one in the south,’ [panna] ‘that one in the east,’
[kanna] ‘that one down there,’ [sanna] ‘that one down in the sea,’
[iŋŋa] ‘that one,’ and so on.

Outside of pronouns, we have the adverbial forms here : there,
hither : thither, hence : thence, now : then; the th-forms, however,
259merge with simple anaphoric use, as in Going to the circus? I'm
going there too
. Similarly, so (and archaically also thus) is both
demonstrative and, more usually, anaphoric (I hope to do so).
Forms like (do it) this way, this sort (of thing), this kind (of thing)
are on the border between substitutes and ordinary linguistic

15. 9. Interrogative substitutes prompt the hearer to supply
either the species or the identification of the individual; in English,
accordingly, interrogative substitutes occur only in supplement-questions.
Of pronouns, we have the independent who? (accusative
whom?) for personals and what? for non-personals; these ask for
both species and individual. For non-personals only we have also
the independent which? asking for identification of the individual
object from a limited field, but not for the species. The dependent
substitutes, asking for the identification of the individual from a
limited field, are which one? which ones?

Outside the pronouns, we have the interrogative substitutes
where? whither? whence? when? how? why? Interrogative verb-substitutes
occur in some languages, as in Menomini [wɛʔse:kew¿]
‘what sort is he?’

The limitation of interrogative forms to certain syntactic positions
is quite common. Frequently we find them restricted to
positions in the predicate of a binary sentence-type. The word-order
and the plural verb-form in who are they? what are those
are features of this kind. In present-day French, the non-personal
quoi? [kwa¿] ‘what?’ is scarcely ever used as actor or goal,
but instead, figures as a predicate complement, appearing in the
conjunct form que [kə], as in qu'est-ce que c'est? [k ɛ s kə s ɛ¿] ‘what
is it that this is? what's this?’ and qu'est-ce qu'il a vu? [k ɛ s k il
a vy¿] ‘what is it that he has seen? what did he see?’ In some
languages the interrogative substitutes are always predicates of
equational sentences, as, in Tagalog, [ˈsiːnu aŋ nagbiˈgaj sa iˈju¿]
‘who the one-who-gave to you? who gave it to you?’ or, in Menomini
[awɛ:ʔ pɛ:muhnɛt¿] ‘who the-one-walking-by? who is walking

15. 10. The various possibilities of selecting individual objects
from a species are represented by all manner of substitute-forms,
especially of pronouns. In English, nearly all forms of this sort
consist of limiting adjectives with the anaphoric one, ones (§ 15.5)
or of substantive uses, by class-cleavage, of the same words. There
260are many distinctions, not always rigidly carried out, between
dependent and independent substitution, and in the latter, between
personal and non-personal classes. The various limiting adjectives
differ in treatment; these differences add another line of classification
among them (§ 12.14).

(1) Some limiting adjectives are, like ordinary adjectives, followed
by one, ones to form anaphoric substitutes. We have seen
that this is the case of the singular this, that and, under certain
conditions, of which? what? It is true also of each, every, whatever,
whichever, and of the phrasal expressions many a, such a, what a.
Thus, we say he was pleased with the children and gave each one a
. As independent substitutes we use this, that, which, what,
whichever, whatever of non-personals only; corresponding to every,
we have personal everybody, everyone and non-personal everything;
each has no independent form.

(2) We have both simple pronoun use or combination with the
anaphoric ones, one, in the case of either, former, latter, last, neither,
other, such, and the ordinals, first, second, etc. The variants differ
chiefly in connotation. Thus, we say Here are the books; take either
. The word other forms a special sub-class, in that it has a
plural form, others: You keep this book and I'll take the others (the
other ones
). In independent use these words serve chiefly as nonpersonals.

(3) The remaining limiting adjectives are peculiar in not taking
the anaphoric one, ones. Thus, we say: Here are the books; take one
(two, three, any, both, all, a few, some, and so on)
. The independent
substitutes show great variety. Thus, all is used as a non-personal:
All is not lost; That's all. On the other hand, one, as an atonic, is
personal: One hardly knows what to say. Several form compounds
for independent use, such as the personal somebody, someone, anybody,
anyone and the non-personal something, anything.

(4) Several limiting adjectives show an eccentric treatment.
The article the with the anaphoric one, ones forms a dependent
substitute, provided some other modifier follows: the one(s) on the
; otherwise it does not appear in pronominal use, and the
definite pronoun serves instead. The article a in combination
with another adjective does not influence the treatment of the
latter: many a one; another (one). Otherwise, the article a is
accompanied by the anaphoric one only in the emphatic form
not a one. All other pronominal uses show us one replacing a: to
261take an apple there corresponds the pronominal take one. The determiner
no is paralleled by the dependent substitute none, but ordinarily
we use instead the combination of not with any (I didn't see
); the independent substitutes are the compounds nobody, no
, nothing (archaic naught).

Among these substitution-types, the negative is, of course,
represented in all languages, and often shows special peculiarities;
to it belong also the non-pronominal nowhere, never, and sub-standard
nohow. In many languages, as in most forms of sub-standard
English, these substitutes are accompanied by the general negative
adverb: I can't see nothing. The numerative types (all, one, two,
three, and so on) seem also to be universal. As to the selective
types, however, there is great room for variety; other languages
have substitution-types that are not exactly matched in English.
Thus, Russian [ˈne-xto] ‘someone’ implies that the speaker can
(but does not) identify the individual (‘someone told me the other
day that…’), while [xto-ni-ˈbut] does not imply this ability
(‘there's someone at the door’). Still another type, [ˈkoj-xto]
implies that a different individual is selected on different occasions
(‘now and then someone tries’).

15. 11. Substitutes frequently are tied up with special syntactic
functions; thus, we have seen that interrogative substitutes in
English and many other languages are confined to certain positions
in the sentence. Some languages have special pronouns for predicative
use. Thus, in Menomini, beside such forms as [nenah]
‘I,’ [enuh] ‘that one’ (animate), [eneh] ‘that’ (inanimate), there
are parallel forms which occur only as predicates; the normal substitute
appears in [kɛhke:nam eneh] ‘he-knows-it that (thing);
he knows that,’ but the predicative form in [eneʔ kɛ:hkenah]
‘that (thing) that-which-he-knows; that is what he knows,’ or in
[enuʔ kɛ:hkenah] ‘that (person) the-one-who-knows-it; that one
is the one who knows it.’ These predicative forms vary inflectionally
for the same categories as a verb, such as interrogative [enet
kɛ:hkenah?] ‘is it that which he knows? is that the thing he
knows?’ or surprised present [enesa? kɛ:hkenah!] ‘and so that is
what he knows!’ and so on.

Our relative substitutes belong to a fairly widespread, but by no
means universal type: the substitute indicates that the phrase in
which it figures is an included (or completive) form. In English,
the phrase has the favorite full-sentence structure (actor-action
262construction), and is marked by the relative substitute as not constituting
a full sentence. Our relatives who (whom), which, where,
when, that differ from other substitutes by class-cleavage. They,
or their immediate phrase, come first in the clause. We have,
firstly, the anaphoric type, that, and personal who, non-personal
which: the boy who (that) ran away, the book which (that) he read;
the house in which we lived. If the relative substitute fills in its
clause the position of verbal goal, prepositional axis, or predicate
complement, we have here also a zero-substitute: the man I saw,
the house we lived in, the hero he was. In ordinary speech, English
relative clauses identify the individual antecedent; in more formal
style we have also non-identifying relative clauses with paratactic
sentence-modulation: the man, who was carrying a big bag, came up
to the gate

In languages with case-forms, the inflection of the relative pronoun
is normally determined by the forms in its clause: I saw the
boy who ran away
; the boy whom I saw ran away. In Latin, a normal
form would be in hāc vītā quam nunc ego dēgō ‘in this life which I
now lead,’ where the antecedent, vītā happens to be in the ablative
case (as axis of the preposition in), and the relative pronoun, quam
‘which,’ in the accusative case, as goal of the verb dēgō. However,
languages with complicated inflection now and then show attraction
of the relative pronoun into an inflectional form that belongs
properly to the antecedent: the Latin form vītā in hāc quā nunc
ego dēgō
, with the same denotation as the above normal form, has
the relative pronoun quā in the ablative case, concording with the
antecedent, instead of the accusative case demanded by its position
in the clause.

Independent relative substitutes, having no antecedent, allow
the clause to replace an indication of species: take what(ever) you
; ask whom(ever) you like; whoever says so is mistaken. In
English such clauses are used also as paratactic modifiers of a full
sentence: whatever he says, I don't believe him. The same difference
between dependent and independent use appears in our adverbial
substitutes: dependent the time (when) he did it; the house where we
; independent we'll see him when he gets here; we visit them
whenever we can; we take them where
(ver) we find them.263

Chapter 16
Form-classes and Lexicon

16. 1. The meaningful features of linguistic signaling are of two
kinds: lexical forms, which consist of phonemes, and grammatical
forms, which consist of tagmemes (features of arrangement,
§ 10.5). If we extend the term lexical to cover all forms that can be
stated in terms of phonemes, including even such forms as already
contain some grammatical features (e.g. poor John or duchess or
ran), then the parallelism of lexical and grammatical features can
be exhibited in a set of terms like the following:

(1) Smallest and meaningless unit of linguistic signaling:

(a) lexical: phoneme;

(b) grammatical: taxeme;

(2) Smallest meaningful unit of linguistic signaling: glosseme;
the meaning of a glosseme is a noeme;

(a) lexical: morpheme; the meaning of a morpheme is a

(b) grammatical: tagmeme; the meaning of a tagmeme is an

(3) Meaningful unit of linguistic signaling, smallest or complex:
linguistic form; the meaning of a linguistic form is a linguistic

(a) lexical: lexical form; the meaning of a lexical form is a
lexical meaning;

(b) grammatical: grammatical form; the meaning of a grammatical
form is a grammatical meaning.

Every lexical form is connected in two directions with grammatical
forms. On the one side, the lexical form, even when taken
by itself, in the abstract, exhibits a meaningful grammatical
structure. If it is a complex form, it shows some morphologic or
syntactic construction (duchess, poor John), and if it is a morpheme,
it may still exhibit morphologic features (a modified morpheme,
e.g. men or ran, § 13.7); in an unmodified morpheme (man, run)
we may view the absence of grammatical construction as a positive
264characteristic. On the other side, the lexical form in any actual,
utterance, as a concrete linguistic form, is always accompanied by
some grammatical form: it appears in some function, and these
privileges of occurrence make up, collectively, the grammatical
function of the lexical form. The lexical form appears in certain
sentence-types or, if it is a bound form, in none at all; it appears in
certain positions of certain constructions or, if it is an interjection,
in few or none; it appears as replaced form in certain substitutions,
or, if it be a substitute, as substitute in certain substitutions. The
functions of lexical forms are created by the taxemes of selection
which help to make up grammatical forms. Lexical forms which
have any function in common, belong to a common form-class.

The functions of lexical forms appear as a very complex system.
Some functions are common to a great number of forms and define
a large form-class; for instance, the functions which define the
English form-class of substantive expressions (serving in the
sentence-type of call, filling the positions of actor with a verb, of
goal with a verb, of axis with a preposition; underlying a possessive
adjective, and so on), are common to an almost unlimited number
of words and phrases. Different functions may create overlapping
form-classes; thus, the function of filling the actor position is
common to substantive expressions and to marked infinitive
phrases (to scold the boys would be foolish). Other functions may be
limited to a very few lexical forms or to only a single one; thus,
phrases with the noun way as center seem to be the only substantive
expressions which function as adverbs of manner, with the interrogative
substitute how? (this way, the way I do, and so on).

Particular lexical forms may, by class-cleavage (§ 12.14) exhibit
unusual combinations of function. Thus, egg is in English a
bounded noun, (the egg, an egg) but occurs also as a mass noun (he
spilled egg on his necktie
). Salt is a mass noun and accordingly
underlies a plural only in the specialized meaning ‘kinds of,’ but,
by class-cleavage, there is also a plural salts (as in Epsom salts)
with the meaning ‘consisting of particles,’ in a class with oats,
grits, and the like. Man is a (bounded, personal) male noun (a
man, the man,… he), but by class-cleavage is treated also as a
proper noun, parallel in this with God, as in man wants but little,
man is a mammal. The word one by a complicated class-cleavage
belongs to five form-classes: as a determiner (§ 12.14) it fulfils
the requirement that bounded singular nouns be preceded by a
265modifier of this class (one house, one mile); as an ordinary numerative
it occurs with the definite determiners (the one man, this one
, my one friend); it replaces a with anaphora of the noun
(§ 15.10) when no other modifier is present (Here are some apples;
take one); it occurs as an independent pronoun for ‘any person in
general’ and in this use is always atonic and underlies the derivatives
one's and oneself (one can't help oneself); finally, it is the
anaphoric substitute for nouns after an adjective, and in this use
forms a plural, ones (the big box and the small one, these boxes and the
ones in the kitchen
, § 15.5).

16. 2. The grammar of a language includes, then, a very complex
set of habits (taxemes of selection) by which every lexical form is
used only in certain conventional functions; every lexical form is
assigned always to the customary form-classes. To describe the
grammar of a language, we have to state the form-classes of each
lexical form, and to determine what characteristics make the
speakers assign it to these form-classes.

The traditional answer to this question appears in our school
grammars, which try to define the form-classes by the class-meaning
— by the feature of meaning that is common to all the
lexical forms in the form-class. The school grammar tells us, for
instance, that a noun is “the name of a person, place, or thing.”
This definition presupposes more philosophical and scientific knowledge
than the human race can command, and implies, further, that
the form-classes of a language agree with the classifications that
would be made by a philosopher or scientist. Is fire, for instance,
a thing? For over a century physicists have believed it to be an
action or process rather than a thing: under this view, the verb
burn is more appropriate than the noun fire. Our language supplies
the adjective hot, the noun heat, and the verb to heat, for what
physicists believe to be a movement of particles (molecules) in a
body. Similarly, school grammar defines the class of plural nouns
by its meaning “more than one” (person, place, or thing), but
who could gather from this that oats is a plural while wheat is a
singular? Class-meanings, like all other meanings, elude the linguist's
power of definition, and in general do not coincide with the
meanings of strictly-defined technical terms. To accept definitions
of meaning, which at best are makeshifts, in place of an identification
in formal terms, is to abandon scientific discourse.

Class-meanings are merely composites, or, one might say, greatest
266common factors, of the grammatical meanings which accompany
the forms. To state a class-meaning is to find some formula
that includes the grammatical meanings in which the forms occur.
An English finite verb expression (runs, ran away, is very kind,
scolded the boys, and so on) occurs only in one position of one construction,
namely as action in the actor-action construction (John
ran away
). Even when it is used alone, it appears only as a completive
sentence which, accordingly, presupposes an actor. Now,
we can state the meaning of the actor-action construction very
roughly as ‘A performs B,’ where A is the nominative expression
(John) and B the finite verb expression (ran away). This statement
defines for us the meanings of the two positions; the meaning
of the actor-position is ‘performer of B,’ and that of the action-position
is ‘performed by A.’ Therefore, since English finite verb
expressions occur only and always in this latter position, their
class-meaning is the same as that of their one position, namely,
‘performed by an object.’ If we define the class-meaning of the
larger form-class of verbs as ‘action,’ then the class-meaning of
English finite verb expressions is ‘(action) performed by an actor.’

When a form-class has more than one function, its class-meaning
is harder to state, but is still merely a derivative of the grammatical
meanings in which the forms occur. English substantive expressions
occur, for instance, in the position of actor in the actor-action
construction (John ran), with the positional meaning ‘performer
of an action.’ They occur in the position of goal in the action-goal
construction (hit John), with a positional meaning something like
‘undergoer of an action.’ They occur in the position of axis in
the relation-axis construction (beside John), with a positional
meaning of, say, ‘center from which a relation holds good.’ They
occur in morphologic construction with the possessive suffix
(John's), with the positional meaning of ‘possessor.’ Without listing
all the other functions of English substantive expressions, we
can say that the class-meaning common to all the lexical forms
in this form-class is ‘that which can be the performer of an action,
the undergoer of an action, tae center from which a relation holds
good, the possessor of objects,’ and so on. Whether we can sum
this up in a shorter formula, depends upon our resources of terminology;
for instance, we can sum up the class-meaning just
given, under the term ‘object.’

These instances suffice to show that class-meanings are not
267clearly-definable units which could serve as a basis for our work,
but only vague situational features, undefinable in terms of our
science. The people who speak English and keep their substantive
expressions within the accepted functions, do not guide themselves
by deciding whether each lexical form denotes an object. Form-classes,
like other linguistic phenomena, can be defined, not in
terms of meaning, but only in terms of linguistic (that is, lexical
or grammatical) features.

16. 3. The form-class of a lexical form is determined for the
speakers (and consequently for the relevant description of a
language) by the structure and constituents of the form, by the
inclusion of a special constituent (a marker), or by the identity of
the form itself.

(1) A complex form is usually assigned to a form-class by its
structure and constituents. An endocentric phrase, for instance,
such as fresh milk, belongs to the same form-class as its head or
center (§ 12.10). An exocentric phrase, such as in the house, contains
some characteristic constituent (as, in our example, the
preposition in) which determines its form-class. Thus, the form-class
of a phrase is usually determined, at bottom, by the form-class
of one or more of the included words. For this reason the
speaker (and the grammarian) need not deal separately with
each phrase; the form-class of almost any phrase is known if we
know the syntactic constructions and the form-classes of words.
The form-classes of words are therefore fundamental for syntax.
Our school grammar recognizes this: it tries, by a mistaken method,
to be sure, to determine the form-classes of words, particularly
the most inclusive of these form-classes (parts of speech), and then
shows how phrases are constructed.

(2) Sometimes the function of a phrase is determined by some
special constituent, a marker. For instance, in English, a phrase
consisting of the preposition to and an infinitive expression, belongs
to the special form-class of marked infinitive phrases, whose function
differs from that of unmarked infinitive expressions, since
they serve as actors (to scold the boys was foolish) and as attributes
of nouns, verbs, and adjectives (a chance to go; he hopes to go; glad
to go
). The determining adjectives form noun phrases which are
distinguished by closure: this fresh milk cannot take adjective modifiers
as can fresh milk or milk (§ 12.10). Whenever a form-class of
small extent determines a peculiar function in phrases, we may
268regard its forms as markers. Thus, our determining adjectives,
our prepositions, our co-ordinating conjunctions, and our subordinating
conjunctions, may be viewed as markers; they are small
form-classes, and the presence of any of their forms in a phrase determines
something about the form-class of this phrase. Other examples
of markers are the particles of Chinese or Tagalog (§ 12.13).

(3) Finally, lexical forms may belong arbitrarily or irregularly
to a form-class that is indicated neither by their structure nor
by a marker. For instance, the phrase in case has the structure
of preposition plus substantive and yet serves as a subordinating
conjunction: In case he isn't there, don't wait for him. The phrases
this way, that way, the other way, the same way have substantive
structure, but are used as verb-modifiers of the special sub-class
(manner) that has the interrogative substitute how? Similarly,
quite a few English nouns and noun phrases serve as verb-modifiers
in the when? class, either alone or in phrases: Sunday, last winter,
tomorrow morning. The form-classes of English words are largely
arbitrary: there is nothing to tell us that man, boy, lad, son, father
are male nouns, that run, bother are verbs, that sad, red, green
are adjectives, and so on. In particular, of course, the form-class
of every morpheme is arbitrarily determined. A complete description
of a language will list every form whose function is not determined
either by structure or by a marker; it will include, accordingly,
a lexicon, or list of morphemes, which indicates the
form-class of each morpheme, as well as lists of all complex forms
whose function is in any way irregular.

16. 4. Form-classes are not mutually exclusive, but cross each
other and overlap and are included one within the other, and so
on. Thus, in English, the nominative expressions (which serve as
actors) include both substantives and marked infinitives (to scold
the boys would be foolish
). On the other hand, among the substantives
are some pronoun-forms which, by over-differentiation, do
not serve as actors: me, us, him, her, them, whom. One group of
substantives, the gerunds (scolding), belongs to a form-class with
infinitives and with other verb-forms, in serving as head for certain
types of modifiers, such as a goal (scolding the boys). For
this reason a system of parts of speech in a language like English
cannot be set up in any fully satisfactory way: our list of parts
of speech will depend upon which functions we take to be the most

One can often distinguish, however, between great form-classes
like the above, and petty form-classes like that of foot, goose, tooth
or of ox (with irregular plural-forms). Large form-classes which
completely subdivide either the whole lexicon or some important
form-class into form-classes of approximately equal size, are called
categories. Thus, the English parts of speech (substantive, verb,
adjective, and so on) are categories of our language. So are
singular and plural substantives, since these two form-classes, of
approximately equal size, completely subdivide the form-class of
substantives. In general, inflectional forms, what with the parallel
occurrence in every paradigm, represent categories — for instance,
the various forms of the verb-paradigm, including the congruence-forms
of finite verbs (am : is : are or was : were) and, crossing
these, the tenses and modes of finite verbs (he is : he was : he were).

Not all categories, however, are inflectional. The selection of the
pronouns he versus she divides our personal nouns into the categories
of male and female; yet there is no inflection or regular
derivation to distinguish these, but only a sporadic use of markers
(count : countess, Paul : Pauline, Albert : Alberta) or of entirely
irregular derivation (duck : drake, goose : gander) or of composition
(he-goat, billy-goat, bull-buffalo) or suppletion (son : daughter, ram :
ewe) or merely class-cleavage (a teacher … he; a teacher … she;
Francis : Frances).

Again, some categories are syntactic, and appear not in inflection,
but in phrases. Such are the categories of indefinite and
definite substantives (a book : the book), or, in our verbs, the aspects
(wrote : was writing), completion (wrote : had written), or
voice (wrote : was written).

The categories of a language, especially those which affect
morphology (book : books, he : she), are so pervasive that anyone
who reflects upon his language at all, is sure to notice them. In
the ordinary case, this person, knowing only his native language, or
perhaps some others closely akin to it, may mistake his categories
for universal forms of speech, or of “human thought,” or of the
universe itself. This is why a good deal of what passes for “logic”
or “metaphysics” is merely an incompetent restating of the chief
categories of the philosopher's language. A task for linguists of the
future will be to compare the categories of different languages and
see what features are universal or at least widespread. Thus, a
form-class comparable to our substantive expressions, with a
270class-meaning something like ‘object,’ seems to exist everywhere,
though in many languages it is not an arbitrary class, like our
substantive part of speech, but depends largely upon the presence
of markers, as in Malayan or Chinese (§ 12.13).

16. 5. Our knowledge of the practical world may show that some
linguistic categories agree with classes of real things. It may be,
for instance, that our non-linguistic world consists of objects,
actions, qualities, manners, and relations, comparable with the
substantives, verbs, adjectives, adverbs, and prepositions of our
language. In this case it would still be true, however, that many
other languages do not recognize these classes in their part-of-speech
system. Moreover, we should still have to determine the
English parts of speech not by their correspondence with different
aspects of the practical world, but merely by their functions in
English syntax.

This appears plainly in the circumstance that languages with an
elaborate part-of-speech system always contain abstract forms;
they have parallel forms with the same lexical meaning for use in
different syntactic positions. Thus, a verb like run or an adjective
like smooth cannot serve as an actor, but we have for this function
the abstract noun forms run (as in the run will warm you up) and
smoothness. It is an error to suppose that abstract forms like these
occur only in the languages of literate peoples; they occur in all
languages that limit different form-classes to different syntactic

Linguistic categories, then, cannot be defined in philosophical
terms; having defined them in formal terms, we may have great
difficulty in describing their meaning. To show this, we need only
glance at some of the more familiar categories.

Number, as it appears in our singulars and plurals, seems to be
close to some universal trait of human response; yet, cases like
oats versus wheat, or Epsom salts versus table salt, seem to have
little non-linguistic justification.

The categories of gender in English are close to our non-linguistic
recognition of personality and sex, but even here some animals
(the bull … he or it) and other things (the good ship … she or
it) are variously treated. The gender-categories of most Indo-European
languages, such as the two of French or the three of
German (§ 12.7), do not agree with anything in the practical world,
and this is true of most such classes. In the Algonquian languages,
271all persons and animals belong to one category, an ‘animate’
gender, but so do some other objects, such as ‘raspberry,’ ‘kettle,’
and ‘knee;’ all other objects (including, for instance, ‘strawberry,’
‘bowl,’ ‘elbow’) belong to the other, ‘inanimate’ gender. Some
of the Bantu languages run up to as high as twenty such classes;
distinctions of number, however, are merged with the gender-classification.

Case-categories, ranging from two, as in English (he : him),
up to twenty or so, as in Finnish, resemble various situations of the
practical world, but never with any consistency. Thus, in German,
the goal of a verb is in the accusative case, as in er bat mich [e :r
ˈba:t mix] ‘he asked me (for something),’ but certain verbs have
it in the dative case, as er dankte mir [e:r ˈdaŋkte miːr] ‘he thanked
me;’ compare the Latin examples in § 12.8.

The categories of tense have a surface rationality, especially in a
language like Latin, which distinguishes present (cantat ‘he sings’),
past (cantāvit ‘he sang’), and future (cantābit ‘he will sing’), but
even here one soon finds that these categories disagree with our
non-linguistic analysis: the “historical present” is used in Latin,
as in English, of past events, and the meanings of the Latin tense-forms
are mixed up with considerations other than relative time.

The English categories of aspect distinguish between ‘punctual’
action (some grammarians call it ‘perfective’), envisaged as a unit
(he wrote the letter), and ‘durative’ action (some call it ‘imperfective’),
which extends over a segment of time during which other
things can happen (he was writing the letter). This distinction is at
best hard to define for the practical world, and in English suffers
marked dislocations; some verbs, for instance, appear persistently
in punctual form (I think he is there; he is funny) and are durative
only in special constructions or meanings (I am thinking of him;
he is being funny). In Russian, which has much the same aspects
as English, certain verbs, such as ‘eat’ and ‘drink,’ appear persistently
in durative form.

A common verb-category that is lacking in English, is iteration,
which distinguishes between an action occurring once and a repeated
action, as, in Russian [on beˈžal doˈmoj] ‘he was running
home’ (on one particular occasion) and [on ˈbegal doˈmoj] ‘he
ran home; he was running home’ (repeatedly, e.g. every day). 121272

Perfection contrasts contemporary, ‘imperfectic’ action with
‘perfectic’ action, whose effect is contemporary: he writes versus
he has written; he is writing versus he has been writing; he wrote
versus he had written; he was writing versus he had been writing.
The difference is scarcely definable in terms of practical situation,
and different languages show different distributions.

English has many modes, distinguishing various approaches of
an action to its actual occurrence. Morphologically, English distinguishes
between ‘real’ (he is here) and ‘unreal’ (if he were here);
syntactically, English recognizes a whole series by the peculiarity
of certain irregular (‘auxiliary’) verbs which are followed by an
infinitive without to: he will write, shall write, can write, must write,
may write. We may observe that in these combinations the infinitive
is rather persistently punctual, and only now and then
durative (I shall be writing); in Russian, the future tense, which
corresponds fairly well to our shall and will phrases, distinguishes
aspect just as exactly as do the present and past tenses. The uses
of different modes are tied up in many languages with differences
of syntactic position and congruence. In English, for instance, the
unreal appears only in clauses introduced by if or though, or in
combination with the phrasal mode-forms (he would help us, unreal
of he will help us). Similar complications appear in the uses of the
various modes of other languages, as, in French, je pense qu'il
[žə pɑns k i vjɛn] ‘I think he is coming,’ with the verb of the
clause in the ‘indicative’ (actual) mode, but je ne pense pas qu'il
[žə n pɑns pɑ k i vjɛn] ‘I don't think he is coming,’ with the
verb of the clause in the ‘subjunctive’ (possible) mode.

16. 6. We saw in § 16.3 that the function of some forms is
determined by their constituents or their construction. Any function
that is so determined is said to be regular, and a function
which is not so determined is said to be irregular. Thus, if we know
that the words fox and ox are singular common nouns, wavering
between non-personal and male personal gender, then we can say
that fox has the regular function of combining with the plural-suffix
[-ez] in the form foxes (since this function is shared by an
unlimited number of singular nouns), but that ox has the irregular
273function of combining with the plural-suffix [-n̩]. Linguists usually
apply the terms regular and irregular to the form itself, saying, for
instance, that the noun fox is regular and the noun ox irregular;
we must specify, of course, the function with respect to which
these terms hold good, since in their other functions the nouns fox
and ox are quite alike. By another extension of these terms, linguists
apply them also to the resultant forms in which the functions
appear, saying, for instance, that the plural noun foxes is regular
and the plural noun oxen irregular.

The speaker can use a form in a regular function even when he
has never heard the resultant form: he may utter a form like foxes,
for instance, even when he has never heard this particular plural.
He can use a form in an irregular function only if he has heard it
used in this function: the form oxen is uttered only by speakers
who have heard it from other speakers. In the description of a
language, accordingly, regular functions are stated for whole form-classes,
in the mass: we can state the regular plural-formation of
English nouns without attempting to list all the nouns in the
language. Irregular functions, on the other hand, force us to list
all the forms of the class: we have to mention the noun ox as
taking -en in the plural, and the nouns foot, tooth, goose as taking
substitution of [ij] in the plural, and so on.

If we insist on this distinction, we may say that any form which
a speaker can utter without having heard it, is regular in its immediate
constitution and embodies regular functions of its constituents,
and any form which a speaker can utter only after he
has heard it from other speakers, is irregular. Strictly speaking,
then, every morpheme of a language is an irregularity, since the
speaker can use it only after hearing it used, and the reader of a
linguistic description can know of its existence only if it is listed
for him. The lexicon is really an appendix of the grammar, a list
of basic irregularities. This is all the more evident if meanings
are taken into consideration, since the meaning of each morpheme
belongs to it by an arbitrary tradition. In a language like English,
where each morpheme is arbitrarily assigned to some grammatical
class, this feature also is an irregularity: the speaker must learn
from experience and the describer must list the fact that pin is
a noun, spin a verb, thin an adjective, in a preposition, and so on.
This task also is customarily assigned to the lexicon; the grammar
lists only the kinds of irregularity that are not present in all
274the morphemes of a language, and the terms regular and irregular
are used only of features that appear in the grammar.

If we make this restriction, it is obvious that most speech-forms
are regular, in the sense that the speaker who knows the constituents
and the grammatical pattern, can utter them without ever
having heard them; moreover, the observer cannot hope to list
them,, since the possibilities of combination are practically infinite.
For instance, the classes of nominative expressions and finite
verb expressions in English are so large that many possible actor-action
forms — say, a red-headed plumber bought five oranges
may never before have been uttered; by the same token, however,
we cannot be sure that this is true of any particular combination
which we may chance to hear. A grammatical pattern (sentence-type,
construction, or substitution) is often called an analogy.
A regular analogy permits a speaker to utter speech-forms which
he has not heard; we say that he utters them on the analogy of
similar forms which he has heard.

An irregular analogy, on the other hand, may cover a number
of forms, but a speaker will rarely utter a new form on the analogy
of those which he has heard. For instance, the phrases at least, at
, at best, at worst, at first, at last are built up on the same pattern
(at plus adjective in -st), but the analogy is limited to a very
few forms. In at all (where the adjective does not end in -st and
the sandhi is irregular) or in don't we have a unique analogy.
When the automobile came into use, one speaker was as well able
as another to form the compound automobile-driver, on the analogy
of cab-driver, truck-driver, and so on; a compound like cranberry,
on the other hand, with its unique first member, is uttered only
by speakers who have heard it. If we take meanings into consideration,
we can say the same of a speaker who uses the term blackbird
of the species of bird to which it customarily applies, for the compound
bears this meaning by an arbitrary tradition. A form like
charlestoner ‘one who performs the dance called Charleston’ is
formed on the regular analogy of dancer, waltzer, two-stepper, and
so on; a form like duchess (§ 10.6) is unique. On the border-line
we have cases like the feminines in -ess, which on the whole are
limited to traditional forms: we say poetess, sculptress, but not
*paintress; occasionally, however, a speaker will extend this analogy,
uttering such forms as, say, profiteeress, swindleress. Even
our rcot-forming morphemes (§ 14.9) have some flexibility; hearing
275a form like squunch in a meaning ‘step with suction-noise on
wet ground,’ we cannot tell whether the speaker has heard it or
is using the analogy of [skw-], as in squirt, squash, and [-onč],
as in crunch.

The regular analogies of a language are habits of substitution.
Suppose, for instance, that a speaker had never heard the form
give Annie the orange, but that he had heard or spoken a set of
forms like the following:

Baby is hungry. Poor Baby! Baby's orange. Give Baby the orange!

Papa is hungry. Poor Papa! Papa's orange. Give Papa the orange!

Bill is hungry. Poor Bill! Bill's orange. Give Bill the orange!

Annie is hungry. Poor Annie! Annie's orange.

He has the habit, now, — the analogy, — of using Annie in
the same positions as Baby, Papa, Bill, and accordingly, in the
proper situation, will utter the new form Give Annie the orange!
When a speaker utters a complex form, we are in most cases unable
to tell whether he has heard it before or has created it on the analogy
of other forms. The utterance of a form on the analogy of
other forms is like the solving of a proportional equation with an
indefinitely large set of ratios on the left-hand side:

image Baby is hungry : Annie is hungry | Poor Baby : Poor Annie | Baby's orange : Annie's orange | Give Baby the orange : x


image dog : dogs | pickle : pickles | potato : potatoes | piano : pianos | radio : x

16. 7. The power or wealth of a language consists of the morphemes
and the tagmemes (sentence-types, constructions, and
substitutions). The number of morphemes and tagmemes in any
language runs well into the thousands. In every language, moreover,
many complex forms carry specialized meanings which
cannot figure in a purely linguistic description but are practically
of great importance. The linguist can determine, for instance,
that English compounds of the type blackbird, bluebird, whitefish,
or phrases of the type give out, fall out, throw up, bear specialized
meanings, but he cannot evaluate these meanings, although in
practical life they are fully as useful as any sememe.276

Popularly, the wealth of a language is supposed to depend upon
the number of different words which it uses, but this number is
indeterminate, since words are freely formed according to the
analogies of morphologic construction. For instance, having
counted play, player, and dance, shall we count dancer as a fourth
word, even though it contains no additional glosseme? If so,
then the number of words in any language is practically infinite.
When we are told that Shakspere used 20,000 different words in
his writings, and Milton in his poems some 8,000, we mistakenly
conclude that less eloquent speakers use far fewer. It is an indication
of Shakspere's genius that he used so many different words
in so small a volume of speech as is contained in his works, but
this volume of speech is small compared to the amount which
even a taciturn person will utter in the course of a year. The
myths about peasants, workingmen, or savages who use only
a few hundred words have no foundation in fact; in so far as one
can count words (ignoring, for instance, the inflected forms of a
language like ours), every adult speaker uses at least somewhere
round 20,000 to 30,000 words; if he is educated — that is, if he
knows technical and learned words — he uses many more. Everyone,
moreover, understands more words than he uses.

The relative frequency of the various lexical and grammatical
units (morphemes and tagmemes) in a language can be studied
wherever we have copious records of normal utterances. In the
next chapters we shall see that our lack of such records is one of
the impediments to the historical study of language — for fluctuations
in the frequency of glossemes play an important part in
the changes that occur in every language.

The frequency of most lexical forms is doubtless subject to a
great deal of superficial fluctuation, according to the practical
circumstances. A word like thimble, say, or stove, might not occur
at all in long stretches of speech; yet such forms as these are used
by everyone when the occasion presents itself. The most frequent
forms, on the other hand, both lexical and, especially, grammatical,
are constantly demanded by the structure of the language. Such
counting as has been done has been confined to words. It is found
that the commonest words (the, to, is, etc.) make up a consistently
high percentage of what is spoken.

16. 8. The practical question as to what things can be said in
different languages, is often confused with questions of word-meanings
277and of categories. One language will use a phrase where
another uses a single word and still another a bound form. A meaning
that is categoric in one language (as, for instance, plurality
of objects in English) may appear only under particular practical
stimuli in another language. As to denotation, whatever can be
said in one language can doubtless be said in any other: the difference
will concern only the structure of the forms, and their connotation.
What one language expresses by a single morpheme
will in another language require perhaps a long phrase; what
one language says in a word may appear in another language as
a phrase or as an affix. Elements of meaning that appear in one
language because they belong to some category, even though they
are irrelevant to the practical situation, will be absent in another
language. In English we say Pike's Peak is high with a present-tense
verb; in Chinese or in Russian there would be no present-tense
element in a similar message.

It is a striking fact that the smallest units of signaling, the glossemes,
of different languages, differ vastly in practical value. This
is true even of closely related languages. Where we say ride,
German says reiten [ˈrajten] for riding on an animal, but fahren
[ˈfa:ren] for other kinds of riding, as in a vehicle. Where we say
on, German says auf when the force of gravity helps the contact,
as in ‘on the table,’ but otherwise an, as in ‘on the wall.’ Our
morning matches the French matin [matɛn], except when the
morning is viewed as a segment of time during which something
else can happen, as in ‘I slept all morning’ or ‘during the morning;’
in this case French uses a derivative matinée [matine]. Even things
which are easily denned and classified, receive the most diverse
treatment in different languages. Nothing could be more definite
than terms for simple biological relationship between persons.
Yet, beside words corresponding to our brother and sister, German
has a plural Geschwister [geˈšvister] that includes both sexes, as in
Wieviele Geschwister haben Sie? [viː ˈfiːle geˈšvister ˈha:ben ziː?]>
‘How many brothers and (or) sisters have you?’ Some languages
have here one word, regardless of gender, as Tagalog [kapaˈtid];
our brother corresponds to a Tagalog phrase [kapaˈtid na laˈla:ki],
where the last word means ‘male,’ and our sister to [kapaˈtid na
baˈba:ji], with the attribute ‘female.’ On the other hand, some
languages insist upon relative age: Chinese [ˈko1 ko1] ‘elder
brother,’ [ˈcjuŋ1 ti4] ‘younger brother,’ [ˈčje3 čje3] ‘elder sister,’
278[ˈmej4 mej4] ‘younger sister.’ An even more complicated terminology
appears in Menomini, which we can best elucidate if we use
the term sibling to mean ‘brother or sister.’ In Menomini the
terms are [nɛʔnɛh] ‘my elder brother,’ [neme:h] ‘my elder sister,’
[nehse:h] ‘my younger sibling,’ [neko:ʔsemaw] ‘my sibling of
opposite sex’ (i.e. ‘my brother’ when a woman says it, ‘my sister’
when a man says it), [ne:hkah] ‘my brother (man speaking)m,’
[ne:tɛkɛh] ‘my sister (woman speaking).’ The general term
[niːtɛsjanak] ‘my siblings’ is used in the plural when the siblings
are of both sexes and not all younger than the possessor.

Terms of relationship not only vary as in the above examples,
but also are used in situations that one cannot define. The Menomini
terms for ‘brother’ and ‘sister’ are used also for cousins,
provided the related parents are of the same sex: a man says
[ne:hkah] of his father's brother's son, and so on. Moreover, these
and some other terms are inherited: my father's brother's son's son
is also [ne:hkah]. Consequently, the meaning really hinges on the
consistency with which these relationships are remembered and

In the same way, plant-names, for example, are perhaps nowhere
used in a way that would be consistent with a botanist's classification
— even aside from such vague terms as tree, shrub, bush,
herb, reed, grass.

Even in such a sphere as that of the numbers, languages show
many deviations. Our system of decimal numbers (twenty-two,
thirty-five, etc.) shows traces of a duodecimal or twelves system
(eleven, twelve instead of *one-teen, two-teen). Other irregularities
are formal, as two : twenty : second : half, or three : thirteen, thirty,
third. Furthermore, the connotation of certain numbers like three,
seven, thirteen, and of additional terms like dozen, score, gross, cannot
be stated mathematically. In Danish there is an admixture of a
vigesimal or twenties system. In French one counts from ‘sixty’
to ‘seventy-nine’ without a special word for the intervening
multiple of ten: ‘seventy’ is soixante-dix [swasɑnt-dis] ‘sixty-ten;’
‘seventy-one’ is soixante et onze [swasɑnt e onz] ‘sixty and eleven,’
and so on; ‘eighty’ is quatre-vingt [katrə vɛn] ‘four-twenties,’
and then one counts up twenty more to reach one-hundred; thus,
‘ninety-two’ is quatre-vingt douze [katrə vɛn duːz] ‘four-twenties-twelve.’
Peoples who have little use for higher numbers may
use very few: the Kham Bushmen are said to count by simple
279numbers only to ‘three,’ and to use ‘two and two’ for ‘four,’
and so on.

In other spheres which are subject to scientific analysis, this
may still provide no gauge for the linguistic classification. Color,
for instance, is a matter of frequency of refracted or reflected light-waves.
The visible spectrum is an unbroken scale of frequencies.
Different languages use different color-names (such as our red,
orange, yellow, green, blue, violet, § 9.1) for different parts of this
scale. We should have a hard time deciding at what points on the
actual scale the domain of each English color-name begins and
ends. If we showed people colors in minute grades of variety, we
should find that between the frequencies which were named consistently,
say, as yellow and as green, there would be a border-zone,
where the naming wavered. If we went outside the European
culture-sphere, we should find entirely different distributions.

For most of our meanings we have not even this approach to an
external standard. Terms which relate to social behavior, such as
love, friend, kind, hate could be defined in terms of ethnology, folklore,
and sociology, provided these studies had reached a perfection
and accuracy undreamed of today. Terms which relate to states
of the speaker's body that are perceptible only to him, such as
queasy, qualmish, sad, gay, glad, happy, could be defined only if
we had a minute knowledge of what goes on inside a living person's
body. Even all this would not suffice for linguistic meanings that
have less practical bearing, such as categories of noun-gender or
verbal aspect. There seems to be no practical criterion by which
the gender of a noun in German, French, or Latin could be determined:
to define the meaning of the episememe ‘masculine’ in such
a language would be simply to list the markers of masculine nouns
and the nouns that belong arbitrarily to the class, and to say that
whatever is common, in the practical world, to all these objects,
is the “meaning.” of the masculine gender-category. The same is
true of the verbal aspects of English: the difference between wrote
and was writing is so elusive and differs so much for different verbs
and in different phrases, that the definer, after stating the main
principles, cannot do better than to resort to a demonstration by
means of examples.280

Chapter 17
Written Records

17. 1. The language of any speech-community appears to an
observer as a complicated signaling-system, of the kind that has
occupied us in the preceding chapters of this book. A language
presents itself to us, at any one moment, as a stable structure of
lexical and grammatical habits.

This, however, is an illusion. Every language is undergoing, at
all times, a slow but unceasing process of linguistic change. We
have direct evidence of this change in the case of communities
which possess written records of their earlier speech. The English
of the King James Bible or of Shakspere is unlike the English of
today. The fourteenth-century English of Chaucer is intelligible
to us only if we use a glossary. The ninth-century English of King
Alfred the Great, of which we have contemporary manuscript
records, seems to us like a foreign language; if we could meet
English-speakers of that time, we should not understand their
speech, or they ours.

The speed of linguistic change cannot be stated in absolute
terms. A speaker has no difficulty, in youth, in conversing with his
grandparents, or, in age, in conversing with his grandchildren, yet
a thousand years — say, thirty to forty generations — have
sufficed to change the English language to the extent we have just
indicated. During these generations, it must have seemed to each
London-English mother that her children were learning to speak
the same kind of English as she had learned in her infancy. Linguistic
change is far more rapid than biological change, but probably
slower than the changes in other human institutions.

Linguistic change interests us especially because it offers the
only possibility of explaining the phenomena of language. Speakers
acquire their habits from earlier speakers; the only explanation
of their habits lies in the habits of these earlier speakers. If we
ask, for instance, why present-day speakers use the form dog for
the animal ‘canis domesticus,’ or, let us say, why they add the
suffix [-ez, -z, -s] to derive plural from singular nouns, the obvious
281answer is that they acquired these habits, in infancy, from the
older people round them; if we then ask the same questions about
the habits of these older people, we are referred to the habits of
still older people, and so on, back into time, without limit. If we
could realize our diagram of density of communication (§ 3.4), in
which every speaker was represented by a dot and every utterance
by an arrow from the dot that represented the speaker to the dot
or dots that represented the hearer or hearers, we should find that
the network reached indefinitely back into time.

In the normal case, then, the explanation for a speech-habit is
simply the existence of the same habit at an earlier time. Where
linguistic change has been at work, however, the explanation will
be the existence of some other habit at an earlier time, plus the
occurrence of the change. Our lexical habit, for instance, of using
the word meat ‘edible flesh,’ is not very old; a few centuries ago,
the word flesh was used in this meaning, and the word meat meant
‘food.’ The explanation of our present-day habit, in this case,
consists in (1) the earlier habit, and (2) the intervening change.
Since linguistic change never stops, it sooner or later affects every
habit in a language; if we know enough of the speech of the past,
the second type of explanation will apply to every present-day

Since written records give us direct information about the
speech-habits of the past, the first step in the study of linguistic
change, wherever we have written records, is the study of these

We today are so used to reading and writing that we often confuse
these activities with language itself (§ 2.1). Writing is a relatively
recent invention. It has been in use for any considerable
length of time in only a few speech-communities, and even in these
its use has been confined, until quite recently, to a very few persons.
A speech-utterance is the same, whether it receives a written
record or not, and, in principle, a language is the same, regardless
of the extent to which speech-utterances of this language are
recorded in writing. For the linguist, writing is, except for certain
matters of detail, merely an external device, like the use of the
phonograph, which happens to preserve for our observation some
features of the speech of past times.

17. 2. Writing is an outgrowth of drawing. Probably all peoples
make pictures by painting, drawing, scratching, or carving. These
282pictures, aside from other uses (§ 2.9), sometimes serve as messages
or reminders — that is, they modify the conduct of the
beholder — and they may be persistently used in this way. The
Indians of North America are skilful draftsmen, and in older
times made extensive practical use of pictures. Thus, we are told
of an Ojibwa Indian who owned a long strip of birch-bark with a
series of pictures, which he used to remind himself of the succession
of verses in a sacred song. The third picture, for instance, represents
a fox, because the third verse of the song says something
about a fox, and the sixth picture represents an owl, because the
sixth verse says, “It is an ill omen.” A Mandan Indian sent the
following picture to a fur-trader: in the center are two crossed
lines; at one side of these lines are outline drawings of a gun and of
a beaver, with twenty-nine parallel strokes above the picture of the
beaver; at the other side of the crossed lines are drawings of a
fisher, an otter, and a buffalo. This means: “I am ready to trade
a fisher-skin, an otter-skin, and a buffalo-hide for a gun and thirty

Records and messages of this sort are usually spoken of as
“picture-writing,” but this term is misleading. The records and
messages, like writing, have the advantage of being permanent
and transportable, but they fall short of writing in accuracy, since
they bear no fixed relation to linguistic forms and accordingly do
not share in the delicate adjustment of the latter.

We have no record of any people's progress from this use of
pictures to the use of real writing, and can only guess at the
steps. In the use of pictures we can often see the beginnings of
the transition, and traces of it remain in the actual systems of

Real writing uses a limited number of conventional symbols.
We must suppose, therefore, that in the transition the pictures
became conventionalized. The way of outlining each animal, for
instance, becomes so fixed that even a very imperfect sketch leaves
no doubt as to the species of animal. To some degree this is true
of the pictures of American Indians. In actual systems of writing
we often find symbols which still betray this origin. In the so-called
hieroglyphic writing of ancient Egypt, most of the symbols are
conventional but realistic pictures, and many of them actually
denote the name of the object which they represent; thus, the
picture of a goose (drawn always in the same way) denotes the
283word [sʔ] 122 which means ‘goose.’ In Chinese writing, some of the
symbols, such as, for instance, the symbol for the word [ma3]
‘horse,’ still resemble a picture of the meaning of the word, and
this is sometimes true of the older shapes of characters whose
modern form shows no such resemblance.

When the picture has become rigidly conventionalized, we may
call it a character. A character is a uniform mark or set of marks
which people produce under certain conditions and to which,
accordingly, they respond in a certain way. Once this habit is
established, the resemblance of the character to any particular
object is of secondary importance, and may be obliterated by
changes in the convention of forming the character. These changes
are often due to the nature of the writing-materials. Some of
the characters of the cuneiform writing of the ancient Mesopotamian
peoples still betray their origin in pictures, but for the most
part this is not the case: the characters consist of longer and shorter
wedge-shaped strokes in various arrangements, and evidently
got this shape because they were scratched into tough clay. In the
hieroglyphic writing of ancient Egypt the characters were carefully
painted, but for rapid writing with a reed brush on papyrus the
Egyptians developed a simplified and rounded version (known as
hieratic writing) whose characters have lost all resemblance to
pictures. Our own writing is ultimately derived from the ancient
Egyptian, but no one could recognize pictures in our letters; as a
matter of fact, our letter F still has the two horns of the snail
which was pictured in the hieroglyphic ancestor of this letter.

The other, more important phase of the transition from the use
of pictures to real writing, is the association of the characters with
linguistic forms. Most situations contain features that do not lend
themselves to picturing; the picture-user resorts to all sorts of
devices that will elicit the proper response. Thus, we saw the
Indian drawing twenty-nine strokes above his beaver to represent
the number of beaver-pelts. Instead of depicting the process of
exchange by a series of pictures, he represented it by two crossed
lines with the sets of traded objects at either side. The Ojibwa
represented “ill omen” by an owl, in accordance, no doubt, with
some tribal belief.

When the picture-user was confronted by a problem of this kind,
we may suppose that he actually spoke to himself, and tried out
284various wordings of the troublesome message. Language, after all,
is our one way of communicating the kind of things that do not
lend themselves to drawing. If we make this supposition, we can
understand that the picture-users might, in time, arrange the
characters in the order of the spoken words of their language, and
that they might develop a convention of representing every part
— say, every word — of the spoken utterance by some character.
We can only guess at the steps of this transition: real writing
presupposes it.

In real writing, some characters have a twofold value, for they
represent both a picturable object and a phonetic or linguistic
form; other characters, having lost their pictorial value, represent
only a phonetic or linguistic form; purely pictorial characters that
are not associated with speech-forms sink into subsidiary use. The
linguistic value predominates more and more, especially as the
characters become conventionalized in shape, losing their resemblance
to pictured objects. The characters become symbols — that
is marks or groups of marks that conventionally represent some
linguistic form. A symbol “represents” a linguistic form in the
sense that people write the symbol in situations where they utter
the linguistic form, and respond to the symbol as they respond to
the hearing of the linguistic form. Actually, the writer utters the
speech-form before or during the act of writing and the hearer
utters it in the act of reading; only after considerable practice do
we succeed in making these speech-movements inaudible and

17. 3. Apparently, words are the linguistic units that are first
symbolized in writing. Systems of writing which use a symbol for
each word of the spoken utterance, are known by the misleading
name of ideographic writing. The important thing about writing
is precisely this, that the characters represent not features of the
practical world (“ideas”), but features of the writers' language;
a better name, accordingly, would be word-writing or logographic

The main difficulty about logographic writing is the providing
of symbols for words whose meaning does not lend itself to pictorial
representation. Thus, the Egyptians used a character that
represented a tadpole, to symbolize a word that meant ‘one-hundred
thousand,’ presumably because tadpoles were very numerous
in the swamps. The Chinese symbol for the word
285‘good’ is a combination of the symbols for ‘woman’ and for

The most important device of this sort is to use the symbol of
some phonetically similar word whose meaning is picturable.
Thus, the ancient Egyptians used the character that depicted a
goose, not only for the word [sʔ] ‘goose,’ but also for the word
[sʔ] ‘son,’ and they used the character that depicted a conventionalized
checkerboard, not only for [mn] ‘checkers,’ but also for
[mn] ‘remain.’ Chinese writing used the conventionalized character
depicting a wheat-plant not only for a word that meant ‘wheat,’
but also for the homonymous word that meant ‘come’ — in
present-day North Chinese, [laj2]. The ambiguity that arises in
this way, leads to a further development: one adds some character
that shows which of the similar words is to be read; these additional
characters are called classifiers or determinants. In Chinese
writing, which carries the logographic system to perfection, the
phonetic (as the basic symbol is called) and the classifier are united
into a single compound character. Thus, the symbol for [ma3]
‘horse’ and the symbol for [ny3] ‘woman’ are united into a compound
character, which serves as the symbol for the word [ma1]
‘mother.’ The symbol for [faŋ1] ‘square’ combines with the symbol
for [thu2] ‘earth’ into a compound symbol for [fan1] ‘district;’
with the symbol for [sr1] ‘silk,’ it forms a compound symbol representing
the word [faŋ3] ‘spin.’ The phonetic part of the compound
symbol, as these examples show, does not always accurately
represent the sound of the word; we have to suppose, however,
that at the time and in the dialect where this development took
place, the compound symbols (that is, such as were there and then
created) were phonetically accurate.

The logographic system, as we see it in Chinese writing, has the
disadvantage that one has to learn a symbol for every word of
the language. The compound symbols of Chinese writing can
all be analyzed into 214 constituents (“radicals”), but, even so,
the labor of learning to read and write is enormous. On the other
hand, this system has a great advantage in that the symbols are
non-committal as to the phonetic shape of the words. The Chinese
speak a number of mutually unintelligible dialects, but in writing
and printing they adhere to certain conventions of lexicon and
word-order and are thus able to read each others' writings and,
with some training, also the writings of their ancient literature.286

Our numerals (derived from ancient India) are examples of
logographic writing. A symbol like 4 is intelligible to many nations,
although we read it as [fowr], the Germans as [fiːr], the
French as [katr], and so on. Moreover, since we arrange the numerals
according to a fixed convention, we can read each others'
numeral phrases even though our languages differ as to the structure
of these phrases: 91, for instance, is everywhere intelligible,
although we say not [ˈnajn ˈwon] but [ˈnajntij ˈwon], and the Germans
say, in opposite order, [ˈajn unt ˈnojntsik] ‘one and ninety,’
and the French [katrə vɛn onz] ‘four twenties eleven,’ and the
Danes [ˈeʔn ɔ hal ˈfɛmʔs] ‘one and half five-times.’

17. 4. In the device of representing unpicturable words by
phonetically similar picturable words, we see the emergence of
the phonetic factor in writing. Once a symbol is associated with
a particular word, the phonetic features of this word may suffice
to bring about the writing of the symbol. In Chinese, where the
words are of uniform structure, this transference has been made
only from word to word, and the compound characters, in accordance
with this structure, are written as units and held down to
uniform size. In the writing of other languages, where words are
of various lengths, we find word-symbols used for phonetically
similar parts of longer words. Thus, the Egyptians wrote the
symbol for [mn] ‘checkerboard’ twice over to represent the word
[mnmn] ‘move.’ By a succession of the symbols for [mç] ‘duster’
and [Dr] ‘basket,’ they wrote the word [mçdr] ‘ear.’ In accordance
with the structural variety, they represented words not always
by one symbol, but also by various arrangements of logograms,
phonetics, and classifiers. Similarly, in Aztec writing, the place-name
Teocaltitlan, literally ‘god-house-people,’ was represented by
the symbols for tentli ‘lips,’ otli ‘path,’ calli ‘house,’ and tlantli
‘teeth;’ this is the more intelligible as the -tli in these words is
an inflectional suffix.

The symbols in this way may take on a more and more constant
phonographic value: they become phonograms — that is, symbols
not for linguistic forms, but for phonetic forms. The commonest
result seems to be a set of syllabic symbols, each one of which denotes
one syllabic sound with (or without) preceding and following
non-syllabics. The cuneiform writing of the ancient Mesopotamians
reached this stage; it had characters for such syllables
as [ma, mi, mu, am, im, um, muk, mut, nam, tim]. Throughout
287its use, as it passed from nation to nation, it carried along logographic
features. For instance, the ancient Sumerian word for
‘god’ was [an]; when the Babylonians learned the use of writing,
they took over the Sumerian symbol as a logogram for the Babylonian
word [ilu] ‘god,’ and as a classifier which they placed before
the names of gods. This kind of retention often occurs when a
system of writing is adapted to a new language; thus, we retain
Latin abbreviations, such as & (Latin et) for and; etc. (Latin et
‘and other things’) for and so forth; i.e. (Latin id est) for
that is; e.g. (Latin exempli gratia ‘for the sake of an example’)
for for instance; lb. (Latin libra) for pound, and so on.

In Babylonian writing the syllabic principle was never fully
carried out; thus, a single symbol (a vertical wedge with two small
wedges aslant at the left) represented the syllables [ud, ut, ut, tam,
par, pir, lax, xiš] and, logographically, the words [uːmu] ‘day,’
[šamšu] ‘sun,’ and [piçu] ‘white.’ In its Old Persian form, cuneiform
writing had developed into a genuine syllabary, with a relatively
small number of symbols, each representative of some one
syllable. In general, syllabic systems of writing are widespread
and seem to be easily devised. The ancient Greeks on the island
of Cyprus used a syllabary of some sixty-five symbols. The Japanese
largely use Chinese logographs, but supplement them with
two syllabaries, both of which are derived from Chinese characters.
The Vai, in Guinea, are said to have a system of 226 syllabic
signs. When persons acquainted with modern writing devise a
system for an illiterate people, they sometimes find it easiest to
teach syllabic writing. Thus, Sikwaya, a Cherokee, devised a set
of eighty-five syllabic symbols for his language; the Fox Indians
have several syllabaries, all based on English script forms; and
the Cree have a syllabary consisting of simple geometrical characters.

17. 5. It seems that only once in the history of writing there
has been any advance beyond the syllabic principle. Some of
the Egyptian hieroglyphic and hieratic symbols were used for
syllables containing only one consonant; in the use of these, differences
of the accompanying vowel were disregarded, and the resultant
ambiguities were removed by the use of classifiers and
logograms. In all, there were twenty-four of these symbols for one-consonant
syllables. At an early date — certainly before 1500 B.C.
— Semitic-speaking people became acquainted with Egyptian
288writing, and hit upon the idea of setting down words of their language
by means of the twenty-four simplest Egyptian symbols.
This was feasible because the structure of Semitic identifies each
root by its consonant-scheme (§ 14.8); the non-indication of vowels
could leave a reader in doubt only as to some features of word-derivation
which he might, in most instances, guess from the context.

Our oldest examples of this Semitic writing are the Sinai Inscriptions,
which date from somewhere round 1800 to 1500 B.C. One
later style of writing these characters is known as the South Semitic;
it is represented by old inscriptions and, in modern times, by
the Ethiopian alphabet. The other, North Semitic, style, was
used by the Phoenicians, the Hebrews, and the Arameans. The
Aramaic varieties include the style which we see in the modern
“Hebrew” type, the Syrian style, and the writing of modern
Arabic. It is the North Semitic character, in its Phoenician and
its Aramaic varieties, that has spread, with many changes, over
Asia and Europe.

The syllabaries used in India seem to be derived in part from
Aramaic, and mostly from Phoenician writing. For the languages
of India, indication of the vowel phonemes was necessary. The
Indians used each Semitic character for the syllable of consonant
plus [a] and then devised additional marks (diacritical signs) which
they added to the symbol to designate the combination of the
consonant with some other vowel. Thus, a simple sign means [ba],
and the same sign with various marks means [baː, bi, biː, bu, buː]
and so on. Further, the Indians devised a mark which meant that
the consonant was followed by no vowel at all, and a set of symbols
for vowels without any consonant. At the same time, they
increased the number of basic symbols until they had one for
each consonant phoneme. In this way they arrived at a system
which recorded their speech-forms with entire phonetic accuracy.

17. 6. Of all the offshoots, immediate and other, of Semitic
writing, we need trace only the one which includes our own system
of writing. The ancient Greeks took over the Phoenician system
and made a decisive change. Some of the Phoenician symbols
represented syllables containing consonants that were foreign to
Greek; thus, Α represented glottal stop plus vowel, Ο a laryngal
spirant plus vowel, and Ι the consonant [j] plus vowel. The Greeks
used these superfluous symbols to indicate vowel values, combining
289two symbols, such as ΤΑ or ΤΟ or ΤΙ, to represent a single syllable.
In this way they arrived at the principle of phonemic or alphabetic
writing — the principle of using a symbol for each phoneme. They
fell short of complete accuracy only because they failed to invent
enough symbols for vowels: they never distinguished between the
long and short quantities, distinctive in their language, of the
vowels [a, i, u]. They did later devise diacritical marks to indicate
the position and the two qualities of their word-accent, and some
signs of punctuation to indicate sentence-modulation.

From the Greeks the alphabet spread to other Mediterranean
peoples. The Romans received it apparently through the mediation
of the Etruscans. In the Middle Ages it passed from the Greeks
to the Bulgarians, Serbians, and Russians, and from the Romans,
directly or indirectly, to the other nations of Europe.

The transfer of writing to a new language occurs, apparently,
in this way, that some bilingual person who knows writing in one
language, hits upon the notion of using the alphabet also for his
other language. He may retain whatever defects the alphabet had
in the first language and he may retain letters that are necessary in
the first language but superfluous in the new one, and he may fail
to devise new letters for additional phonemes of the new language.
On the other hand, he or his successors may be clever enough to
mend these defects, either by inventing new characters or by
putting superfluous characters to good use, or by semi-phonetic
devices, such as using combinations of letters for a single

The phonetic pattern of Latin was such that the Greek alphabet,
as the Romans got it (probably from the Etruscans), was almost
sufficient. One defect, the use of the symbol C for both [k] and
[g], they mended by inventing the modified symbol G for [g]. A
more serious matter was the lack of symbols to distinguish long
and short vowels; the practice of placing a stroke over the letter or
of writing the letter twice to indicate length, never gained much
ground. There was no need for indicating the word-accent, since
this in Latin was automatically regulated according to the primary

The Germanic-speaking peoples took over the Graeco-Roman
alphabet, we do not know when or where, in a shape somewhat
different from the ordinary Greek or Latin styles. This form of the
alphabet, known as the runes, was used for short inscriptions,
290chiefly of magic or religious character, such as epitaphs. The runes
were not used skilfully, but they did include letters for some typically
Germanic phonemes, [θ, w, j]. The customary order of the
alphabet, too, was different from that of the Graeco-Roman prototype;
it ran: [f u θ arkgwhnijpɛzstbemlŋod]. For
this reason the runic alphabet is sometimes called the futhark.
The oldest runic inscriptions date from round 300 A.D. Later, as
the Germanic-speaking peoples were christianized by Romance and
Irish missionaries, they gave up the runes in favor of the Latin
alphabet. However, the Gothic bishop Ulfila, who in the fourth
century devised an alphabet for his Bible-translation, retained
several runic letters, and the Old English priests, in the eighth
century, when they took to writing English, retained the runic
characters for [θ] and [w], since the Latin alphabet provided none.
It was only after the Norman Conquest that English writers gave
up these letters in favor of the combinations th and vv (whence our
w). The five Latin vowel letters have never sufficed for English;
on the other hand, we retain the superfluous letters c, q, and x.
The writing of present-day English lacks symbols for the phonemes
[a, ɛ, ɔ, θ, ð, š, ž, č, ŋ] and for the stress-accent. This lack is only
partially repaired by the use of digraphs, such as th, sh, ch, ng.

Occasionally we find our alphabet fully adapted to the phonetic
system of some language. In the ninth century, the apostles Cyril
and Method added enough extra letters to the Greek alphabet to
make it cover the primary phonemes of the Old Bulgarian language.
This Slavic alphabet, in its modern form, is well suited to the
Slavic languages; for Serbian, some extra characters have been
added. Several modern languages have adequate forms of the
Latin alphabet; in the case of Bohemian and of Finnish, this result
has been reached by the use of diacritical marks, and in the case of
Polish by the use also of digraphs, such as cz for [č] and sz for [š].

17. 7. The principle of alphabetic writing —- one symbol for
each phoneme — is applicable, of course, to any language. The
inadequacy of the actual systems is due largely to the conservatism
of the people who write. The writer does not analyze the phonetic
system of his speech, but merely writes each word as he has seen
it in the writings of his predecessors. When the art of writing becomes
well established in a community, not only the spellings of
words, but even lexical and grammatical forms become conventional
for written records. In this way, a literary dialect may become
291established and obligatory for written records, regardless of the
writer's actual dialect.

This conservatism, as time goes on, works also in another way:
the conventions of writing remain unaltered even though the
speech-forms have undergone linguistic change. For instance, in
Latin writing the letter C represented the phoneme [k]. When the
Irish and the English took over the Latin alphabet, they used this
letter for their [k]-phonemes; in Old English, cu spelled [kuː] ‘cow,’
cinn spelled [kinn] ‘chin,’ and scip spelled [skip] ‘ship.’ Later on,
the phoneme [k] underwent certain changes in the various dialects
of Latin. In Italy, [k] before front vowels became [č]; Latin [ˈkentum]
‘hundred,’ for instance, became Italian [ˈčɛnto]. The Romans
wrote their word as centum; the Italians still write cento. In France,
the Latin [k] before front vowels has become [s], as in [sɑn] ‘hundred,’
but the French still write this word as cent. In English, we
have taken our foreign-learned words from French, with the [s]
pronunciation, but also with the traditional spelling with C, as in
the word cent [sent]. In Latin, the letters A, E, I, O, U were used
for the phonemic types [a, e, i, o, u], and they were taken into
English writing in these values. Thus, in medieval English writing,
a graph like name represented a form like [ˈna:me] ‘name.’ In the
fifteenth century, English spelling became conventionally fixed in
much its present shape. Since that time, however, our vowel
phonemes have undergone a great deal of change. The result has
been that we use the Latin vowel-letters not only in entirely new
values — this, after all, would do no harm -— but in inconsistent
ways. We have kept on using the letter A in graphs like name, hat,
all, far, although these words have now entirely different syllabic
phonemes. Sounds which existed when our spelling became habitual,
but have since been lost by linguistic change, are still represented
in our writing by silent letters, as in name, know, gnat,
bought, would.

Once a system of spelling has become antiquated in its relation
to the spoken sounds, learned scribes are likely to invent pseudo-archaic
spellings. The words debt, doubt, subtle contained no [b]-sound
in Old French, whence English received them, and were
written both in French and in English as dette, doute, sutil; the
present-day spellings with b were invented by scribes who knew
the far-off Latin antecedents of the French words, debitum, dubito,
subtilis. The letter s in isle reflects the Old French spelling isle
292(from Latin insula); at the time when the word was taken into
English it no longer had an [s] (compare modern French île [iːl])
and was appropriately spelled ile. The scribes not only favored the
spelling with s, but even introduced the letter s into two similar
words which had never contained any [s]-sound, namely the native
English island (from Old English iglond) and the French loan-word
aisle (French aile, from Latin āla). People who saw the runic letter
þ in ancient English writings but did not know its value [θ], took
it to be a form of the letter y and arrived at the notion that the
article the was in older English ye.

17. 8. It is evident, from all this, that written records give us
only an imperfect and often distorted picture of past speech, which
has to be deciphered and interpreted, often at the cost of great
labor. To begin with, the values, logographic or phonographic, of
the written signs may be unknown. In this case, the problem of
decipherment is sometimes desperate. The best help is a bilingual
inscription, in which by the side of the undeciphered text there is a
version in some known language; other aids are some knowledge
of the language or of the contents of the inscription. In 1802
Georg Friedrich Grotefend succeeded in deciphering cuneiform
inscriptions in Old Persian, and round the middle of the nineteenth
century a succession of workers (E. Hincks, Rawlinson, Oppert)
deciphered those in Babylonian-Assyrian; in both instances the
decipherers made ingenious use of their knowledge of related
languages. The cuneiform texts in other languages (Sumerian, the
language of Van, and Hittite) were deciphered thanks to bilingual
texts, such as dictionary-like tablets of word-lists in Sumerian,
Assyrian, and Hittite. In 1821 Jean Francois Champollion began
the decipherment of ancient Egyptian writings by using the famous
Rosetta Stone (found by the French in 1799; now in the British
Museum), which bears parallel inscriptions in hieroglyphics, in a
later form of Egyptian writing, and in Greek. In 1893 Vilhelm
Thomsen deciphered the Old Turkish Orkhon inscriptions; Thomsen
saw that the writing was alphabetical and the language of the
Turk family. The hieroglyph-like inscriptions of the Hittites and
those of the ancient Cretans have never been deciphered; of the
Maya picture-writing in Central America only some characters,
denoting months, days, numbers, and colors, have been interpreted.

If the system of writing is known, but the language is not, the
situation is little better. The most famous instance of this is the
293Etruscan language in ancient Italy; we have extensive texts in
a form of the Greek alphabet, but cannot interpret them, beyond
reading personal names and a few other words. We have dice
with the first six numbers written on the faces, but cannot determine
the order of these numbers. The Lydian inscriptions in Asia
Minor are intelligible, thanks to a bilingual text in Lydian and
Aramaic; the alphabet is Greek, and the language apparently
related to Etruscan.

17. 9. When both the system of writing and the language are
intelligible, we aim, of course, to learn from the texts all we can
get as to phonetics, grammar, and lexicon. The phonetic values
of the characters in ancient writings can never be surely known;
thus, the actual sounds represented even by the alphabetic symbols
of languages like Ancient Greek, Latin, Gothic, or Old English,
are in part uncertain. When the writing has become conventional
and unphonetic, the lapses of scribes or the way they write
uncommon words, may betray the real phonetic values. Our
Old English manuscripts show the same inflectional system from
the ninth century until well into the eleventh century, distinguishing
the vowels of unstressed syllables and the presence of
final m and n; but occasional lapses of the scribes betray the fact
that already in the tenth century most of these vowels had changed
to [e] and the final [m] and [n] had been lost; such lapses are, for
instance, spellings like worde for usual worda ‘of words,’ fremme
for normal fremman ‘to make,’ gode for godum ‘to good ones.’
When an English writer in the fifteenth century spells behalf
without an l, we infer that he no longer pronounced the [l] in this
word, although the tradition of writing insists upon the symbol
to this day. So-called inverse spellings tell the same story. Old
English had a sound [x] in words like light, bought, eight, which
is still reflected in our spelling with gh. When we find the word
deleite (a loan from Old French deleiter), which never contained
the sound [x], spelled delight, then we may be sure that the [x]
was no longer spoken in words like light: for the writers, the gh
was now a mere silent graph, indicative only of vowel-quantity.

A serious factor in the linguistic interpretation of written documents
is their transmission. Inscriptions, chiefly on stone or metal
or, as in the cuneiform texts, on clay, are generally original notations;
we need reckon only with one scribe's errors of spelling or
dictation. Most writing, however, is made on perishable material,
294and has come to our time through successive copyings. Our manuscripts
of Greek and Latin writings date from the Middle Ages,
often from the later Middle Ages or from the early modern period;
only fragments have been preserved on papyrus in the sands of
Egypt. It is rare good fortune when we have a contemporary manuscript
of an ancient text, like the Hatton manuscript of Alfred
the Great's translation of Pope Gregory's Pastoral Care. The
scribes not only made mistakes in copying, especially where they
did not understand the text, but they even tampered with it, by
way of improving the language or falsifying the content. The
study of ancient writing, paleography, and the technique of reconstructing
ancient texts from one or more imperfect copies, textual
, have developed into separate branches of science.
Unfortunately, textual critics have sometimes lacked linguistic
knowledge; our printed editions of ancient texts may fail to report
linguistically valuable forms that appear in the manuscripts.

Sometimes the text which appears in our written records has
undergone re-spelling into a new alphabet or a new system of
orthography. This is the case with our text of the ancient Greek
Homeric poems, and with our texts of the Avesta. We try, in
such cases, to reconstruct the original spellings and to detect
misleading or erroneous features in the traditional text.

17. 10. There are a few side-issues which sometimes help us
in the linguistic interpretation of written records. In the forms
of composition which we group together under the name of verse,
the author binds himself to observe certain phonetic patterns.
In modern English verse, for instance, the author shapes his wording
so that stress-phonemes come at certain intervals, and that
words of like ending, from the stressed syllabic to the end, occur
in pairs or larger sets, again at certain intervals. Thus, if we know
that a poet composed under a convention of exact rimes, we can
gather from his rime-words a great deal of information that may
not appear in the spellings. Chaucer rimed — to quote the words
in their present-day spellings — mean with clean, but not with
keen, queen, green: he evidently spoke different vowels in these
two sets of words. On the other hand, inconsistencies are equally
illuminating. When the Alsatian poet Brant, at the end of the
fifteenth century, rimes the word for ‘not’ both in the Alsatian
form [nit], as, for instance, with Bitt [bit] ‘request,’ and in the
present-day standard German form [nixt], as, for instance, with
295Geschicht [geˈšixt] ‘story,’ we know that in his day the modern
standard form, nicht [nixt] ‘not’ had already gained currency
alongside the provincial form of the word. Even when rimes are
used traditionally after they cease to be phonetically true, as,
in modern English poetry, rimes like move : love or scant : want,
a study of the tradition may be of interest.

Other types of verse lead to similar deductions. In old Germanic
poetry, high-stressed words occurred in alliterative sets with the
same initial consonant, as in house and home, kith and kin. Accordingly
when in ancient Icelandic verses of the Eddie poems we
find [ˈwega, ˈvega] ‘strike’ alliterating with [rejðr] ‘wroth,’ we
conclude that the men who coined this alliteration still pronounced
the latter word with an initial [wr-], although the spelling of our
manuscripts, in accordance with the later language, no longer
shows the [w]. In Greek and Latin verse the succession of long
and short syllables was regulated; a syllable containing a long
vowel or a diphthong, or any vowel followed by more than one
consonant, counted as long; the position of words in verse thus
often informs us as to vowel-quantities, which are only in part
shown by Greek orthography and not at all by Latin.

Another occasional help toward the interpretation of written
records is the transcription of speech-forms from one language
into another. At the beginning of the Christian era we find the
name of Caesar written in Greek texts as kaisar: since the Greek
language has not undergone a change of [k] to [č] or the like, and
the Greek k, accordingly, represented always a phoneme of the [k]
type, this transcription makes it likely that Latin at that time
still preserved the [k-]. The old Chinese transcriptions of Indo-Aryan
names in Buddhist texts give information about the sounds
which were attached to Chinese logographic symbols.

Finally, written records may contain statements of a linguistic
nature, as in the case of Sanskrit grammar and lexicon (§ 1.6);
the Hindus, moreover, were excellent phoneticians and interpreted
the written symbols in physiologic terms. Often enough, however,
we have to distrust the information in our texts. The Latin grammarians
give us little help as to speech-sounds; the English phoneticians
of the early modern period, likewise, confused sounds with
spellings and give very poor guidance as to the actual pronunciation
of their time.296

Chapter 18
The Comparative Method

18. 1. We saw in Chapter 1 that some languages resemble each
other to a degree that can be explained only by historical connection.
Some resemblance, to be sure, may result from universal
factors. Such features as phonemes, morphemes, words, sentences,
constructions, and substitution-types, appear in every language;
they are inherent in the nature of human speech. Other features,
such as noun-like and verb-like form-classes, categories of number,
person, case, and tense, or grammatical positions of actor, verbal
goal, and possessor, are not universal, but still so widespread
that better knowledge will doubtless some day connect them with
universal characteristics of mankind. Many features that are
not widespread — among them some very specific and even minute
ones — are found in distant and wholly unrelated languages;
these features, too, may be expected some day to throw light on
human psychology.

Other resemblances between languages bear no significance
whatever. Modern Greek [ˈmati] means ‘eye,’ and so does the
Malay word [mata]. If we knew nothing of the history of these
languages, we should have to work through their lexicons and
grammars in search of other resemblances, and then weigh the
probabilities of historical connection, taking into account both
the number of resemblances and their structural position. Actually,
our knowledge of the past forms both of Greek and of
Malay shows us that the resemblance of the two words for ‘eye’
is accidental. Modern Greek [ˈmati] is a relatively recent development
from an ancient Greek [omˈmation] ‘little eye,’ and this
word was in ancient Greek connected, as a secondary derivative,
with an underlying word [ˈomma] ‘eye.’ The Malay word [mata],
on the other hand, had in ancient times much the same phonetic
shape as today. Even if, against all present seeming, it should turn
out, some day, that these two languages are related, the relationship
would lie far back of Primitive Indo-European and Primitive
Malayo-Polynesian time, and the resemblance of the modern
words for ‘eye’ would have nothing to do with this relationship.297

Still other resemblances are due to the borrowing of speech-forms.
In modern Finnish there are many words like abstraktinen
‘abstract,’ almanakka ‘almanac,’ arkkitehti ‘architect,’ ballaadi
‘ballad,’ and so on through the dictionary—cultural words of
general European distribution, which have been borrowed, in the
last centuries, from one European language into the other, and
evidence nothing about kinship. To be sure, we cannot always
distinguish between this sort of transmission and the normal handing
on of linguistic habits within a speech-community, but for the
most part the two processes are very different. If the Finno-Ugrian
languages should be related to the Indo-European, then the kinship
dates from a time when the words abstract, almanac, etc., were
not yet in use.

18. 2. When we say, in contrast with these cases, that a resemblance
between languages is due to relationship, we mean that
these languages are later forms of a single earlier language. In the
case of the Romance languages, we have written records of this
parent language, namely, Latin. After the Latin language had
spread over a large area, it underwent different linguistic changes
in different parts of this area, so that today these different parts
differ greatly in speech, and we call the divergent speech-forms
“Italian,” “French,” “Spanish,” and so on. If we could follow
the speech, say of Italy, through the last two-thousand years, we
could not pick out any hour or year or century when “Latin”
gave way to “Italian;” these names are entirely arbitrary. By
and large, any feature that is common to all the modern territorial
forms of Latin, was present in the Latin of two-thousand years ago;
on the other hand, when the modern forms of Latin disagree as to
any feature, then some or all of them have, in this feature, undergone
some change during the last two-thousand years. The resemblances
appear especially in features that are common in
everyday speech — in the commonest constructions and form-classes
and in the intimate basic vocabulary. The features of
difference, moreover, appear in systematic groups, with each territorial
form diverging in its own characteristic way.

In most cases we are less favorably situated, in that we possess no
written records of the uniform parent speech. The Germanic languages,
for instance, resemble each other much as do the Romance,
but we have no records from a time when the differences had not
yet arisen. The comparative method, however, makes the same inferences
298in both cases. In the latter case we merely lack the
confirmation of the written record. We assume the existence, at
some time in the past, of a Primitive Germanic parent language,
but the speech-forms of this language are known to us only by
inference. When we write them down, we indicate this by placing
an asterisk before them.

18. 3. Compare, for instance, the following words in present-day
standard English, Dutch, German, Danish, and Swedish:

tableau English | Dutch | German | Danish | Swedish | ‘man’ | ‘hand’ | ‘foot’ | ‘finger’ | ‘house’ | ‘winter’ | ‘summer’ | ‘drink’ | ‘bring’ | ‘lived’

This list could be extended almost indefinitely; the resemblances
are so many and they so thoroughly pervade the basic vocabulary
and grammar, that neither accident nor borrowing will explain
them. We need only turn to languages outside the Germanic
group to see the contrast, as in ‘hand’: French [mɛn], Russian
[ruˈka], Finnish käsi; or ‘house’: French [mezon], Russian [dom],
Finnish talo. Another remarkable feature is the systematic grouping
of the differences within the Germanic family. Where Swedish
has the compound intonation, there Danish lacks the glottal stop;
where the others have initial [f], there Dutch has initial [v]; where
the others have [d], there German has [t]. In fact, whole series of
forms show the same divergences from one Germanic language to
the other. Thus, the divergent syllabic phonemes in the word
house are paralleled in a whole set of forms:

tableau English | Dutch | German | Danish | Swedish | ‘house’ | ‘mouse’ | ‘louse’ | ‘out’ | ‘brown’299

The fact that the differences themselves follow a system, — that
the divergence, say, of English and German [aw] and Dutch [øy] appears
in a whole series of forms — confirms our surmise that these
forms are historically connected. The divergence, we suppose, is due
to characteristic changes undergone by some or all of the related languages.
If we extend our observation to cover more of the dialects
in each area, we find many other varieties, with a similar parallelism.
In particular, we find, in our example, that forms with the vowel
[uː], such as [huːs, muːs] etc., occur also in local dialects of the English,
Dutch, and German areas — as, for instance, in Scotch English.

Further, availing ourselves of the written records of these
languages, we find that the oldest records from the English and
Dutch-German areas, dating round the eighth and ninth centuries
of our era, write the forms in our example uniformly with the
letter u, as hus, mus, lus, ut (southern German uz), brun. Since the
writing of these peoples was based on Latin, where the letter u
represented vowels of the. type [u], we conclude that the divergences
in the syllabic of our forms had not yet arisen in the ninth century,
and that the syllabic in those days was [u] in all the Germanic
languages; other evidence leads us to believe that the vowel was
long [uː]. Accordingly, we conclude that the Primitive Germanic
parent language spoke these forms with [uː] as the syllabic. It is
important to observe, however, that this description of the phoneme
is only a supplementary detail; even if we made no surmise as
to the acoustic character of the Primitive Germanic phoneme, the
regularity of the correspondences, in the way of agreement and in
the way of parallel disagreement, could still be explained only on the
supposition that some one phoneme of the parent language appeared
in the syllabic position of the forms house, mouse, and so on.

18. 4. It is interesting to compare these inferences with the
inferences that are made in the more favorable case, where the
parent language is known to us from written records. The resemblance
between the Romance languages is much like that between
the Germanic languages.

tableau Italian | Ladin | French | Spanish | Roumanian | ‘nose’ | ‘head’ | ‘goat’ | ‘bean’ 123300

Here we follow the same procedure as with the Germanic correspondences,
observing the local types in each area, and the
spellings of the older records. The difference is only this, that
written notations of the form of the parent language, Latin, are in
most instances available. The Romance words in our example are
modern forms of the Latin words which appear in our records as
nasum, caput, capram, fabam.

After we have learned to draw inferences from the Romance
forms, we may find discrepancies between the result of our inferences
and the written records of Latin. These discrepancies are
especially interesting because of the light they throw on the value
of our inferences in cases where no record of the parent language
is available. Take, for instance, the syllabic in the following types:

tableau Italian | Ladin | French | Spanish | Roumanian | ‘flower’ | ‘knot’ | ‘vow’ | ‘tail’ 124 225

The Latin prototypes appear in the first three of these words, as
well as in a number of similar cases, with a syllabic o, which we
interpret as [o:]: florem, nodum, uotum. In our fourth word, accordingly,
we infer that the Latin prototype contained this same
vowel and had the form *[ˈko:dam]. An inference of this kind is a
reconstruction; we mark the reconstructed form, *[ˈko:dam] or
*cōdam, with an asterisk. Now, in the written records of Latin,
the word for ‘tail’ appears in a different shape, namely as caudam
(accusative singular; the nominative is cauda). This disagrees
with our reconstruction, for ordinarily Latin au (presumably
[aw]) is reflected in the Romance languages by a different type of
vowel-correspondence. Thus, Latin aurum ‘gold’ and causam
‘thing, affair’ appear as:

tableau Italian | Ladin | French | Spanish | Roumanian | ‘gold’ | ‘thing’

It is true that our Latin manuscripts, written in the Middle Ages,
occasionally spell the word for ‘tail’ as coda, but this may be due
merely to the errors of copyists; the older manuscripts from which
301ours were copied may have had the usual Latin form cauda. This
error would be natural for copyists whose school pronunciation of
ancient Latin did not distinguish between Latin o and au, and
would be almost inevitable for copyists who spoke a form of
Latin in which our word already had, as in the present-day languages,
the vowel of florem, nodum, votum and not that of aurum,
causam. That some people were in this latter position appears
from the gloss, preserved to us in ninth-century manuscripts,
which explains the word cauda by saying that it means coda:
apparently, the former seemed antique and difficult, while the
latter was intelligible. The conclusive support for our reconstruction
appears in this, that inscriptions of early date show occasional
spellings of o in words that ordinarily have au, as Pola for the
name Paulla in an inscription dating from the year 184 B.C.
Further, we learn that this o-pronunciation for aw-forms was a
vulgarism. Suetonius (who died about 160 A.D.) tells us that the
rhetorician Florus corrected the Emperor Vespasian (died 79 A.D.)
for saying plostra instead of the more elegant plaustra ‘wagons;’
the next day, the emperor got back at him by calling him Flaurus
instead of Florus. As to our word, a grammarian of the fourth
century A.D. speaks of cauda and coda as variant pronunciations.
Moreover, we occasionally find over-elegant forms, like Vespasian's
Flaurus for Florus; an inscription dating from before the
beginning of the Christian Era has the spelling Avstia for ostia
[o:stia] ‘doors.’ In sum, we conclude that our reconstructed
*coda *[ko:da] is by no means illusory, but represents a less elegant
pronunciation which really existed in ancient time.

Cases like this give us confidence in the reconstructed forms.
Latin writing did not indicate vowel-quantities; a graph like secale
‘rye’ could represent several phonetic types. As this word
does not occur in verse, where its position would show us the vowel-quantities
(§ 17.10), we should be unable to determine its form,
had we not the evidence of the comparative method: forms like
Italian segola [ˈsegola], French seigle [sɛ:gl] show us that the Latin
graph represents the form [ˈse:kale]. Students of the Romance
languages reconstruct a Primitive Romance (“Vulgar Latin”)
form before they turn to the written records of Latin, and they
interpret these records in the light of the reconstructed form.

18. 5. A reconstructed form, then, is a formula that tells us
which identities or systematic correspondences of phonemes appear
302in a set of related languages; moreover, since these identities
and correspondences reflect features that were already present
in the parent language, the reconstructed form is also a kind of
phonemic diagram of the ancestral form.

In the oldest records of the Germanic languages we find the
following forms of the word father:

Gothic, text composed in the fourth century A.D., preserved
in a sixth-century manuscript: fadar, presumably [ˈfadar]; the
phoneme represented by d may have been a spirant.

Old Norse, in thirteenth-century manuscripts of texts that
were, in part, composed much earlier: faðer, faðir, presumably

Old English, ninth-century manuscripts: fæder, presumably
[ˈfɛder]. 126

Old Frisian, thirteenth-century manuscripts of texts that were
composed somewhat earlier: feder, presumably [ˈfeder].

Old Saxon (that is, northerly parts of the Dutch-German area),
ninth-century manuscripts: fader, presumably [ˈfader].

Old High German (southerly parts of the Dutch-German area),
ninth-century manuscripts: fater, presumably [ˈfater].

We sum up these facts by putting down the Primitive Germanic
prototype as *[ˈfader]; moreover, we claim that this summarizing
formula at the same time shows us the phonemic structure of the
prehistoric form.

Our formula embodies the following observations.

(1) All the Germanic languages stress the first syllable of this
word, as of most others. We indicate this in our formula by an
accent-mark, or, since accent on the first syllable is normal in
Germanic, by writing no accent-mark at all. This means, at the
same time, that in the Primitive Germanic parent language this
word shared with most other words a phonemic feature (call it x)
which appears in all the actual Germanic languages as a high
stress on the first syllable of the word. Of course, it is almost a
certainty that this feature x in the parent speech was the same
as appears in all the actual Germanic languages, namely, a high
stress on the first syllable, but this additional surmise in no way
affects the validity of the main conclusion.

(2) All the old Germanic languages begin the word with [f].
303If we had not the older records, we should have to consider the
fact that some present-day dialects of the English and of the Dutch-German
areas have here a voiced spirant of the type [v], but
the geographic distribution would even then show us that [f] was
the older type. In any case, the structural value of the symbol
[f] in our formula is merely this, that the word father in the Germanic
languages begins, and in Primitive Germanic began, with
the same phoneme as the words foot, five, fee, free, fare, and so
on, all of which we symbolize by formulas with initial [f].

(3) The [a] in our formula says that we have here the same correspondence
as in words like the following:

water: Gothic [ˈwato:], Old Norse [vatn], Old English [ˈwɛter],
Old Frisian [ˈweter], Old Saxon [ˈwatar], Old High German [ˈwassar],
Primitive Germanic formulas *[ˈwater, ˈwato:];

acre: Gothic [ˈakrs], Old Norse [akr], Old English [ˈɛker], Old
Frisian [ˈekker], Old Saxon [ˈakkar], Old High German [ˈakxar],
Primitive Germanic formula *[ˈakraz];

day: Gothic [dags], Old Norse [dagr], Old English [dɛj], Old
Frisian [dej], Old Saxon [dag], Old High German [tag], Primitive
Germanic formula *[ˈdagaz].

In this case the deviations, namely Old English [ɛ] and Old
Frisian [e] beside the [a] of the other languages, do not occur in
all forms; all the dialects have [a], for instance, in cases like the

fare: Gothic, Old English, Old Saxon, Old High German [ˈfaran],
Old Norse, Old Frisian [ˈfara], Primitive Germanic formula *[ˈfaranan].

In fact, the English [ɛ] and the Frisian [e] occur under fixed
phonetic conditions — namely, in monosyllables, like day, and
before an [e] of the next syllable, as in father, water, acre. This
deviation, we infer, is due to a later change, perhaps in a common
intermediate Anglo-Frisian parent language. We are safe, in
any case, in setting up, for all these words, a single structural
phonemic unit [a] in the Primitive Germanic parent language.

(4) The acoustic value of the Gothic letter which we have transliterated
as d is doubtful; it may have been a stop of the type
[d] or a spirant of the type [ð], or it may have fluctuated, in which
case [d] and [ð] were variants of one phoneme. The old Scandinavian
graph speaks for [ð] in this area. The West Germanic
languages have an unmistakable [d], which, in this as in other
304cases, appears in South German as [t]. In our Primitive Germanic
formula we indicate all this by the symbol [d] or [ð]; the former
is preferable because easier to print. Our formula identifies the
phoneme with that which appears in cases like the following:

mother: Old Norse [ˈmo:ðer], Old English [ˈmo:dor], Old Frisian
[ˈmo:der], Old Saxon [ˈmo:dar], Old High German [ˈmuotar],
Primitive Germanic formula *[ˈmo:der];

mead: Old Norse [mjɔðr], Old English [ˈmeodo], Old Frisian
[ˈmede], Old High German [ˈmetu], Primitive Germanic formula

ride: Old Norse [ˈriːða], Old English [ˈriːdan], Old Frisian [ˈriːda],
Old High German [ˈriːtan], Primitive Germanic formula *[ˈriːdanan].

(5) The next phoneme shows us a divergence in Gothic, which
is obviously due to later change: Gothic always has ar for the unstressed
er of the other languages, e.g.: Gothic [ˈhwaθar], Old
English [ˈhwɛðer] ‘which of the two.’

(6) The dialects agree as to the last phoneme, [r].

18. 6. While we have no written records to confirm our reconstructions
of Primitive Germanic, we occasionally get almost this
from the very ancient Scandinavian runic inscriptions (§ 17.6).
Take, for instance, the following reconstructions:

guest: Gothic [gasts], Old Norse [gestr], Old English, Old Frisian
[jest], Old Saxon, Old High German [gast], Primitive Germanic
formula *[ˈgastiz];

horn: all the old dialects [horn], Primitive Germanic formula

Here our Primitive Germanic reconstructions are longer than
the actually attested forms. Space forbids our entering into the
reasons that lead us to set up the additional phonemes; suffice it to
say that in most cases, as in guest, these additional phonemes are
made entirely definite by the forms in the actual dialects, while in
others, such as horn, the presence of additional phonemes in
Primitive Germanic is certain from the comparison of the Germanic
languages, although the nature of these phonemes is decided only
by the considerations which we now approach. I have chosen the
words guest and horn as examples because they occur in a runic
inscription on a golden horn, dating probably round 400 A.D.,
found near Gallehus in Denmark. Transliterated, the inscription

ek hlewagastiz holtiŋaz horna tawido305

‘I, Fame-Guest, the Holting (man of the family of Holt), made the
horn.’ The same words in our Primitive Germanic formulas, would
appear as *[ˈek ˈhlewa-ˌgastiz ˈholtingaz hornan ˈtawido:n], and
the inscription confirms the final syllable of our reconstruction of
guest, and the vowel, at any rate, of the final syllable in our reconstruction
of horn.

The Finnish, Esthonian, and Lappish languages, belonging to
the Finno-Ugrian family (§ 4.7) and therefore unrelated to ours,
contain many words which they must have borrowed from a
Germanic language at an ancient time — all evidence points to the
beginning of the Christian Era. As these languages have since
that time gone through entirely different changes than have the
Germanic languages, these borrowed forms give us independent
evidence as to the ancient form of Germanic words. Our reconstructions
of Primitive Germanic forms, like ring, Old English
[hring], Old Norse [hringr], as *[ˈhringaz], or king, Old English
[ˈkyning], as *[ˈkuningaz], or gold, Old English [gold] as *[ˈgolθan],
or yoke, Old English [jok], as *[ˈjokan], are confirmed by such
Finnish loan-words as rengas ‘ring,’ kuningas ‘king,’ kulta ‘gold,’
jukko ‘yoke.’

18. 7. The comparative method gives us an even more powerful
check upon our Primitive Germanic reconstructions. Since the
Germanic languages are a branch of the Indo-European family,
our Primitive Germanic forms enter as units into comparison
with forms of the other Indo-European languages. The reconstructed
forms of Primitive Indo-European give us a scheme of a
still earlier structure, out of which the Primitive Germanic structure
has grown.

Among our last examples there are two good instances. Our
reconstruction of Primitive Germanic *[ˈgastiz] ‘guest’ matches the
Latin form hostis ‘stranger.’ From the comparison of the Slavic
forms, Old Bulgarian [gosti], Russian [gost], and so on, we reconstruct
a Primitive Slavic *[˅gosti]; this, however, is under strong
suspicion of having been borrowed from a Germanic dialect and
must therefore stay out of account. The comparison of the Latin
form, however, leads us to set up a Primitive Indo-European
formula *[ghostis], which tells us, in shorthand fashion, that the
Latin second syllable confirms the final phonemes of our Primitive
Germanic formula.

Similarly, on the basis of Gothic [gaˈjuk] ‘pair’ and the other
306old Germanic forms of the word yoke, namely, Old Norse [ok], Old
English [jok], Old High German [jox], we set up a Primitive
Germanic formula *[ˈjokan], confirmed by the Finnish loan-form
jukko. The phonemes in the second syllable of this reconstructed
form would be in some respects indeterminate, were it not that
this formula enters in turn into comparison with other forms of the
Indo-European group. Sanskrit [juˈgam] leads us to set up a
Primitive Indo-Iranian *[juˈgam]. Further, we have Greek [zuˈgon]
and Latin [ˈjugum]. The Slavic forms, such as Old Bulgarian
[igo], Russian [ˈigo], lead us to set up a Primitive Slavic formula
*[ˈigo]. Cornish iou, Welsh iau, point to a Primitive Celtic *[ˈjugom].
Even languages which have reshaped our word, Lithuanian
[ˈjungas] and Armenian luc, give some evidence as to the structure
of the word in Primitive Indo-European. All of this evidence we
subsume in the formula, Primitive Indo-European *[ǰuˈgom].

The case of the word father shows us an inference of a more
complex character. Sanskrit [piˈta:], Greek [paˈte:r], Latin
[ˈpater], Old Irish [ˈaðir], Primitive Germanic *[ˈfader], are the
principal forms which lead us to set up the Primitive Indo-European
formula as *[pəˈte:r]. The initial phoneme here illustrates
the simplest case, a constant and normal set of correspondences:
initial [p] of the Indo-European languages in general is matched
by [f] in Germanic, and by zero in Celtic; Latin [ˈporkus] ‘pig,’
Lithuanian [˅paršas], corresponds to Primitive Germanic *[ˈfarhaz],
Old English [fearh] (modern farrow), and Old Irish [ork], and the
Primitive Indo-European formula is *[ˈporkos].

The second phoneme in our formula shows a more complex case.
In our Primitive Indo-European formulas we distinguish three
short-vowel phonemes, [a, o, ə], although no Indo-European language
has this threefold distinction. We do this because the
correspondences between the languages show three different combinations.
We use the symbol [a] in those cases where Indo-Iranian,
Greek, Latin, and Germanic agree in having [a], as in

acre: Sanskrit [ˈaǰrah], Greek [aˈgros], Latin [ˈager], Primitive
Germanic *[ˈakraz]: Primitive Indo-European formula *[agros].

We use the symbol [o] for the many cases where Indo-Iranian
and Germanic have [a], but Greek, Latin, and Celtic have [o], as in

eight: Sanskrit [ašˈtarw], Greek [okˈto:], Latin [ˈokto:], Primitive
Germanic *[ˈahtaw], Gothic [ˈahtaw], Old German [ˈahto]:
Primitive Indo-European formula *[okˈto:w].307

We use the symbol [ə] for the cases where Indo-Iranian has [i],
while the other languages have the same phoneme as in the forms
of the first set:

stead: Sanskrit [ˈsthitih] ‘a standing,’ Greek [ˈstasis], Primitive
Germanic *[ˈstadiz], Gothic [staθs], Old High German [stat]:
Primitive Indo-European formula *[sthətis].

Evidently the forms of the word father show this last type of
correspondence; hence we use [ə] in our formula. The morphologic
structure of Primitive Indo-European, as it appears in the totality
of our formulas, confirms our threefold distinction [a, o, ə], in that
these three units take part in three different types of morphologic

The third symbol in our formula, which is the last we shall consider,
illustrates a very interesting type of inference. Ordinarily
when the other Indo-European languages have a [t], the Germanic
languages have a [θ]. Thus,

brother: Sanskrit [ˈbhra:ta:], Greek [ˈphra:te:r] (‘member of a
phratry’), Latin [ˈfra:ter], Old Bulgarian [bratru], Primitive
Germanic *[ˈbro:θer], Gothic [ˈbro:θar], Old Norse [ˈbro:ðer],
Old English [ˈbro:ðor], Old High German [ˈbruoder]: Primitive
Indo-European formula *[ˈbhra:te:r];

three: Sanskrit [ˈtrajah], Greek [˅trejs], Latin [tre:s], Old Bulgarian
[trije], Primitive Germanic *[θriːz], Old Norse [θriːr],
Old High German [driː]: Primitive Indo-European formula

The word father, together with some others, is anomalous in
Primitive Germanic in containing [d] instead of [θ]. One might,
of course, assume that two distinct Primitive Indo-European
phonemes were here involved, which had coincided as [t] in all
the Indo-European languages except Germanic, which alone distinguished
them as [θ] versus [d]. In 1876, however, Karl Verner
(1846-1896), a Danish linguist, showed that in a number of the
cases where Germanic has the troublesome [d], this consonant
follows upon a vowel or diphthong which is unstressed in Sanskrit
and Greek; this correlation occurs in enough instances, and, in the
morphologic structure, systematically enough, to exclude the factor
of accident. The contrast of the words brother and father illustrates
this correlation. Since the place of the word-accent is determined
by the primary phonemes in Italic, Celtic, and Germanic, we can
easily believe that its position in each of these languages is due to
308later change. Sanskrit and Greek, moreover, agree so often, although
the place of the accent in both is highly irregular, that we
do not hesitate to attribute this feature to the parent language.
We thus face a definite succession of events in the period between
Primitive Indo-European and Primitive Germanic — a period to
which we give the name pre-Germanic:

Primitive Indo-European: [t] a unit phoneme; word-accent on
different syllables in different words:

*[ˈbhra:te:r] ‘brother’ *[pəˈte:r] ‘father’

Pre-Germanic period:

first change: [t] becomes [θ]:

*[ˈbra:θe:r] *[faˈθe:r]

second change: [θ] after unstressed syllabic becomes [d], presumably
a voiced spirant:

*[ˈbra:θe:r] *[faˈde:r]

third change: the accent is shifted to the first syllable of each
word; this brings us to

Primitive Germanic *[ˈbro:θer] *[ˈfader].

In a similar way, the correspondences reveal the pre-history of
each branch of the Indo-European family. Thus, in the case of
Latin cauda and cōda ‘tail,’ the Lithuanian word [˅kuodas] ‘tuft’
probably represents the same form of the parent speech; if so, then,
in the light of other correspondences, in which Lithuanian [uo]
and Latin [o:] appear side by side, we may take cōda to be the
older of the two Latin forms, and cauda to be a hyper-urban
(over-elegant) variant (§ 18.4).

Our Primitive Indo-European reconstructions are not subject
to any check by means of earlier recorded or reconstructed forms.
In the last decades, to be sure, it has been ascertained that the
Hittite language, known to us from records in cuneiform writing
from 1400 B.C. onward, is distantly related to Indo-European.
Accordingly, it has been possible to uncover a few features of a
Primitive Indo-Hittite parent language — that is, to trace the
earlier history of a few of the features of Primitive Indo-European.

18. 8. The comparative method tells us, in principle, nothing
about the acoustic shape of reconstructed forms; it identifies the
phonemes in reconstructed forms merely as recurrent units. The
Indonesian languages show us a striking example of this. Each
language has only a few phonemes of the types [d, g, l, r], but the
variety of the correspondences assures us of a larger number of
309phonemes in the parent language. The acoustic character of these
phonemes can only be guessed at; the symbols by which we represent
them are merely labels for correspondences. It is worth
noticing that we have older written records for none of these languages
except Javanese; this in no way affects the application of
the comparative method. The eight normal types of correspondence
will appear sufficiently if we consider three languages: Tagalog
(on the island of Luzon in the Philippines), Javanese, and
Batak (on the island of Sumatra). In the following examples the
consonant under discussion appears in the middle of the word.

tableau Tagalog | Javanese | Batak | Primitive Indonesian | ‘choose’ | ‘lack’ | ‘nose’ | ‘desire’ | ‘point out’ | ‘spur’ | ‘sago’ | ‘addled’ 127 228

18. 9. The comparative method assumes that each branch or
language bears independent witness to the forms of the parent
language, and that identities or correspondences among the related
languages reveal features of the parent speech. This is the
same thing as assuming, firstly, that the parent community was
completely uniform as to language, and, secondly, that this parent
community split suddenly and sharply into two or more daughter
communities, which lost all contact with each other.310

Often enough, the comparative method assumes successive
splittings of this sort in the history of a language. It assumes that
Germanic split off neatly from Primitive Indo-European. After
this split, any change in Germanic was independent of changes in
the sister languages, and any resemblance between Germanic and
the sister languages betokens a common inheritance. The differences
between Primitive Indo-European and Primitive Germanic
are due to changes which occurred during the pre-Germanic period.
In exactly the same way, the comparative method- interprets
the special similarities among the West Germanic languages
(in contrast with Scandinavian and Gothic) by saying that a
West Germanic community split off, neatly and suddenly, from
the uniform Primitive Germanic parent community. After this
splitting off comes a pre-West-Germanic period, during which
there arose the differences that characterize Primitive West
Germanic. Again, on the basis of peculiarities common to English
and Frisian (such as, especially, the [ɛ, e] for Primitive West
Germanic [a], which we noticed above), we may speak of a pre-Anglo-Frisian
period, during which there occurred the changes
which led to Primitive Anglo-Frisian. Upon this there followed
a pre-English period, which leads to the forms that appear in our
earliest records of English. Thus, the comparative method reconstructs
uniform parent languages existing at points in time,
and deduces the changes which took place after each such parent
language split, up to the next following parent language or recorded
language. The comparative method thus shows us the ancestry
of languages in the form of a family-tree, with successive branchings:
the points at which branches separate are designated by
the word primitive; the branches between the points are designated
by the prefix pre-, and represent periods of linguistic change
(Figure 1).

18. 10. The earlier students of Indo-European did not realize
that the family-tree diagram was merely a statement of their
method; they accepted the uniform parent languages and their
sudden and clear-cut splitting, as historical realities.

In actual observation, however, no speech-community is ever
quite uniform (§ 3.3). When we describe a language, we may ignore
the lack of uniformity by confining ourselves to some arbitrarily
chosen type of speech and leaving the other varieties for
later discussion, but in studying linguistic change we cannot do
311this, because all changes are sure to appear at first in the shape
of variant features.

image Germanic | Celtic | Italic | Greek | Albanese | Armenian | Indic | Iranian | Slavic | Baltic | Primite Indo-European

image English (actual records) | pre-English period | Primitive Anglo-Frisian | pre-Anglo-Frisian period | Primitive West Germanic | pre-West Germanic period | Primitive Germanic | pre-Germanic period | Primitive Indo-European

Figure 1. (Above) Family-tree diagram of the relationship of the Indo-European
languages. (Below) Part of a family-tree diagram, showing the
epochs in the history of English.

At times, to be sure, history shows us a sudden cleavage, such
as is assumed by the comparative method. A cleavage of this
sort occurs when part of a community emigrates. After the Angles,
Saxons, and Jutes settled in Britain, they were fairly well cut
off from their fellows who remained on the Continent; from that
312time on, the English language developed independently, and any
resemblance between English and the continental dialects of
West Germanic can be taken, in the ordinary case, as evidence
for a feature that existed before the emigration of the English.
When the Gipsies, in the Middle Ages, started from northwestern
India on their endless migration, the changes in their language,
from that time on, must have been independent of whatever linguistic
changes occurred in their former home.

A less common case of clear-cut division of a speech-community,


Figure 2. Eastern Europe: the splitting of speech-areas by invasion.
Latin, once a unit, was split, in the early Middle Ages, by the intrusion of
Slavic. In the ninth century this area, in turn, was split by the intrusion
of Hungarian.313

is splitting by the intrusion of a foreign community. Under the
Roman Empire, Latin was spoken over a solid area from Italy
to the Black Sea. In the early Middle Ages, Slavs came in from
the north and settled so as to cut this area completely in two:
since that time, the development of Roumanian, in the east, has
gone on independently of the development of the other Romance
languages, and a feature common to both Roumanian and the
western Romance languages is presumably guaranteed as Latin.
In the ninth century, the great Slavic area in turn suffered a
similar split, for the Magyars (Hungarians), coming from the
east, settled so as to cut the Slavic area into a northern and
a southern part (see Figure 2). Since that time, accordingly, the
changes in South Slavic (Slovene, Serbian, Bulgarian) have been independent
of those in the northern area of Slavic, and any common
features of the two areas presumably date from before the split.

Such clear-cut splitting, however, is not usual. The differences
among the Romance languages of the western area are evidently
not due to geographic separation or to the intrusion of foreign
speech-communities. Aside from English and from Icelandic,
the same holds good of the Germanic languages, including the
sharply defined difference between West Germanic and Scandinavian,
which border on each other in the Jutland peninsula. Evidently
some other historical factor or factors beside sudden separation
may create several speech-communities out of one, and in
this case we have no guarantee that all changes after a certain
moment are independent, and therefore no guarantee that features
common to the daughter languages were present in the parent
language. A feature common, let us say, to French and Italian,
or to Dutch-German and Danish, may be due to a common change
which occurred after some of the differences were already in

18. 11. Since the comparative method does not allow for varieties
within the parent language or for common changes in related
languages, it will carry us only a certain distance. Suppose,
for instance, that within the parent language there was some dialectal
difference: this dialectal difference will be reflected as an irreconcilable
difference in the related languages. Thus, certain of
the inflectional suffixes of nouns contain an [m] in Germanic and
Balto-Slavic, but a [bh] in the other Indo-European languages, and
there is no parallel for any such phonetic correspondence.314

(a) Primitive Indo-European *[-mis], instrumental plural:
Gothic [ˈwulfam] ‘to, by wolves,’

Primitive Indo-European *[-miːs], instrumental plural: Lithuanian
[naktiˈmis] ‘by nights,’ Old Bulgarian [noštimi],

Primitive Indo-European *[-mos], dative-ablative plural: Lithuanian
[vilˈkams] ‘to wolves,’ Old Bulgarian [vl̩komu],

(b) Primitive Indo-European *[-bhis], instrumental plural: Sanskrit
[padˈbhih] ‘by feet,’ Old Irish [ˈferav] ‘by men,’

Primitive Indo-European *[-bhjos], dative-ablative plural: Sanskrit
[padˈbhjah] ‘to, from the feet,’

Primitive Indo-European *[-bhos], dative-ablative plural: Latin
[ˈpedibus] ‘to, from the feet,’ Old Celtic [ma:trebo] ‘to the mothers.’

In cases like these, the comparative method does not show us
the form of the parent speech (which is defined as a uniform language),
but shows us irreconcilably different forms, whose relation,
as alternants or as dialectal variants, it does not reveal. Yet these
cases are very many.

On the other hand, if, like the older scholars, we insist that the
discrepancy is due to a common change in the history of Germanic
and Balto-Slavic, then, under the assumptions of the comparative
method, we must say that these two branches had a period of
common development: we must postulate a Primitive Balto-Slavo-Germanic
speech-community, which split off from Primitive
Indo-European, and in turn split into Germanic and Balto-Slavic.
If we do this, however, we are at once involved in contradictions,
because of other, discordant but overlapping, resemblances. Thus,
Balto-Slavic agrees with Indo-Iranian, Armenian, and Albanese,
in showing sibilants in certain forms where the other languages
have velars, as in the word for ‘hundred’ :

Sanskrit [çaˈtam], Avestan [satəm], Lithuanian [˅šimtas], but
Greek [he-kaˈton], Latin [ˈkentum], Old Irish [ke:ð], Primitive
Indo-European *[km̩ˈtom]. We suppose that the parent language
in such cases had palatalized velar stops.

Likewise, where the four branches just named have velar stops,
there the others, in many forms, have combinations of velars with
a labial element, or apparent modifications of these; we suppose
that the parent language had labialized velar stops, as in the interrogative
substitute stem:

Sanskrit [kah] ‘who?’ Lithuanian [kas], Old Bulgarian [ku-to],
315but Greek [ˈpo-then] ‘from where?’ Latin [kwo:] ‘by whom, by
what?’ Gothic [hwas] ‘who?’ Primitive Indo-European *[kwos]
‘who?’ and derivatives.

Only in a limited number of cases do the two sets of languages
agree in having plain velar stops. Accordingly, many scholars
suppose that the earliest traceable division of the Primitive Indo-European
unity was into a western group of so-called “centum-languages”
and an eastern group of “satem-languages,” although,
to be sure, Tocharian, in Central Asia, belonged to the former
group. This division, it will be seen, clashes with any explanation
that supposes Balto-Slavic and Germanic to have had a common
period of special development.


Figure 3. Some overlapping features of special resemblance among the
Indo-European languages, conflicting with the family-tree diagram. —
Adapted from Schrader.

1. Sibilants for velars in certain forms.

2. Case-endings with [m| for [bh].

3. Passive-voice endings with [r].

4. Prefix [ˈe-] in past tenses.

5. Feminine nouns with masculine suffixes.

6. Perfect tense used as general past tense.

Again, we find special resemblances between Germanic and
Italic, as, for instance, in the formation and use of the past-tense
verb, or in some features of vocabulary (goat : Latin haedus;
Gothic gamains : Latin commūnis ‘common’). These, too, conflict
with the special resemblances between Germanic and Balto-Slavic.
In the same way, Italic on the one side shares peculiarities
with Celtic and on the other side with Greek (Figure 3).316

18. 12. As more and more of these resemblances were revealed,
the older scholars, who insisted upon the family-tree diagram, faced
an insoluble problem. Whichever special resemblances one took as
evidence for closer relationships, there remained others, inconsistent
with these, which could be explained only by an entirely
different diagram. The decision, moreover, was too important to
be evaded, since in each case it profoundly altered the value of
resemblances. If Germanic and Balto-Slavic, for instance, have
passed through a period of common development, then any agreement
between them guarantees nothing about Primitive Indo-European,
but if they have not passed through a period of common development,
then such an agreement, on the family-tree principle, is
practically certain evidence for a trait of Primitive Indo-European.

The reason for these contradictions was pointed out in 1872
by Johannes Schmidt (1843-1901), in a famous essay on the interrelationship
of the Indo-European languages. Schmidt showed
that special resemblances can be found for any two branches of
Indo-European, and that these special resemblances are most
numerous in the case of branches which lie geographically nearest
each other. Johannes Schmidt accounted for this by the so-called
wave-hypothesis. Different linguistic changes may spread, like
waves, over a speech-area, and each change may be carried out
over a part of the area that does not coincide with the part covered
by an earlier change. The result of successive waves will be a network
of isoglosses (§ 3.6). Adjacent districts will resemble each
other most; in whatever direction one travels, differences will
increase with distance, as one crosses more and more isogloss-lines.
This, indeed, is the picture presented by the local dialects in the
areas we can observe. Now, let us suppose that among a series of
adjacent dialects, which, to consider only one dimension, we shall
designate as A, B, C, D, E, F, G,… X, one dialect, say F, gains
a political, commercial, or other predominance of some sort, so
that its neighbors in either direction, first E and G, then D and H,
and then even C and I, J, K, give up their peculiarities and in time
come to speak only the central dialect F. When this has happened,
F borders on B and L, dialects from which it differs sharply enough
to produce clear-cut language boundaries; yet the resemblance
between F and B will be greater than that between F and A, and,
similarly, among L, M, N,… X, the dialects nearest to F will
show a greater resemblance to F, in spite of the clearly marked
317boundary, than will the more distant dialects. The presentation of
these factors became known as the wave-theory, in contradistinction
to the older family-tree theory of linguistic relationship. Today we
view the wave process and the splitting process merely as two
types — perhaps the principal types — of historical processes that
lead to linguistic differentiation.

18. 13. The comparative method, then, — our only method for
the reconstruction of prehistoric language, — would work accurately
for absolutely uniform speech-communities and sudden,
sharp cleavages. Since these presuppositions are never fully realized,
the comparative method cannot claim to picture the historical
process. Where the reconstruction works smoothly, as in the Indo-European
word for father, or in observations of less ambitious
scope (such as, say, reconstructions of Primitive Romance or
Primitive Germanic), there we are assured of the structural features
of a speech-form in the parent language. Wherever the comparison
is at all ambitious as to the reach of time or the breadth of
the area, it will reveal incommensurable forms and partial similarities
that cannot be reconciled with the family-tree diagram. The
comparative method can work only on the assumption of a uniform
parent language, but the incommensurable forms (such as *[-mis]
and *[-bhis] as instrumental plural case endings in Primitive Indo-European)
show us that this assumption is not justified. The comparative
method presupposes clear-cut splitting off of successive
branches, but the inconsistent partial similarities show us that
later changes may spread across the isoglosses left by earlier
changes; that resemblance between neighboring languages may be
due to the disappearance of intermediate dialects (wave-theory);
and that languages already in some respects differentiated may
make like changes.

Sometimes additional facts help us to a decision. Thus, the
adjective Sanskrit [ˈpiːva:] ‘fat,’ Greek [ˈpiːo:n] occurs only in
Indo-Iranian and Greek, but its existence in Primitive Indo-European
is guaranteed by the irregular formation of the feminine
form, Sanskrit [ˈpirvariː], Greek [ˈpiːejra]; neither language
formed new feminines in this way. On the other hand, the Germanic
word hemp, Old English [ˈhɛnep], Middle Dutch [ˈhannep],
and so on, corresponds to Greek [ˈkannabis]; nevertheless, we learn
from Herodotus (fifth century B.C.) that hemp was known to the
Greeks only as a foreign plant, in Thrace and Scythia: the word
318came into Greek (and thence into Latin) and into Germanic (and
thence, presumably, into Slavic) from some other language — very
likely from a Finno-Ugrian dialect — at some time before the pre-Germanic
changes of [k] to [h] and of [b] to [p]. But for this piece
of chance information, the correspondence of the Greek and
Germanic forms would have led us to attribute this word to
Primitive Indo-European.

18. 14. The reconstruction of ancient speech-forms throws some
light upon non-linguistic conditions of early times. If we consider,
for instance, that the composition of our earliest Indic records can
scarcely be placed later than 1200 B.C., or that of the Homeric
poems later than 800 B.C., we are bound to place our reconstructed
Primitive Indo-European forms at least a thousand years earlier
than these dates. We can thus trace the history of language, often
in minute detail, much farther back than that of any other of a
people's institutions. Unfortunately, we cannot transfer our
knowledge to the latter field, especially as the meanings of speech-forms
are largely uncertain. We do not know where Primitive
Indo-European was spoken, or by what manner of people; we
cannot link the Primitive Indo-European speech-forms to any
particular type of prehistoric objects.

The noun and the verb snow appear so generally in the Indo-European
languages that we can exclude India from the range of
possible dwellings of the Primitive Indo-European community.
The names of plants, even where there is phonetic agreement,
differ as to meaning; thus, Latin [ˈfa:gus], Old English [bo:k] mean
‘beech-tree,’ but Greek [phe:ˈgos] means a kind of oak. Similar
divergences of meaning appear in other plant-names, such as our
words tree, birch, withe (German Weide ‘willow’), oak, corn, and
the types of Latin salix ‘willow,’ quercus ‘oak,’ hordeum ‘barley’
(cognate with German Gerste), Sanskrit [ˈjavah] ‘barley.’ The
type of Latin glans ‘ acorn’ occurs with the same meaning in Greek,
Armenian, and Balto-Slavic.

Among animal-names, cow, Sanskrit [ga:wh], Greek [˅bows],
Latin [bo:s], Old Irish [bo:], is uniformly attested and guaranteed
by irregularities of form. Other designations of animals appear in
only part of the territory; thus, goat, as we have seen, is confined to
Germanic and Italic; the type Latin caper: Old Norse hafr ‘goat’
occurs also in Celtic; the type Sanskrit [aˈǰah], Lithuanian [o˅žiːs]
is confined to these two languages; and the type of Greek [ˈajks]
319appears also in Armenian and perhaps in Iranian. Other animals
for which we have one or more equations covering part of the
Indo-European territory, are horse, dog, sheep (the word wool
is certainly of Primitive Indo-European age), pig, wolf, bear, stag,
otter, beaver, goose, duck, thrush, crane, eagle, fly, bee (with
mead, which originally meant ‘honey’), snake, worm, fish. The
types of our milk and of Latin lac ‘milk’ are fairly widespread, as
are the word yoke and the types of our wheel and German Rad
‘wheel,’ and of axle. We may conclude that cattle were domesticated
and the wagon in use, but the other animal-names do not
guarantee domestication.

Verbs for weaving, sewing, and other processes of work are
widespread, but vague or variable in meaning. The numbers
apparently included ‘hundred’ but not ‘thousand.’ Among terms
of relationship, those for a woman's relatives by marriage (‘husband's
brother,’ ‘husband's sister,’ and so on) show widespread
agreement, but not those for a man's relatives by marriage; one
concludes that the wife became part of the husband's family,
which lived in a large patriarchal group. The various languages
furnish several equations for names of tools and for the metals gold,
silver, and bronze (or copper). Several of these, however, are loanwords
of the type of hemp; so certainly Greek [ˈpelekus] ‘axe,’
Sanskrit [paraˈçuh] is connected with Assyrian [pilakku], and our
axe and silver are ancient loan-words. Accordingly, scholars place
the Primitive Indo-European community into the Late Stone Age.320

Chapter 19
Dialect Geography

19. 1. The comparative method, with its assumption of uniform
parent languages and sudden, definitive cleavage, has the virtue
of showing up a residue of forms that cannot be explained on this
assumption. The conflicting large-scale isoglosses in the Indo-European
area, for instance, show us that the branches of the
Indo-European family did not arise by the sudden breaking up of
an absolutely uniform parent community (§ 18.11, Figure 3).
We may say that the parent community was dialectally differentiated
before the break-up, or that after the break-up various
sets of the daughter communities remained in communication;
both statements amount to saying that areas or parts of areas
which already differ in some respects may still make changes in
common. The result of successive changes, therefore, is a network
of isoglosses over the total area. Accordingly, the study of local
differentiations in a speech-area, dialect geography, supplements
the use of the comparative method.

Local differences of speech within an area have never escaped
notice, but their significance has only of late been appreciated.
The eighteenth-century grammarians believed that the literary
and upper-class standard language was older and more true to a
standard of reason than the local speech-forms, which were due
to the ignorance and carelessness of common people. Nevertheless,
one noticed, in time, that local dialects preserved one or another
ancient feature which no longer existed in the standard language.
Toward the end of the eighteenth century there began to appear
dialect dictionaries, which set forth the lexical peculiarities of nonstandard

The progress of historical linguistics showed that the standard
language was by no means the oldest type, but had arisen, under
particular historical conditions, from local dialects. Standard
English, for instance, is the modern form not of literary Old
English, but of the old local dialect of London which had become
first a provincial and then a national standard language, absorbing,
321meanwhile, a good many forms from other local and provincial
dialects. Opinion now turned to the other extreme. Because a
local dialect preserved some forms that were extinct in the standard
language, it was viewed as a survival, unchanged, of some
ancient type; thus, we still hear it said that the speech of some
remote locality is “pure Elizabethan English.” Because the admixture
of forms from other dialects had been observed only in the
standard language, one jumped at the conclusion that local dialects
were free from this admixture and, therefore, in a historical sense,
more regular. At this stage, accordingly, we find dialect grammars,
which show the relation of the sounds and inflections of a local
dialect to those of some older stage of the language.

Investigation showed that every language had in many of its
forms suffered displacements of structure, which were due to the
admixture of forms from other dialects. Old English [f], for instance,
normally appears as [f] in standard English, as in father,
foot, fill, five, and so on, but in the words vat and vixen, from Old
English [fɛt] and [ˈfyksen] ‘female fox,’ it appears as [v], evidently
because these forms are admixtures from a dialect which had
changed initial [f] to [v]; and, indeed, this initial [v] appears
regularly in some southern English dialects (Wiltshire, Dorset,
Somerset, Devon), in forms like [ˈvaðə, vut, vil, vajv]. Some
students hoped, therefore, to find in local dialects the phonemic
regularity (that is, adherence to older patterns) that was broken
in the standard language. In 1876 a German scholar, Georg
Wenker, began, with this end in view, to survey the local dialects
in the Rhine country round Düsseldorf: later he extended his
survey to cover a wider area, and published, in 1881, six maps as
a first instalment of a dialect atlas of northern and central Germany.
He then gave up this plan in favor of a survey which was to
cover the whole German Empire. With government aid, Wenker
got forty test-sentences translated, largely by schoolmasters, into
more than forty-thousand German local dialects. Thus it was
possible to mark the different local varieties of any one feature on
a map, which would then show the geographic distribution. Since
1926 these maps, on a reduced scale, have been appearing in print,
under the editorship of F. Wrede.

The result, apparent from the very start, of Wenker's study, was
a surprise: the local dialects were no more consistent than the
standard language in their relation to older speech-forms. Dialect
322geography only confirmed the conclusion of comparative study,
namely, that different linguistic changes cover different portions
of an area. The new approach yielded, however, a close-range view
of the network of isoglosses.

19. 2. At present, then, we have three principal forms of dialect
study. The oldest is lexical. At first, the dialect dictionaries
included only the forms and meanings which differed from standard
usage. This criterion, of course, is irrelevant. Today we expect a
dictionary of a local dialect to give all the words that are current
in non-standard speech, with phonetic accuracy and with reasonable
care in the definition of meanings. A dialect dictionary for a
whole province or area is a much bigger undertaking. It should
give a phonemic scheme for each local type of speech, and therefore
can hardly be separated from a phonologic study. We expect
a statement of the geographic area in which every form is
current, but this statement can be given far better in the form
of a map.

Grammars of local dialects largely confine themselves to stating
the correspondence of the phonemes and of the inflectional forms
with those of an older stage of the language. The modern demand
would be rather for a description such as one might make of any
language: phonology, syntax, and morphology, together with
copious texts. The history of the forms can be told only in connection
with that of the area as a whole, since every feature has
been changed or spared only in so far as some wave of change has
reached or failed to reach the speakers of the local dialect. The
grammar of a whole area represents, again, a large undertaking.
The first work of this kind, the single-handed performance of a
man of the people, was the Bavarian grammar, published in 1821,
of Johann Andreas Schmeller (1785-1852); it is still unsurpassed.
For English, we have the phonology of the English dialects in the
fifth volume of Ellis's Early English Pronunciation, and Joseph
Wright's grammar, published in connection with his English
Dialed Dictionary
. Here too, of course, we demand a statement of
the topographic extent of each feature, and this, again, can be
more clearly given on a map.

Except for the complete and organized description of a single
local dialect, then, the map of distribution is the clearest and most
compact form of statement. The dialect atlas, a set of such maps,
allows us to compare the distributions of different features by
323comparing the different maps; as a practical help for this comparison,
the German atlas provides with each map a loose transparent
sheet reproducing the principal isoglosses or other marks
of the map. Aside from the self-understood demands of accuracy
and consistency, the value of a map depends very largely on the
completeness with which the local dialects are registered: the
finer the network, the more complete is the tale. In order to record
and estimate a local form, however, we need to know its structural
pattern in terms of the phonemic system of the local dialect.
Furthermore, several variant pronunciations or grammatical or
lexical types may be current, with or without a difference of
denotation, in a local dialect, and these variants may be decidedly
relevant to the history of the change which produced them. Finally,
to reproduce the whole grammar and lexicon would require so
vast a number of maps that even a very large atlas can only give
samples of distribution; we ask for as many maps as possible. In
view of all this, a dialect atlas is a tremendous undertaking, and in
practice is likely to fall short in one or another respect. The
sentences on which the German atlas is based, were written down
in ordinary German orthography by schoolmasters and other
linguistically untrained persons; the material does not extend to
great parts of the Dutch-German area, such as the Netherlands
and Belgium, Switzerland, Austria, Baltic German, Yiddish, Transylvanian,
and the other speech-islands. The data are largely
phonologic, since the informant, except for striking lexical or
grammatical differences, would merely transcribe the forms into
a spelling that represented the local pronunciation; yet the phonologic
aspect is precisely what will be least clear in such a transcription.
The data for the French atlas were collected by a trained
phonetician, Edmond Edmont; one man, of course, could visit
only a limited number of localities and stay but a short time in
each. Accordingly, the maps register only something over six-hundred
points in the French area (France and adjoining strips
in Belgium, Switzerland, and Italy), and the forms were collected
in each case from a single informant by means of a questionnaire
of some two-thousand words and phrases. However fine his ear,
Edmont could not know the phonologic pattern of each local
dialect. The results for both phonetics and lexicon are more
copious than those of the German atlas, but the looseness of the
network and the lack of whole sentences are drawbacks. The atlas
324itself was planned and worked out by Jules Gilliéron (1854-1926),
and has appeared in full (1896-1908), together with a supplement
for Corsica. An Italian atlas, by K. Jaberg and J. Jud, has
been appearing since 1928; it tries for great accuracy and pays
close attention to meanings. Smaller atlases exist for Swabia
(by H. Fischer, 28 maps, published, in connection with a careful
treatise, in 1895), for Denmark (by V. Bennicke and M. Kristensen,
1898-1912), for Roumania (by G. Weigand, 1909), for Catalonia
(by A. Griera, 1923 ff.), and for Brittany (by P. Le Roux, 1924 ff.).
Other atlases are in preparation, including a survey of New England
under the direction of H. Kurath. A single-handed observer can
cover a small part of an area, as did Karl Haag in his study of a
district in Southern Swabia (1898); or else, he may restrict himself
to one or two features but follow them over a larger district, as
did G. G. Kloeke in his study of the vowel phonemes of the words
mouse and house in the Netherlands and Belgium (1927).

Needless to say, the map or atlas may be accompanied by a
treatise that interprets the facts or accounts for their origin, as
in the publications of Fischer, Haag, and Kloeke. The great
atlases have given rise to many studies, such as, notably, Gilliéron;s
various books and essays, based on the French atlas, and
a whole series of studies, under the editorship of F. Wrede, by
workers on the German maps.

19. 3. Our knowledge is confined, so far, to the conditions that
prevail in long-settled areas. In these, there is no question of
uniformity over any sizable district. Every village, or, at most,
every cluster of two or three villages, has its local peculiarities of
speech. In general, it presents a unique combination of forms, each
of which also appears, in other combinations, in some of the neighboring
localities. On the map, accordingly, each settlement or
small cluster of settlements will be cut off from each of its neighbors
by one or more isoglosses. As an example, Figure 4, reproducing
a small portion of Haag's map, shows the Swabian village of
Bubsheim (about ten miles east by southeast of Rottweil). The
nearest neighbors, within a distance of less than five miles, are all
separated from Bubsheim by isoglosses; only two of these neighbors
agree with each other as to all of the features that were
studied by Haag. The appended table (Figure 5) shows under the
name of each locality, the forms in which its dialect differs from
the forms of Bubsheim, which are given in the first column; where
325no form is given, the dialect agrees with Bubsheim. The number
before each form is the same as the number attached to the corresponding
isogloss in Figure 4.


Figure 4. Isoglosses around the German village of Bubsheim (Swabia),
after Haag. The village of Denkingen has been added, with a few of its isoglosses,
in order to show the recurrence of Line 6.

If we followed the further course of these isoglosses, we should
find them running in various directions and dividing the territory
into portions of differing size. The isoglosses numbered 1, 2, and
3 in our Figures, cut boldly across the German area; Bubsheim
agrees, as to these features, with the south and southwest. In
contrast with these important lines, others, such as our number 9,
surround only a small district: the form [ˈtruːnke] ‘drunk,’ which
is listed for Denkingen, is spoken only in a small patch of settlements.
The isogloss we have numbered as 6 appears on our map
as two lines; these are really parts of an irregularly winding line:
Denkingen agrees with Bubsheim as to the vowel of the verb
mow, although the intermediate villages speak differently. We
find even isoglosses which divide a town into two parts; thus,
along the lower Rhine, just southwest of Duisburg, the town of
Kaldenhausen is cut through by a bundle of isoglosses: the eastern
and western portions of the town speak different dialects.

The reason for this intense local differentiation is evidently to
be sought in the principle of density (§ 3.4). Every speaker is
constantly adapting his speech-habits to those of his interlocutors;326

tableau Bubsheim | Reichenbach, Egersheim | Königsheim | Mahlstetten | Böttingen | Denkingen | Gosheim | Wehingen | ‘stove’ | ‘up’ | ‘time’ | ‘bean’ | ‘end’ | ‘to mow’ | ‘color’ | ‘old’ | ‘drunk’ | ‘to go’

Figure 5. Ten speech-forms in the local dialect of Bubsheim in Swabia, with the divergent forms of neighboring dialects.
Where no form is given, the dialect agrees with Bubsheim. The numbers are those of the isoglosses on the map, Figure 4. — After

he gives up forms he has been using, adopts new ones, and, perhaps
oftenest of all, changes the frequency of speech-forms without
entirely abandoning any old ones or accepting any that are really
new to him. The inhabitants of a settlement, village, or town,
however, talk much more to each other than to persons who live
elsewhere. When any innovation in the way of speaking spreads
over a district, the limit of this spread is sure to be along some line
of weakness in the network of oral communication, and these lines
of weakness, in so far as they are topographical lines, are the
boundaries between towns, villages, and settlements.

image The netherlands | [muːs, huːs] | [muːs, hy:s] | [my:s, hy:s] | [mø:s, hø:s] | [møys, høys]

Figure 6. Distribution of syllabic sounds in the words mouse and house in
the Netherlands. — After Kloeke.

19. 4. Isoglosses for different forms rarely coincide along their
whole extent. Almost every feature of phonetics, lexicon, or
grammar has its own area of prevalence — is bounded by its own
isogloss. The obvious conclusion has been well stated in the form
of a maxim: Every word has its own history.328

The words mouse and house had in early Germanic the same
vowel phoneme, a long [uː]. Some modern dialects — for instance,
some Scotch dialects of English — preserve this sound apparently
unchanged. Others have changed it, but keep the ancient structure,
in the sense that these two words still have the same syllabic
phoneme; this is the case in standard English and in standard
German, where both words have [aw], and in standard Dutch,
where both have [øɥ]. In the study above referred to, Kloeke
traces the syllabics of these two words through the present-day
local dialects of Belgium and the Netherlands. Our Figure 6 shows
Kloeke's map on a reduced scale.

An eastern area, as the map shows, has preserved the Primitive
Germanic vowel [uː] in both words: [muːs, huːs].

Several patches, of various size, speak [y:] in both words:
[my:s, hy:s].

A district in the extreme west speaks [ø:] in both words: [mø:s,

A great central area speaks a diphthong of the type [øɥ] in both
words: [møɥs, høɥs]. Since this is the standard Dutch-Flemish
pronunciation, it prevails in the usage of standard speakers also
in the other districts, but this fact is not indicated on the map.

In these last three districts, then, the sound is no longer that
of Primitive Germanic and medieval Dutch, but the structure of
our two words is unchanged, in so far as they still agree in their
syllabic phoneme.

Our map shows, however, three fair-sized districts which speak
[uː] in the word mouse, but [y:] in the word house; hence, inconsistently,
[muːs, hy:s]. In these districts the structural relation of
the two words has undergone a change: they no longer agree as to
their syllabic phoneme.

We see, then, that the isogloss which separates [muːs] from
[my:s] does not coincide with the isogloss which separates [huːs]
from [hy:s]. Of the two words, mouse has preserved the ancient
vowel over a larger territory than house. Doubtless a study of other
words which contained [uː] in medieval times, would show us still
other distributions of [uː] and the thher sounds, distributions
which would agree only in part with those of mouse and house.

At some time in the Middle Ages, the habit of pronouncing [y:]
instead of the hitherto prevalent [uː] must have originated in some
cultural center — perhaps in Flanders — and spread from there
329over a large part of the area on our map, including the central
district which today speaks a diphthong. On the coast at the
north of the Frisian area there is a Dutch-speaking district known
as het Bilt, which was diked in and settled under the leadership of
Hollanders at the beginning of the sixteenth century, and, as the
map shows, uses the [y:]-pronunciation. It is [y:], moreover, and
not the old [uː], that appears in the loan-words which in the early
modern period passed from Dutch into the more easterly (Low
German) dialects of the Dutch-German area, and into foreign
languages, such as Russian and Javanese. The Dutch that was
carried to the colonies, such as the Creole Dutch of the Virgin
Islands, spoke [y:]. The spellings in written documents and the
evidence of poets' rimes confirm this: the [y ^-pronunciation spread
abroad with the cultural prestige of the great coastal cities of
Holland in the sixteenth and seventeenth centuries.

This wave of cultural expansion was checked in the eastern part
of our district, where it conflicted with the expansion of another
and similar cultural area, that of the North German Hanseatic
cities. Our isoglosses of mouse and house, and doubtless many
others, are results of the varying balance of these two cultural
forces. Whoever was impressed by the Hollandish official or
merchant, learned to speak [y:]; whoever saw his superiors in the
Hanseatic upper class, retained the old [uː]. The part of the population
which made no pretensions to elegance, must also have long
retained the [uː], but in the course of time the [y:] filtered down
even to this class. This process is still going on: in parts of the area
where [uː] still prevails — both in the district of [muːs, huːs] and
in the district of [muːs, hy:s] — the peasant, when he is on his good
behavior, speaks [y:] in words where his everyday speech has
[uː]. This flavor of the [y:]-variants appears strikingly in the shape
of hyper-urbanisms: in using the elegant [y:], the speaker sometimes
substitutes it where it is entirely out of place, saying, for
instance, [vy:t] for [vuːt] ‘foot,’ a word in which neither older nor
present-day upper-class Dutch ever spoke an [y:].

The word house will occur much oftener than the word mouse
in official speech and in conversation with persons who represent
the cultural center; mouse is more confined to homely and familiar
situations. Accordingly, we find that the word house in the upper-class
and central form with [y:] spread into districts where the
word mouse has persisted in the old-fashioned form with [uː]. This
330shows us also that the Holland influence, and not the Hanseatic,
was the innovator and aggressor; if the reverse had been the case,
we should find districts where house had [uː] and mouse had [y:].

In the sixteenth and seventeenth centuries, even while the
[y:]-pronunciation was making its conquests, there arose, it would
seem in Antwerp, a still newer pronunciation with [øɥ] instead of
the hitherto elegant [y:]. This new style spread to the Holland
cities, and with this its fortune was made. The [øɥ]-pronunciation,
as in standard Dutch huis [høɥs], muis [møɥs], is today the only
truly urbane form. On our map, the area of this [øɥ] looks as if it
had been laid on top of a former solid area of [y:], leaving only
disconnected patches uncovered along the edge. This picture of
disconnected patches at the periphery is characteristic of older
styles, in language or in other activities, that have been superseded
by some new central fashion. It is characteristic, too, that the
more remote local dialects are taking up a feature, the [y:]-pronunciation,
which in more central districts and in the more privileged
class of speakers, has long ago been superseded by a still
newer fashion.

19. 5. The map in our last example could not show the occurrence
of the present-day standard Dutch-Flemish pronunciation with
[øɥ] in the districts where it has not conquered the local dialects.
To show this would be to cover our whole map with a dense and
minute sprinkling of [øɥ]-forms, for the educated or socially better-placed
persons in the whole area speak standard Dutch-Flemish.

The persistence of old features is easier to trace than the occurrence
of new. The best data of dialect geography are furnished by
relic forms, which attest some older feature of speech. In 1876,
J. Winteler published what was perhaps the first adequate study
of a single local dialect, a monograph on his native Swiss-German
dialect of the settlement Kerenzen in the Canton of Glarus. In this
study, Winteler mentions an archaic imperative form, [lɑx] ‘let,’
irregularly derived from the stem [lɑs-], and says that he is not
certain that anyone still used it at the time of publication; most
speakers, at any rate, already used the widespread and more
regular form [lɑs] ‘let.’ A later observer, C. Streiff, writing in
1915, has not heard the old form; it has been totally replaced by

In the same way, Winteler quotes a verse in which the Glarus
people are mocked for their use of the present-tense plural verb-331


Figure 7. The Canton of Glarus, Switzerland. — In 1915 the shaded areas
still used the provincial [hajd, wajd] as plurals of “have” and “want to;” the
unshaded area used the general Swiss-German forms [hand, wand]. — After

forms [hajd] ‘(we, ye, they) have’ and [wajd] ‘(we, ye, they)
want to,’ forms which sounded offensively rustic to their neighbors,
who used the more generally Swiss provincial forms [hand, wand].
Forty years later, Streiff reports a similar verse, in which the
people of the central region of the canton (including the largest

image spanish

Figure 8. The French speech-area. — A discontinuous isogloss encloses the
two marginal shaded areas in which reflexes of Latin multum “much, very”
are still in use. — After Gamillscheg.

community and seat of government, the town of Glarus) mock
the inhabitants of the outlying valleys for their use of these same
forms, [hajd, wajd]. Our Figure 7, based on Streiff's statements,
shows the distribution in 1915: the more urbane and widespread
[hand, wand] prevail in the central district along the river Linth,
333which includes the capital, Glarus, and communicates freely with
the city of Zurich (toward the northwest); the old rustic forms
are used in the three more remote valleys, including the settlement
of Kerenzen.

The relic form, as this example shows, has the best chance of
survival in remote places, and therefore is likely to appear in


Figure 9. The French speech-area. — The unshaded district uses reflexes
of Latin fallit in the meaning “it is necessary.” The shaded areas use other
forms. — After Jaberg.

small, detached areas. The Latin form multum ‘much,’ surviving,
for instance, in Italian molto [ˈmolto] and Spanish mucho
[ˈmučo] ‘much,’ muy [muj] ‘very,’ has been replaced in nearly
all of the French area by words like standard French très [trɛ]
334‘very,’ a modern form of Latin trans ‘through, beyond, exceeding,’
and beaucoup [boku] ‘very,’ which represents a Latin *bonum
‘a good blow or stroke.’ Figure 8 shows the two detached
marginal areas in which modern forms of Latin multum are still
in use.

In Latin, the word fallit meant ‘he, she, it deceives.’ By way
of a meaning ‘it fails,’ this word came to mean, in medieval French,
‘it is lacking,’ and from this there has developed the modern
French use of il faut [i fo] ‘it is necessary; one must.’ This highly
specialized development of meaning can hardly have occurred
independently in more than one place; the prevalence of the modern
locution in the greater part of the French area must be due
to spread from a center, presumably Paris. Figure 9 shows us,
in the unshaded district, the prevalence of phonetic equivalents
of standard French il faut in local dialects. The shaded districts
use other forms, principally reflexes of Latin calet ‘it's hot.’ It
is evident that the modern form spread southward along the
Rhone, which is a great highway of commerce. We see here how
an isogloss running at right angles to a highway of communication,
will not cross it with unchanged direction, but will swerve off,
run parallel with the highway for a stretch, and then either cross
it or, as in our example, reappear on the other side, and then run
back before resuming its former direction. The bend or promontory
of the isogloss shows us which of the two speech-forms has
been spreading at the cost of the other.

19. 6. If we observe a set of relic forms that exhibit some one
ancient feature, we get a striking illustration of the principle
that each word has its own history. The Latin initial cluster
[sk-] has taken on, in the French area, an initial [e-J, a so-called
prothetic vowel, as, for example, in the following four words with
which our Figure 10 is concerned:

tableau Latin | Modern Standard French | ‘ladder’ | ‘bowl’ | ‘write’ | ‘school’

Our figure shows us six disconnected and, as to commerce, remote
districts which still speak forms without the added vowel, such
as [kwe:l] ‘bowl,’ in one or more of these four words. These
335districts include 55 of the 638 places that were observed by Edmont
(§ 19.2). The districts are:

A. A fairly large area in Belgium, overlapping the political
border of the French Republic at one point (Haybes, Department
of the Ardennes), and covering 23 points of the Atlas.

image spanish

Figure 10. The French speech-area. — The shaded districts speak reflexes
of Latin [sk-] without an added initial vowel. — After Jaberg.

B. A somewhat smaller area in the Departments of the Vosges
and of Meurthe-et-Moselle, overlapping into Lorraine, 14 points.

C. The village of Bobi in Switzerland, 1 point.

D. Mentone and two other villages in the Department of Alpes-Maritimes
on the Italian border, 3 points.

E. A fair-sized district along the Spanish border, in the Department
336of Hautes-Pyrenees, and overlapping into the neighboring
Departments, 11 points.

F. A small interior district in the hill-country of the Auvergne,
Departments of Haute-Loire and Puy-de-Dôme, 3 points.

tableau Words in which forms without added vowel are still spoken | Number of places where forms without added vowel are still spoken | by districts | total | ladder, bowl, write, school | ladder, bowl, write | ladder, bowl, school | bowl, write, school | ladder, bowl | ladder, write | ladder, school | bowl, write | ladder | bowl | write

Figure 11. Prothetic vowel in French. —Occurrence of the forms in the
shaded areas of Figure 10, by communities.

What interests us is the fact that most of the settlements in
these backward districts have adopted the prothetic vowel in one,
two, or three of our words. Thus, in district B, the village of Sainte-Marguerite
(Vosges) says [čo:l] ‘ladder’ and [kwe:l] ‘bowl,’ but,
in the modern style, [ekrir] ‘write’ and [eko:l] ‘school.’ Moreover,
the dialects do not agree as to the words in which the innovation is
337made; thus, in contrast with the preceding case, the village of
Gavarnie (Hautes-Pyrénées), in our district E, says [ˈska:lo]
‘ladder’ and [ˈsko:lo] ‘school,’ but [eskuˈde:lo] ‘bowl’ and
[eskriˈbe] ‘write.’ Only two points, both in district A, have preserved
the old initial type in all four of our words; the others show
various combinations of old and new forms. Figure 11 gives, in the
first column, the combinations of words in which the old form is
still in use, then the number of points (by districts and in total)
where each combination has survived. In spite of the great variety

tableau Words in which forms without added vowel are still spoken | Number of places where forms without added vowel are still spoken | by districts | total | ‘ladder’ | ‘bowl’ | ‘write’ | ‘school’

One point is doubtful

Figure 12. Prothetic vowel in French. — Occurrence of the forms in the
shaded areas of Figure 10, by words.

that appears in this table, the survey by individual words, in
Figure 12, shows that the homely terms ‘ladder’ and ‘bowl’ appear
more often in the old form than do ‘write’ and ‘school,’ which are
associated with official institutions and with a wider cultural outlook.
To be sure, at Bobi (district C) it is precisely ‘ladder’ which
has the new form, but wherever the field of observation is larger,
as in districts A, B, and E, or in the total, the terms for ‘ladder’
and ‘bowl’ tend to lead in the number of conservative forms.338

19. 7. The final result of the process of spread is the complete
submergence of the old forms. Where we find a great area in which
some linguistic change has been uniformly carried out, we may be
sure that the greater part of the uniformity is due to geographic
leveling. Sometimes place-names show us the only trace of the
struggle. In the German area generally, two ancient diphthongs,
which we represent as [ew] and [iw] are still distinct, as in standard
New High German, with [iː] for ancient [ew], Fliege ‘fly’ (noun),
Knie ‘knee,’ Stiefvater ‘step-father,’ tief ‘deep,’ but, with [oj] for
ancient [iw], scheu ‘shy,’ teuer ‘dear,’ neun ‘nine.’ The dialect of
Glarus has apparently lost the distinction, as have adjoining
dialects, wherever a labial or velar consonant followed the diphthong:

old [ew] before labial or velar:

tableau Primitive Germanic Type | Glarus | fly | knee | step-

old [iw]:

tableau Primitive Germanic Type | Glarus | shy | dear | nine

Apparently, then, these two old types are both represented in
Glarus by modern [y:], in accordance with the general South-German
development. A single form suggests that the [y:] for old
[ew] is really an importation, namely, the word deep, Primitive
Germanic type *[ˈdewpaz], which appears in Glarus as [tœjf]. Our
suspicion that the diphthong [œj] is the older representative of
[ew] before labials and velars in this region, is confirmed by a
place-name: [ˈxnœj-grɑ:t], literally ‘Knee-Ridge.’

The southwestern corner of German-speaking Switzerland has
changed the old Germanic [k] of words like drink to a spirant [x]
and has lost the preceding nasal, as in [ˈtriːxə] ‘to drink.’ This is
today a crass localism, for most of Switzerland, along with the
rest of the Dutch-German area, speaks [k]. Thus, Glarus says
[ˈtriŋkə] ‘to drink,’ in accord with standard German trinken.
Place-names, however, show us that the deviant pronunciation
once extended over a much larger part of Switzerland. Glarus,
339well to the east, alongside the common noun [ˈwɪŋkəl] ‘angle,
corner,’ has the place-name of a mountain pasture [ˈwɪxlə] ‘Corners,’
and alongside [xrɑŋk] ‘sick’ (formerly, ‘crooked’) the name
of another pasture [ˈxrawx-tɑ:l] ‘Crank-Dale,’ that is, ‘Crooked-Valley,’

19. 8. Dialect geography thus gives evidence as to the former
extension of linguistic features that now persist only as relic forms.
Especially when a feature appears in detached districts that are
separated by a compact area in which a competing feature is
spoken, the map can usually be interpreted to mean that the detached
districts were once part of a solid area. In this way, dialect
geography may show us the stratification of linguistic features;
thus, our Figure 6, without any direct historical supplementation,
would tell us that the [uː]-forms were the oldest, that they were
superseded by the [y:]-forms, and these, in turn, by the diphthongal

Since an isogloss presumably marks a line of weakness in the
density of communication, we may expect the dialect map to show
us the communicative conditions of successive times. The inhabitants
of countries like England, Germany, or France, have
always applied provincial names to rough dialectal divisions, and
spoken of such things as “the Yorkshire dialect,” “the Swabian
dialect,” or “the Norman dialect.” Earlier scholars accepted these
classifications without attempting to define them exactly; it was
hoped, later, that dialect geography would lead to exact definitions.
The question gained interest from the wave-theory (§ 18.12), since
the provincial types were examples of the differentiation of a
speech-area without sudden cleavage. Moreover, the question took
on a sentimental interest, since the provincial divisions largely
represent old tribal groupings: if the extension of a dialect, such
as, say, the “Swabian dialect” in Germany, could be shown to
coincide with the area of habitation of an ancient tribe, then language
would again be throwing light on the conditions of a bygone

In this respect, however, dialect geography proved to be disappointing.
It showed that almost every village had its own dialectal
features, so that the whole area was covered by a network of
isoglosses. If one began by setting up a list of characteristic provincial
peculiarities, one found them prevailing in a solid core,
but shading off at the edges, in the sense that each characteristic
340was bordered by a whole set of isoglosses representing its presence
in different words — just as the house and mouse isoglosses
for [y:] and [uː] do not coincide in the eastern Netherlands (Figure
6). A local dialect from the center of Yorkshire or Swabia
or Normandy could be systematically classed in terms of its province,
but at the outskirts of such a division there lie whole bands
of dialects which share only part of the provincial characteristics.
In this situation, moreover, there is no warrant for the initial
list of characteristics. If these were differently selected — say,
without regard to the popularly current provincial classification
— we should obtain entirely different cores and entirely different
zones of transition.

Accordingly, some students now despaired of all classification
and announced that within a dialect area there are no real boundaries.
Even in a domain such as that of the western Romance languages
(Italian, Ladin, French, Spanish, Portuguese) it was urged
that there were no real boundaries, but only gradual transitions:
the difference between any two neighboring points was no more
and no less important than the difference between any two other
neighboring points. Opposing this view, some scholars held fast
to the national and provincial classifications, insisting, perhaps
with some mystical fervor, on a terminology of cores and

It is true that the isoglosses in a long-settled area are so many
as to make possible almost any desired classification of dialects
and to justify almost any claim concerning former densities of
communication. It is easy to see, however, that, without prejudice
of any kind, we must attribute more significance to some isoglosses
than to others. An isogloss which cuts boldly across a
whole area, dividing it into two nearly equal parts, or even an
isogloss which neatly marks off some block of the total area, is
more significant than a petty line enclosing a localism of a few
villages. In our Figures 4 and 5, isoglosses 1, 2, 3, which mark
off southwestern German from the rest of the German area, are
evidently more significant than, say, isogloss 9, which encloses
only a few villages. The great isogloss shows a feature which has
spread over a large domain; this spreading is a large event, simply
as a fact in the history of language, and, may reflect, moreover,
some non-linguistic cultural movement of comparable strength.
As a criterion of description, too, the large division is, of course,
341more significant than small ones; in fact, the popular classification
of dialects is evidently based upon the prevalence of certain
peculiarities over large parts of an area.

Furthermore, a set of isoglosses running close together in much
the same direction — a so-called bundle of isoglosses — evidences
a larger historical process and offers a more suitable basis of classification
than does a single isogloss that represents, perhaps, some
unimportant feature. It appears, moreover, that these two characteristics,
topographic importance and bundling, often go hand
in hand. Thus, France is divided by a great bundle of isoglosses
running east and west across the area. This division reflects the
medieval division of France into the two cultural and linguistic
domains of French and Provencal.

The most famous bundle of this kind, perhaps, is the east-and-west
bundle which runs across the Dutch-German area, separating
Low German from High German. The difference is in the treatment
of the Primitive Germanic unvoiced stops [p, t, k], which in
the south have been shifted to spirants and affricates. If we take
standard Dutch and standard German as representatives of the
two types, our isoglosses separate forms like these:

tableau Northern | Southern | make | I | sleep | thorp ‘village’ | pound | bite | that | to

The isoglosses of these and other forms that contain Primitive
Germanic [p, t, k] run in a great bundle, sometimes coinciding,
but at other times diverging, and even crossing each other. Thus,
round Berlin, the isogloss of make, together with a good many
others, makes a northward bend, so that there one says [ik] I
with unshifted [k], but [ˈmaxen] ‘make’ with [k] shifted to [x];
on the other hand, in the west the isogloss of I swerves off in a
northwesterly direction, so that round Düsseldorf one says [ix]
‘I’ with the shifted sound, but [ˈma:ken] ‘make’ with the old
[k] preserved.342

In this way we find that the topographic distribution of linguistic
features within a dialect area is not indifferent, and
exhibits decided cleavages. We must make only two obvious
reservations: we cannot guarantee to preserve the popular terminology
by provinces, but, if we retain provincial names, must redefine
them; and we can bound our divisions either imperfectly,
by zones, or arbitrarily, by selecting some one isogloss as the representative
of a whole bundle.

19. 9. Having found the linguistic divisions of an area, we may
compare them with other lines of cleavage. The comparison
shows that the important lines of dialectal division run close to
political lines. Apparently, common government and religion,
and especially the custom of intermarriage within the political
unit, lead to relative uniformity of speech. It is estimated that,
under older conditions, a new political boundary led in less than
fifty years to some linguistic difference, and that the isoglosses
along a political boundary of long standing would persist, with
little shifting, for some two-hundred years after the boundary
had been abolished. This seems to be the primary correlation.
If the important isoglosses agree with other lines of cultural division
— as, in northern Germany, with a difference in the construction
of farm-houses — or if they agree with geographic
barriers, such as rivers or mountain-ranges, then the agreement
is due merely to the fact that these features also happen to concord
with political divisions.

This has been shown most plainly in the distribution of the
important German isoglosses along the Rhine. Some forty kilometers
east of the Rhine the isoglosses of the great bundle that
separates Low German and High German begin to separate and
spread out northwestward and southwestward, so as to form what
has been called the “Rhenish fan” (Figure 13). The isogloss of
northern [k] versus southern [x] in the word make, which has been
taken, arbitrarily, as the critical line of division, crosses the Rhine
just north of the town of Benrath and, accordingly, is called the
“Benrath line.” It is found, now, that this line corresponds
roughly to an ancient northern boundary of the territorial domains
of Berg (east of the Rhine) and Jülich (west of the Rhine).
The isogloss of northern [k] versus southern [x] in the word I
swerves off northwestward, crossing the Rhine just north of the
village of Ürdingen, and is known accordingly, as the “Ürdingen
343line;” some students take this, rather than the line of make, as
the arbitrary boundary between Low and High German. The
Ürdingen line corresponds closely to the northern boundaries of
the pre-Napoleonic Duchies, abolished in 1789, of Jülich and
Berg — the states whose earlier limit is reflected in the Benrath
line — and of the Electorate of Cologne. Just north of Ürdingen,
the town of Kaldenhausen is split by the Ürdingen line into a
western section which says [ex] and an eastern which says [ek];


Figure 13. The Dutch-German speech-area, showing the isogloss of [k]
versus [x] in the word make, and, in the western part, the divergence of three
other isoglosses which in the east run fairly close to that of make. — After

we learn that up to 1789 the western part of the town belonged
to the (Catholic) Electorate of Cologne, and the eastern part to
the (Protestant) County of Mörs. Our map shows also two isoglosses
branching south west ward. One is the line between northern
[p] and southern [f] in the word [dorp —dorf] ‘village;’ this
line agrees roughly with the southern boundaries in 1789 of Jülich,
Cologne, and Berg, as against the Electorate of Treves. In a still
more southerly direction there branches off the isogloss between
northern [t] and southern [s] in the word [dat — das] ‘that,’ and this
344line, again, coincides approximately with the old southern boundary
of the Electorate and Archbishopric of Treves.

All this shows that the spread of linguistic features depends
upon social conditions. The factors in this respect are doubtless
the density of communication and the relative prestige of different
social groups. Important social boundaries will in time attract
isogloss-lines. Yet it is evident that the peculiarities of the
several linguistic forms themselves play a part, since each is likely
to show an isogloss of its own. In the Netherlands we saw a new
form of the word house spreading farther than a new form of the
homely word mouse (§ 19.4). We can hope for no scientifically
usable analysis, such as would enable us to predict the course of
every isogloss: the factors of prestige in the speakers and of meaning
(including connotation) in the forms cut off our hope of this.
Nevertheless, dialect geography not only contributes to our understanding
of the extra-linguistic factors that affect the prevalence
of linguistic forms, but also, through the evidence of relic forms
and stratifications, supplies a great many details concerning the
history of individual forms.345

Chapter 20
Phonetic Change

20. 1. Written records of earlier speech, resemblance between
languages, and the varieties of local dialects, all show that languages
change in the course of time. In our Old English records
we find a word stan ‘stone,’ which we interpret phonetically as
[sta:n]; if we believe that the present-day English word stone
[stown] is the modern form, by unbroken tradition, of this Old
English word, then we must suppose that Old English [a:] has
here changed to modern [ow]. If we believe that the resemblances
are due not to accident, but to the tradition of speech-habits,
then we must infer that the differences between the resemblant
forms are due to changes in these speech-habits. Earlier students
recognized this; they collected sets of resemblant forms (etymologies)
and inferred that the differences between the forms of a
set were due to linguistic change, but, until the beginning of the
nineteenth century, no one succeeded in classifying these differences.
The resemblances and differences varied from set to set.
An Old English bat, which we interpret phonetically as [ba:t],
is in one meaning paralleled by modern English boat [bowt], but
in another meaning by modern English bait [bejt]. The initial consonants
are the same in Latin dies and English day, but different
in Latin duo and English two. The results of linguistic change
presented themselves as a hodge-podge of resemblances and differences.
One could suspect that some of the resemblances were
merely accidental (“false etymologies”), but there was no test.
One could reach no clear formulation of linguistic relationship —
the less so, since the persistence of Latin documents through the
Middle Ages alongside of documents in the Romance languages
distorted one's whole view of linguistic chronology.

It is not useless to look back at those times. Now that we have
a method which brings order into the confusion of linguistic resemblances
and throws some light on the nature of linguistic
relationship, we are likely to forget how chaotic are the results
of linguistic change when one has no key to their classification.
346Since the beginning of the nineteenth century we have learned to
classify the differences between related forms, attributing them
to several kinds of linguistic change. The data, whose variety
bewildered earlier students, lend themselves with facility to this
classification. Resemblances which do not fit into our classes of
change, are relatively few and can often be safely ruled out as
accidental; this is the case, for instance, with Latin dies : English
day, which we now know to be a false etymology.

The process of linguistic change has never been directly observed;
we shall see that such observation, with our present facilities,
is inconceivable. We are assuming that our method of
classification, which works well (though not by any means perfectly),
reflects the actual factors of change that produced our
data. The assumption that the simplest classification of observed
facts is the true one, is common to all science; in our case, it is
well to remember that the observed facts (namely, the results of
linguistic change as they show themselves in etymologies) resisted
all comprehension until our method came upon the scene.
The first step in the development of method in historical linguistics
was the seeking out of uniform phonetic correspondences;
we take these correspondences to be the results of a factor of
change which we call phonetic change.

20. 2. At the beginning of the nineteenth century we find a
few scholars systematically picking out certain types of resemblance,
chiefly cases of phonetic agreement or correspondence.
The first notable step was Rask's and Grimm's observation (§ 1.7)
of correspondences between Germanic and other Indo-European
languages. From among the chaotic mass of resemblant forms, they
selected certain ones which exhibited uniform phonetic correlations.
Stated in present-day terms, these correlations appear as follows:

(1) Unvoiced stops of the other languages are paralleled in
Germanic by unvoiced spirants:

[p — f] Latin pēs : English foot; Latin piscis : English fish; Latin
pater : English father;

[t — θ] Latin trēs : English three; Latin tenuis : English thin;
Latin tacēre ‘to be silent’ : Gothic [ˈθahan];

[k — h] Latin centum : English hundred; Latin caput : English
head; Latin cornū, : English horn.

(2) Voiced stops of the other languages are paralleled in Germanic
by unvoiced stops:347

[b — p] Greek [ˈkannabis] : English hemp;

[d — t] Latin duo : English two; Latin dens : English tooth; Latin
edere : English eat;

[g — k] Latin grānum : English corn; Latin genus : English kin;
Latin ager : English acre.

(3) Certain aspirates and spirants of the other languages
(which we denote today as “reflexes of Primitive Indo-European
voiced aspirates”) are paralleled in Germanic by voiced stops and

Sanskrit [bh], Greek [ph], Latin [f], Germanic [b, v]: Sanskrit
[ˈbhara:mi] ‘I bear,’ Greek [ˈphero:], Latin ferō : English bear;
Sanskrit [ˈbhra:ta:], Greek [ˈphra:te:r], Latin frāter : English
brother; Latin frangere : English break;

Sanskrit [dh], Greek [th], Latin [f], Germanic [d, ð]: Sanskrit
[ˈa-dha:t] ‘he put,’ Greek [ˈthe:so:] ‘I shall put,’ Latin fēcī ‘I
made, did’ : English do; Sanskrit [ˈmadhu] ‘honey, mead,’ Greek
[ˈmethu] ‘wine’ : English mead; Sanskrit [ˈmadhjah], Latin medius :
English mid;

Sanskrit [h], Greek [kh], Latin [h], Germanic [g, ɣ]: Sanskrit
[hanˈsah] : English goose; Sanskrit [ˈvahati] ‘he carries on a vehicle,’
Latin vehit : Old English wegan ‘to carry, move, transport;’ Latin
hostis ‘stranger, enemy’ : Old English giest ‘guest.’

The only reason for assembling cases like these is the belief
that the correlations are too frequent or in some other way too
peculiar to be due to chance.

20. 3. Students of language have accepted these correlations
(calling them, by a dangerous metaphor, Grimm's “law”), because
the classification they introduce is confirmed by further study:
new data show the same correspondences, and cases which do not
show these correspondences lend themselves to other classifications.

For instance, from among the cases which do not show Grimm's
correspondences, it is possible to sort out a fair-sized group in
which unvoiced stops [p, t, k] of the other languages appear also in
Germanic; thus, the [t] of the other languages is paralleled by
Germanic [t] in cases like the following:

Sanskrit [ˈasti] ‘he is,’ Greek [ˈesti], Latin est: Gothic [ist] ‘is;’

Latin captus ‘taken, caught’ : Gothic [hafts] ‘restrained;’

Sanskrit [ašˈta:w] ‘eight,’ Greek [okˈto:] Latin octō : Gothic

Now, in all these cases the [p, t, k] in Germanic is immediately
preceded by an unvoiced spirant [s, f, h], and a survey of the cases
which conform to Grimm's correspondences shows that in them
the Germanic consonant is never preceded by these sounds.
Grimm's correlations have thus, by leaving a residue, led us to
find another correlation: after [s, f, h] Germanic [p, t, k] parallel
the [p, t, k] of the other Indo-European languages.

Among the residual forms, again, we find a number in which
initial voiced stops [b, d, g] of Germanic are paralleled in Sanskrit
not by [bh, dh, gh], as Grimm would have it, but by [b, d, g], and in
Greek not by the expected [ph, th, kh], but by [p, t, k]. An example
is Sanskrit [ˈbo:dha:mi] ‘I observe,’ Greek [ˈpewthomaj] ‘I
experience’ : Gothic [ana-ˈbiwdan] ‘to command,’ Old English
[ˈbe:odan] ‘to order, announce, offer,’ English bid. In 1862,
Hermann Grassmann (1809-1877) showed that this type of correlation
appears wherever the next consonant (the consonant after
the intervening vowel or diphthong) belongs to Grimm's third
type of correspondences. That is, Sanskrit and Greek do not have
aspirate stops at the beginning of two successive syllables, but,
wherever the related languages show this pattern, have the first
of the two stops unaspirated: corresponding to Germanic *[bewda-],
we find in Sanskrit not *[bho:dha-] but [bo:dha-], and in Greek
not *[phewtho-] but [pewtho-]. Here too, then, the residual data
which are marked off by Grimm's correspondences, reveal a

In this case, moreover, we get a confirmation in the structure of
the languages. In Greek, certain forms have a reduplication (§ 13.8)
in which the first consonant of the underlying stem, followed by a
vowel, is prefixed: [ˈdo:so:] ‘I shall give,’ [ˈdi-do:mi] ‘I give.’
We find, now, that for stems with an initial aspirate stop the
reduplication is made with a plain stop: [ˈthe:so:] ‘I shall put,’
[ˈti-the:mi] ‘I put.’ The same habit appears elsewhere in Greek
morphology; thus, there is a noun-paradigm with nominative
singular [ˈthriks] ‘hair,’ but other case-forms like the accusative
[ˈtrikha]: when the consonant after the vowel is aspirated, the
initial consonant is [t] instead of [th]. Similarly, in Sanskrit, the
normal reduplication repeats the first consonant: [ˈa-da:t] ‘he
gave,’ [ˈda-da:mi] ‘I give,’ but for an initial aspirate the reduplication
has a plain stop: [ˈa-dha:t] ‘he put,’ [ˈda-dha:mi] ‘I put,’
and similar alternations appear elsewhere in Sanskrit morphology.
349These alternations are obviously results of the sound-change discovered
by Grassmann.

20. 4. If our correspondences are not due to chance, they must
result from some historical connection, and this connection the
comparative method reconstructs, as we have seen, by the assumption
of common descent from a parent language. Where the related
languages agree, they are preserving features of the parent language,
such as, say, the [r] in the word brother, the [m] in the words
mead and mid (§ 20.2), or the [s] in the verb-forms for ‘he is’
(§ 20.3). Where the correspondence connects markedly different
phonemes, we suppose that one or more of the languages have
changed. Thus we state Grimm's correspondences by saying:

(1) Primitive Indo-European unvoiced stops [p, t, k] changed
in pre-Germanic to unvoiced spirants [f, θ, h];

(2) Primitive Indo-European voiced stops [b, d, g] changed in
pre-Germanic to unvoiced stops [p, t, k];

(3) Primitive Indo-European voiced aspirate stops [bh, dh, gh]
changed in pre-Germanic to voiced stops or spirants [b, d, g], in
pre-Greek to unvoiced aspirate stops [ph, th, kh], in pre-Italic
and pre-Latin to [f, f, h]. In this case the acoustic shape of the
Primitive Indo-European phonemes is by no means certain, and
some scholars prefer to speak of unvoiced spirants [f, θ, x]; similarly,
we do not know whether the Primitive Germanic reflexes
were stops or spirants, but these doubts do not affect our conclusions
as to the phonetic pattern.

The correspondences where [p, t, k] appear also in Germanic
demand a restriction for case (1): immediately after a consonant
(those which actually occur are [s, p, k]), the Primitive Indo-European
unvoiced stops [p, t, k] were not changed in pre-Germanic.

Grassmann's correspondences we state historically by saying
that at a certain stage in the history of pre-Greek, forms which
contained two successive syllables with aspirate stops, lost the
aspiration of the first stop. Thus, we reconstruct:

tableau *Primitive Indo-European | pre-Greek | Greek

On the other hand, in the nominative singular of the word for
‘hair,’ we suppose that there never was an aspirate after the vowel:350

Primitive Indo-European *[dhriks] appears as Greek [thriks]. We
infer a similar change for pre-Indo-Iranian: a Primitive Indo-European
*[bhewdho-] appearing in Sanskrit as [bo:dha-], a
Primitive Indo-European *[dhedhe:-] as [dadha:-], and so on.

A further step in the reconstruction of the historical events
proceeds from the fact that the loss of aspiration results in Sanskrit
in [b, d, g], but in Greek in [p, t, k]. This implies that the
Primitive Indo-European [bh, dh, gh] had already become unvoiced
[ph, th, kh] in pre-Greek when the loss of aspiration took
place. Since this unvoicing does not occur in Indo-Iranian, we
conclude that the de-aspiration in pre-Greek and the de-aspiration
in pre-Indo-Iranian took place independently.

The interpretation, then, of the phonetic correspondences that
appear in our resemblant forms, assumes that the phonemes of a
language are subject to historical change
. This change may be limited
to certain phonetic conditions; thus, in pre-Germanic, [p, t, k]
did not change to [f, θ, h] when another unvoiced consonant
immediately preceded, as in *[kəptos] > Gothic [hafts]; in pre-Greek,
[ph, th, kh] became [p, t; k] only when the next syllable
began with an aspirate. This type of linguistic change is known as
phonetic change (or sound change). In modern terminology, the
assumption of sound-change can be stated in the sentence: Phonemes

20. 5. When we have gathered the resemblant forms which show
the recognized correlations, the remainders will offer two self-evident
possibilities. We may have stated a correlation too
narrowly or too widely: a more careful survey or the arrival of new
data may show the correction. A notable instance of this was
Grassmann's discovery. The fact that residues have again and
again revealed new correlations, is a strong confirmation of our
method. Secondly, the resemblant forms may not be divergent
pronunciations of the same earlier form. Grimm, for instance,
mentioned Latin dies : English day as an etymology which did not
fall within his correlations, and since his time no amount of research
has revealed any possibility of modifying the otherwise
valid correlation-classes so that they may include this set. Similarly,
Latin habēre ‘to have’ : Gothic haban, Old High German
habēn, in spite of the striking resemblance, conflicts with types of
correlation that otherwise hold good. In such cases, we may
attribute the resemblance to accident, meaning by this that it is
351not due to any historical connection; thus, Latin dies : English day
is now regarded by everyone as a “false etymology.” Or else, the
resemblance may be due to grammatical resemblance of forms in
the parent language; thus, Latin habēre ‘to have’ and Old High
German habēn ‘to have’ may be descendants, respectively, of two
stems, *[ghaˈbhe:-] and *[kaˈbhe:-] which were morphologically
parallel in Primitive Indo-European. Finally, our resemblant
forms may owe their likeness to a historical connection other than
descent from a common prototype. Thus, Latin dentālis ‘pertaining
to a tooth’ and English dental resemble each other, but do not
show the correlations (e.g. Latin d : English t) which appear in
Latin and English reflexes of a common Primitive Indo-European
prototype. The reason is that dental is merely the English-speaker's
reproduction of the Latin word.

To sum up, then, the residual forms which do not fit into recognized
types of phonetic correlation may be:

(1) descendants of a common ancestral form, deviant only because
we have not correctly ascertained the phonetic correlation,
e.g. Sanskrit [ˈbo:dha:mi] and English bid, before
Grassmann's discovery;

(2) not descendants of a common ancestral form, in which case
the resemblance may be due to

(a) accident, e.g. Latin dies : English day;

(b) morphologic partial resemblance in the parent language,
e.g. Latin habēre : English have;

(c) other historical relations, e.g. Latin dentālis : English

If this is correct, then the study of residual resemblant forms
will lead us to discover new types of phonetic correlation (1), to
weed out false etymologies (2a), to uncover the morphologic structure
of the parent speech (2b), or to recognize types of linguistic
change other than sound-change (2c). If the study of residual
forms does not lead to these results, then our scheme is incorrect.

20. 6. During the first three quarters of the nineteenth century
no one, so far as we know, ventured to limit the possibilities in the
sense of our scheme. If a set of resemblant forms did not fit into
the recognized correlations, scholars felt free to assume that these
forms were nevertheless related in exactly the same way as the
normal forms — namely, by way of descent from a common ancestral
352form. They phrased this historically by saying that a
speech-sound might change in one way in some forms, but might
change in another way (or fail to change) in other forms. A
Primitive Indo-European [d] might change to [t] in pre-Germanic
in most forms, such as two (: Latin duo), ten (: Latin decent), tooth
(: Latin dens), eat (: Latin edere), but remain unchanged in some
other forms, such as day (: Latin dies).

On the whole, there was nothing to be said against this view —
in fact, it embodied a commendable caution — unless and until
an extended study of residual forms showed that possibilities
(1) and (2a, b, c) were realized in so great a number of cases as to
rule out the probability of sporadic sound-change. In the seventies
of the nineteenth century, several scholars, most notably, in the
year 1876, August Leskien (§ 1.9), concluded that exactly this had
taken place: that the sifting of residual forms had resulted so
often in the discovery of non-contradictory facts (1, 2b, 2c) or in
the weeding out of false etymologies (2a), as to warrant linguists
in supposing that the change of phonemes is absolutely regular.
This meant, in terms of our method, that all resemblances between
forms which do not fall into the recognized correspondence-classes
are due to features of sound-change which we have failed to
recognize (1), or else are not divergent forms of a single prototype,
either because the etymology is false (2a), or because some factor
other than sound-change has led to the existence of resemblant
forms (2b, c). Historically interpreted, the statement means that
sound-change is merely a change in the speakers' manner of
producing phonemes and accordingly affects a phoneme at every
occurrence, regardless of the nature of any particular linguistic
form in which the phoneme happens to occur. The change may
concern some habit of articulation which is common to several
phonemes, as in the unvoicing of voiced stops [b, d, g] in pre-Germanic.
On the other hand, the change may concern some habit
of articulating successions of phonemes, and therefore take place
only under particular phonetic conditions, as when [p, t, k] in
pre-Germanic became [f, θ, h] when not preceded by another
sound of the same group or by [s]; similarly, [ph, th, kh] in pre-Greek
became [p, t, k] only when the next syllable began with an
aspirate. The limitations of these conditioned sound-changes are,
of course, purely phonetic, since the change concerns only a habit
of articulatory movement; phonetic change is independent of
353non-phonetic factors, such as the meaning, frequency, homonymy,
or what not, of any particular linguistic form. In present-day
terminology the whole assumption can be briefly put into the
words: phonemes change, since the term phoneme designates a
meaningless minimum unit of signaling.

The new principle was adopted by a number of linguists, who
received the nickname of “neo-grammarians.” On the other hand,
not only scholars of the older generation, such as Georg Curtius
(1820-1885), but also some younger men, most notably Hugo
Schuchardt (1842-1927), rejected the new hypothesis. The discussion
of the pro's and con's has never ceased; linguists are as
much divided on this point today as in the 1870's.

A great part of this dispute was due merely to bad terminology.
In the 1870's, when technical terms were less precise than today,
the assumption of uniform sound-change received the obscure and
metaphorical wording, “Phonetic laws have no exceptions.” It
is evident that the term “law” has here no precise meaning, for
a sound-change is not in any sense a law, but only a historical
occurrence. The phrase “have no exceptions” is a very inexact
way of saying that non-phonetic factors, such as the frequency or
meaning of particular linguistic forms, do not interfere with the
change of phonemes.

The real point at issue is the scope of the phonetic correspondence-classes
and the significance of the residues. The neo-grammarians
claimed that the results of study justified us in making
the correspondence-classes non-contradictory and in seeking a
complete analysis of the residues. If we say that Primitive Indo-European
[d] appears in Germanic as [t], then, according to the
neo-grammarians, the resemblance of Latin dies and English day
or of Latin dentālis and English dental, cannot be classed simply
as “an exception” — that is, historically, as due to the pre-Germanic
speakers' failure to make the usual change of habit — but
presents a problem. The solution of this problem is either the
abandonment of the etymology as due to accidental resemblance
(Latin dies : English day), or a more exact formulation of the
phonetic correspondence (Grassmann's discovery), or the recognition
of some other factors that produce resemblant forms (Latin
dentālis borrowed in English dental). The neo-grammarian insists,
particularly, that his hypothesis is fruitful in this last direction:
it sorts out the resemblances that are due to factors other than
354phonetic change, and accordingly leads us to an understanding of
these factors.

The actual dispute, then, concerns the weeding-out of false
etymologies, the revision of our statements of phonetic correspondence,
and the recognition of linguistic changes other than

20. 7. The opponents of the neo-grammarian hypothesis claim
that resemblances which do not fit into recognized types of phonetic
correspondence may be due merely to sporadic occurrence or
deviation or non-occurrence of sound-change. Now, the very
foundation of modern historical linguistics consisted in the setting
up of phonetic correspondence-classes: in this way alone did Rask
and Grimm bring order into the chaos of resemblances which had
bewildered all earlier students. The advocates of sporadic sound-change,
accordingly, agree with the neo-grammarians in discarding
such etymologies as Latin dies : English day, and retain only a
few, where the resemblance is striking, such as Latin habēre : Old
High German habēn, or Sanskrit [ko:kilah], Greek [ˈkokkuks],
Latin cuculus : English cuckoo. They admit that this leaves us
no criterion of decision, but insist that our inability to draw a
line does not prove anything: exceptional sound-changes occurred,
even though we have no certain way of recognizing

The neo-grammarian sees in this a serious violation of scientific
method. The beginning of our science was made by a procedure
which implied regularity of phonetic change, and further
advances, like Grassmann's discovery, were based on the same
implicit assumption. It may be, of course, that some other assumption
would lead to an even better correlation of facts, but
the advocates of sporadic sound-change offer nothing of the kind;
they accept the results of the actual method and yet claim to explain
some facts by a contradictory method (or lack of method)
which was tried and found wanting through all the centuries that
preceded Rask and Grimm.

In the historical interpretation, the theory of sporadic sound-change
faces a very serious difficulty. If we suppose that a form
like cuckoo resisted the pre-Germanic shift of [k] to [h] and still
preserves a Primitive Indo-European [k], then we must also suppose
that during many generations, when the pre-Germanic
people had changed their way of pronouncing Primitive Indo-European
355[k] in most words, and were working on through successive
acoustic types such as, say, [kh — kx — x — h], they were
still in the word cuckoo pronouncing an unchanged Primitive Indo-European
[k]. If such things happened, then every language would
be spotted over with all sorts of queer, deviant sounds, in forms
which had resisted sound-change or deviated from ordinary
changes. Actually, however, a language moves within a limited
set of phonemes. The modern English [k] in cuckoo is no different
from the [k] in words like cow, calf, kin, which has developed
normally from the Primitive Indo-European [g]-type. We should
have to suppose, therefore, that some later change brought the
preserved Primitive Indo-European [k] in cuckoo into complete
equality with the Germanic [k] that reflects a Primitive Indo-European
[g], and, since every language moves within a limited
phonetic system, we should have to suppose that in every case
of sporadic sound-change or resistance to sound-change, the
discrepant sound has been reduced to some ordinary phonemic
type in time to escape the ear of the observer. Otherwise we should
find, say, in present-day standard English, a sprinkling of forms
which preserved sounds from eighteenth-century English, early
modern English, Middle English, Old English, Primitive Germanic,
and so on — not to speak of deviant sounds resulting from
sporadic changes in some positive direction.

Actually, the forms which do not exhibit ordinary phonetic
correlations, conform to the phonemic system of their language
and are peculiar only in their correlation with other forms. For
instance, the modern standard English correspondents of Old
English [o:] show some decided irregularities, but these consist
simply in the presence of unexpected phonemes, and never in
deviation from the phonetic system. The normal representation
seems to be:

[a] before [s, z] plus consonant other than [t]: goshawk, gosling,

[ɔ] before Old English consonant plus [t]: foster, soft, sought
(Old English sōhte), brought, thought;

[u] before [k] book, brook (noun), cook, crook, hook, look, rook,
shook, took;

[o] before [n] plus consonant other than [t] and before consonant
plus [r̩]: Monday, month; brother, mother, other, rudder;

[ow] before [nt] and [r] and from the combination of Old English
356[o:w]: don't; floor, ore, swore, toward, whore; blow (‘bloom’), flow,
glow, grow, low (verb), row, stow;

[uw] otherwise: do, drew, shoe, slew, too, to, woo, brood, food, mood,
hoof, roof, woof, cool, pool, school, stool, tool, bloom, broom, doom,
gloom, loom, boon, moon, noon, soon, spoon, swoon, whoop, goose,
loose, boot, moot, root, soot, booth, sooth, tooth, smooth, soothe, behoove,
prove, ooze.

If we take the correlation of Old English [o:] with these sounds
as normal under the phonetic conditions of each case, then we have
the following residue of contradictory forms:

[ɑ] shod, fodder;

[aw] bough, slough;

[e] Wednesday;

[o] blood, flood, enough, tough, gum, done, must, doth, glove;

[ow] woke;

[u] good, hood, stood, bosom, foot, and optionally hoof, roof,
broom, soot;

[uw] moor, roost.

All of these seven deviant types contain some ordinary English
phoneme; the [o], for instance, in blood, etc., is the ordinary [o]-phoneme,
which represents Old English [u] in words like love,
tongue, son, sun, come. In every case, the discrepant forms show
not queer sounds, but merely normal phonemes in a distribution
that runs counter to the expectations of the historian.

20. 8. As to the correction of our correspondence-groups by a
careful survey of the residual cases, the neo-grammarians soon
got a remarkable confirmation of their hypothesis in Verner's
treatment of Germanic forms with discrepant [b, d, g] in place
of [f, θ, h] (§ 18.7). Verner collected the cases like Latin pater :
Gothic [ˈfadar], Old English [ˈfɛder], where Primitive Indo-European
[t] appears in Germanic as [d, ð], instead of [θ]. Now,
the voicing of spirants between vowels is a very common form of
sound-change, and has actually occurred at various times in the
history of several Germanic languages. Primitive Germanic [θ]
appears as a voiced spirant, coinciding with the reflex of Primitive
Germanic [d], in Old Norse, which says, for instance, [ˈbro:ðer],
with the same consonant as [ˈfaðer]. In Old English, too, the
Primitive Germanic [θ] had doubtless become voiced between
vowels, as in [ˈbro:ðor], although it did not coincide with [d],
the reflex of Primitive Germanic [d], as in [ˈfɛder]. In both Old
357Norse and Old English, Primitive Germanic [f] had become voiced
[v] between vowels, as in Old English ofen [ˈoven] ‘oven’ (Old
High German ofan [ˈofan]), coinciding with the [v] that represented
Primitive Germanic [b], as in Old English yfel [ˈyvel]
‘evil’ (Old High German ubil [ˈybil]). Nothing could be more
natural, therefore, if one admitted the possibility of irregular
sound-change, than to suppose that the voicing of intervocalic
spirants had begun sporadically in some words already in pre-Germanic
time, and that a Primitive Germanic *[ˈfader] alongside
*[ˈbro:θer] represented merely the beginning of a process that was
to find its completion in the Old Norse, Old English, and Old Saxon
of our actual records. Yet in 1876 Verner's study of the deviant
forms showed an unmistakable correlation: in a fair number of
cases and in convincing systematic positions, the deviant [b, d, g]
of Germanic appeared where Sanskrit and Greek (and therefore,
presumably, Primitive Indo-European) had an unaccented vowel
or diphthong before the [p, t, k], as in Sanskrit [piˈta:], Greek [paˈte:r]
: Primitive Germanic *[ˈfader], contrasting with Sanskrit
[ˈbhra:ta:], Greek [ˈphra:te:r] : Primitive Germanic *[ˈbro:θer].
Similarly, Sanskrit [ˈçvaçurah] ‘father-in-law,’ reflecting, presumably
a Primitive Indo-European *[ˈswekuros], shows in Germanic
the normal reflex of [h] for [k], as in Old High German
[ˈswehar], but Sanskrit [çvaˈçruːh] ‘mother-in-law,’ reflecting a
Primitive Indo-European *[sweˈkruːs] appears in Germanic with
[g], as in Old High German [ˈswigar], representing the Primitive
Indo-European [k] after the unstressed vowel.

A confirmation of this result was the fact that the unvoiced
spirant [s] of Primitive Indo-European suffered the same change
under the same conditions: it appears in Germanic as [s], except
when the preceding syllabic was unaccented in Primitive Indo-European;
in this case, it was voiced in pre-Germanic, and appears
as Primitive Germanic [z], which later became [r] in Norse and in
West Germanic. In a number of irregular verb-paradigms the
Germanic languages have medial [f, θ, h, s] in the present
tense and in the singular indicative-mode forms of the past
tense, but [b, d, g, z] in the plural and subjunctive forms of
the past tense and in the past participle, as, for instance, in Old

[ˈweorθan] ‘to become,’ [he: ˈwearθ] ‘he became,’ but [we:ˈ
wurdon] ‘we became;’358

ke:osan] ‘to choose,’ [he: ˈkeras] ‘he chose,’ but [we: ˈkuron]
‘we chose;’

[ˈwesan] ‘to be,’ [he: ˈwɛs] ‘he was,’ but [we: ˈwɛ:ron] ‘we were.’

This alternation, Verner showed, corresponds to the alternation
in the position of the word-accent in similar Sanskrit paradigms,
as, in the verb-forms cognate with the above:

[ˈvartate:] ‘he turns, becomes,’ [va-ˈvarta] ‘he turned,’ but
[va-vr̩tiˈma] ‘we turned;’

*[ˈǰo:šati] ‘he enjoys,’ [ǰu-ˈǰo:ša] ‘he enjoyed,’ but [ǰu-ǰušiˈma]
‘we enjoyed;’

[ˈvasati] ‘he dwells,’ [u-ˈva:sa] ‘he dwelt,’ but [uːšiˈma] ‘we

This was so striking a confirmation of the hypothesis of regular
sound-change, that the burden of proof now fell upon the opponents
of the hypothesis: if the residual forms can show such a
correlation as this, we may well ask for very good reasons before
we give up our separation of forms into recognized correspondences
and remainders, and our principle of scanning residual forms for
new correspondences. We may doubt whether an observer who
was satisfied with a verdict of “sporadic sound-change” could
ever have discovered these correlations.

In a small way, the accidents of observation sometimes furnish
similar confirmations of our method. In the Central Algonquian
languages — for which we have no older records — we find the
following normal correspondences, which we may symbolize by
“Primitive Central Algonquian” reconstructed forms:

tableau Fox | Ojibwa | Menomini | Plains Cree | Primitive Central Algonquian


(1) Fox [kehkjɛ:wa] ‘he is old,’ Menomini [kečkiːw], PCA

(2) Fox [aškutɛ:wi] ‘fire,’ Ojibwa [iškudɛ:], Menomini
[esko:tɛ:w], Cree [iskute:w], PCA *[iškute:wi].359

(3) Fox [mahkesɛ:hi] ‘moccasin,’ Ojibwa [mahkizin], Menomini
[mahkɛ:sen], Cree [maskisin], PCA *[maxkesini].

(4) Fox [no:hkumesa] ‘my grandmother,’ Ojibwa [no:hkumis],
Menomini [no:hkumɛh], Cree [no:hkum], PCA *[no:hkuma].

(5) Fox [takeškawɛ:wa] ‘he kicks him,’ Ojibwa [tangiškawa:d],
Menomini [tahkɛ:skawɛ:w], Cree [tahkiskawerw], PCA
* [tankeška wɛ: wa].

Now, there is a residual morpheme in which none of these correspondences
holds good, namely the element which means ‘red’:

(6) Fox [mešusiwa] ‘he is red,’ Ojibwa [miškuzi], Menomini
[mɛhko:n], Cree [mihkusiw], PCA *[meçkusiwa].

Under an assumption of sporadic sound-change, this would
have no significance. After the sixth correspondence had been set
up, however, it was found that in a remote dialect of Cree, which
agrees in groups (1) to (5) with the Plains Cree scheme, the morpheme
for ‘red’ has the peculiar cluster [htk], as in [mihtkusiw]
‘he is red.’ In this case, then, the residual form showed a special
phonetic unit of the parent speech.

The assumption of regular (that is, purely phonemic) sound-change
is justified by the correlations which it uncovers; it is inconsistent
to accept the results which it yields and to reject it
whenever one wants a contradictory assumption (“sporadic
sound-change”) to “explain” difficult cases.

20. 9. The relation of our residual forms to factors of linguistic
history other than sound-change, is the crucial point in the dispute
about the regularity of sound-change. The neo-grammarians
could not claim, of course, that linguistic resemblances ever run
in regular sets. The actual data with which we work are extremely
irregular, — so irregular that centuries of study before the days
of Rask and Grimm had found no useful correlations. The neo-grammarians
did claim, however, that factors of linguistic change
other than sound-change will appear in the residual forms after
we have ruled out the correlations that result from sound-change.
Thus, Old English [a:] in stressed syllables appears in modern
English normally as [ow], as in boat (from Old English [ba:t]),
sore, whole, oath, snow, stone, bone, home, dough, goat, and many
other forms. In the residue, we find forms like Old English [ba:t] :
bait, Old English [ha:l] : hale, Old English [swa:n] ‘herdsman’ :
swain. Having found that Old English [a:] appears in modern
standard English as [ow], we assign the forms with the discrepant
360modern English [ej] to a residue. The forms in this residue are
not the results of a deviant, sporadic sound-change of Old English
[a:] to modern English [ej]; their deviation is due not to sound-change,
but to another factor of linguistic change. The forms
like bait, hale, swain are not the modern continuants of Old English
forms with [a:], but borrowings from Scandinavian. Old Scandinavian
had [ej] in forms where Old English had [a:]; Old Scandinavian
(Old Norse) said [stejnn, bejta, hejll, swejnn] where Old
English said [sta:n, ba:t, ha:l, swa:n]. The regularity of correspondence
is due, of course, to the common tradition from Primitive
Germanic. After the Norse invasion of England, the English
language took over these Scandinavian words, and it is the Old
Norse diphthong [ej] which appears in the deviant forms with
modern English [ej].

In cases like these, or in cases like Latin dentālis : English dental,
the opponents of the neo-grammarian hypothesis raise no objection,
and agree that linguistic borrowing accounts for the resemblance.
In many other cases, however, they prefer to say that
irregular sound-change was at work, and, strangely enough, they
do this in cases where only the neo-grammarian hypothesis yields
a significant result.

Students of dialect geography are especially given to this confusion.
In any one dialect we usually find an ancient unit phoneme
represented by several phonemes — as in the case of Old English
[o:] in modern English food, good, blood, and so on (§ 20.7). Often
one of these is like the old phoneme and the others appear to embody
one or more phonetic changes. Thus, in Central-Western
American English, we say gather with [ɛ], rather with [ɛ] or with
[a], and father always with [a]. Some speakers have [juw] in words
like tune, dew, stew, new; some have [uw] in the first three types,
but keep [juw] ordinarily after [n-]; others speak [uw] in all of them.
Or, again, if we examine adjacent dialects in an area, we find a gradation:
some have apparently carried out a sound-change, as when,
say, in Dutch, some districts in our Figure 6 have [y:] for ancient
[uː] in the words mouse and house; next to these we may find
dialects which have apparently carried out the change in some of
the forms, but not in others, as when some districts in our Figure 6
say [hy:s] with the changed vowel, but [muːs] with the unchanged;
finally, we reach a district where the changed forms are lacking,
such as, in Figure 6, the area where the old forms [muːs, huːs] are
361still being spoken. Under a hypothesis of sporadic sound-change,
no definite conclusions could be drawn, but under the assumption
of regular sound-change, distributions of this sort can at once be
interpreted: an irregular distribution shows that the new forms,
in a part or in all of the area, are due not to sound-change, but to
borrowing. The sound-change took place in some one center and,
after this, forms which had undergone the change spread from this
center by linguistic borrowing. In other cases, a community may
have made a sound-change, but the changed forms may in part be
superseded by unchanged forms which spread from a center which
has not made the change. Students of dialect geography make this
inference and base on it their reconstruction of linguistic and
cultural movements, but many of these students at the same time
profess to reject the assumption of regular phonetic change. If
they stopped to examine the implications of this, they would soon
see that their work is based on the supposition that sound-change
is regular, for, if we admit the possibility of irregular sound-change,
then the use of [hy:s] beside [muːs] in a Dutch dialect, or of [ˈraðr̩]
rather beside [ˈgɛˈðr̩] gather in standard English, would justify no
deductions about linguistic borrowing.

20. 10. Another phase of the dispute about the regularity of
sound-change concerns residual forms whose deviation is connected
with features of meaning. Often enough, the forms that deviate
from ordinary phonetic correlation belong to some clearly marked
semantic group.

In ancient Greek, Primitive Indo-European [s] between vowels
had been lost by sound-change. Thus, Primitive Indo-European
*[ˈgewso:] ‘I taste’ (Gothic [ˈkiwsa] ‘I choose’) appears in Greek
as [ˈgewo:] ‘I give a taste;’ Primitive Indo-European *[ˈgenesos]
‘of the kin’ (Sanskrit [ˈǰanasah]) appears as Greek [ˈgeneos], later
[ˈgenows]; Primitive Indo-European *[ˈe:sm̩] ‘I was’ (Sanskrit
[ˈa:sam]) appears in Greek as [˅e:a], later [˅e:].

Over against cases like these, there is a considerable residue of
forms in which an old intervocalic [s] seems to be preserved in
ancient Greek. The principal type of this residue consists of aorist-tense
(that is, past punctual) verb-forms, in which the suffix [-s-] of
this tense occurs after the final vowel of a root or verb-stem. Thus,
the Greek root [plew-] ‘sail’ (present tense [ˈplewo:] ‘I sail,’ paralleled
by Sanskrit [ˈplavate:] ‘he sails’) has the aorist form [ˈeplewsa]
‘I sailed;’ the Greek aorist [ˈetejsa] ‘I paid a penalty’ parallels
362Sanskrit [ˈača:jšam] ‘I collected;’ the Greek root [ste:-] ‘stand’
(present tense [ˈhiste:mi] ‘I cause to stand’) has the aorist form
[ˈeste:sa] ‘I caused to stand,’ parallel with Old Bulgarian [staxu]
‘I stood up,’ Primitive Indo-European type *[ˈesta:sm̩]; a Primitive
Indo-European aorist type *[ˈebhuːsm̩] (Old Bulgarian
[byxu] ‘I became’) is apparently represented by Greek [ˈephuːsa]
‘I caused to grow.’ Opponents of the neo-grammarian method
suppose that when intervocalic [s] was weakened and finally lost
during the pre-Greek period, the [s] of these forms resisted the
change, because it expressed an important meaning, namely that
of the aorist tense. A sound-change, they claim, can be checked in
forms where it threatens to remove some semantically important

The neo-grammarian hypothesis implies that sound-change is
unaffected by semantic features and concerns merely the habits of
articulating speech-sounds. If residual forms are characterized
by some semantic feature, then their deviation must be due not
to sound-change, but to some other factor of linguistic change —
to some factor which is connected with meanings. In our example,
the sound-change which led to the loss of intervocalic [s] destroyed
every intervocalic [s]; forms like Greek [ˈesteisa] cannot be continuants
of forms that existed before that sound-change. They
were created after the sound-change was past, as new combinations
of morphemes in a complex form, by a process which we call
analogic new combination or analogic change. In many forms where
the aorist-suffix was not between vowels, it had come unscathed
through the sound-change. Thus, a Primitive Indo-European
aorist *[ˈele:jkwsm̩] ‘I left’ (Sanskrit [ˈara:jkšam]) appears in
Greek, by normal phonetic development, as [ˈelejpsa]; Primitive
Indo-European *[eǰe:wksm̩] ‘I joined’ (Sanskrit [ˈaja:wkšam])
appears as Greek [ˈezewksa]; the Primitive Indo-European root
*[gews-] ‘taste’ (Greek present [ˈgewo:], cited above), combining
with the aorist-suffix, would give a stem *[ge:ws-s-]: as double
[ss] was not lost in pre-Greek, but merely at a later date simplified
to [s], the Greek aorist [ˈegewsa] ‘I gave a taste’ is the normal
phonetic type. Accordingly, the Greek language possessed the
aorist suffix [-s-]; at all times this suffix was doubtless combined
with all manner of verbal stems, and our aorists with the [-s-] between
vowels are merely combinations which were made after the
sound-change which affected [-s-] had ceased to work. On models
363like the inherited present-tense [ˈgewo:] with aorist [ˈegewsa], one
formed, for the present-tense [ˈplewo:], a new aorist [ˈeplewsa]. In
sum, the residual forms are not due to deflections of the process of
sound-change, but reveal to us, rather, a different factor of linguistic
change — namely, analogic change.

In much the same way, some students believe that sounds which
bear no important meaning are subject to excess weakening and
to loss by irregular sound-change. In this way they explain, for
instance, the weakening of will to [l] in forms like I'll go. The neo-grammarian
would attribute the weakening rather to the fact that
the verb-form in phrases like these is atonic: in English, unstressed
phonemes have been subjected to a series of weakenings and losses.

20. 11. The neo-grammarians define sound-change as a purely
phonetic process; it affects a phoneme or a type of phonemes either
universally or under certain strictly phonetic conditions, and is
neither favored nor impeded by the semantic character of the forms
which happen to contain the phoneme. The effect of sound-change,
then, as it presents itself to the comparatist, will be a set of regular
phonemic correspondences, such as Old English [sta:n, ba:n,
ba:t, ga:t, ra:d, ha:l]: modern English [stown, bown, bowt, gowt,
rowd, howl] stone, bone, boat, goat, road (rode), whole. However,
these correspondences will almost always be opposed by sets or
scatterings of deviant forms, such as Old English [ba:t, swa:n,
ha:l] versus modern English [bejt, swejn, hejl] bait, swain, hale,
because phonetic change is only one of several factors of linguistic
change. We must suppose that, no matter how minute and accurate
our observation, we should always find deviant forms, because,
from the very outset of a sound-change, and during its entire
course, and after it is over, the forms of the language are subject to
the incessant working of other factors of change, such as, especially,
borrowing and analogic combination of new complex forms. The
occurrence of sound-change, as defined by the neo-grammarians,
is not a fact of direct observation, but an assumption. The neo-grammarians
believe that this assumption is correct, because it
alone has enabled linguists to find order in the factual data, and
because it alone has led to a plausible formulation of other factors
of linguistic change.

Theoretically, we can understand the regular change of phonemes,
if we suppose that language consists of two layers of habit.
One layer is phonemic: the speakers have certain habits of voicing,
364tongue-movement, and so on. These habits make up the phonetic
system of the language. The other layer consists of formal-semantic
habits: the speakers habitually utter certain combinations
of phonemes in response to certain types of stimuli, and respond
appropriately when they hear these same combinations. These
habits make up the grammar and lexicon of the language.

One may conceivably acquire the phonetic habits of a language
without using any of its significant forms; this may be the case of
a singer who has been taught to render a French song in correct
pronunciation, or of a mimic who, knowing no French, can yet
imitate a Frenchman's English. On the other hand, if the phonemes
of a foreign language are not completely incommensurable
with ours, we may utter significant forms in this language without
acquiring its phonetic habits; this is the case of some speakers of
French and English, who converse freely in each others' languages,
but, as we say, with an abominable pronunciation.

Historically, we picture phonetic change as a gradual favoring
of some non-distinctive variants and a disfavoring of others. It
could be observed only by means of an enormous mass of mechanical
records, reaching through several generations of speakers. The
hypothesis supposes that such a collection — provided that we
could rule out the effects of borrowing and analogic change —
would show a progressive favoring of variants in some one direction,
coupled with the obsolescence of variants at the other extreme.
Thus, Old English and Middle English spoke a long mid
vowel in forms like gos ‘goose’ and ges ‘geese.’ We suppose that
during a long period of time, higher variants were favored and
lower variants went out of use, until, in the eighteenth century, the
range of surviving variants could be described as a high-vowel
type [uː, iː]; since then, the more diphthongal variants have been
favored, and the simple-vowel types have gone out of use.

The non-distinctive acoustic features of a language are at all
times highly variable. Even the most accurate phonetic record
of a language at any one time could not tell us which phonemes
were changing. Moreover, it is certain that these non-distinctive,
sub-phonemic variants are subject to linguistic borrowing (imitation)
and to analogic change (systematization). This appears
from the fact that whenever the linguist deals with a sound-change
— and certainly in some cases his documents or his observations
must date from a time very shortly after the occurrence of the
365change — he finds the results of the sound-change disturbed by
these other factors. Indeed, when we observe sub-phonemic variants,
we sometimes find them distributed among speakers or
systematized among forms, quite in the manner of linguistic
borrowing and of analogic change. In the Central-Western type
of American English, vowel-quantities are not distinctive, but
some speakers habitually (though perhaps not invariably) use
a shorter variant of the phoneme [a] before the clusters [rk, rp],
as in dark, sharp, and before the clusters [rd, rt] followed by a
primary suffix [-r̩, n̩-], as in barter, Carter, garden, marten (Martin).
Before a secondary suffix, [-r̩, -n̩], however, the longer variant is
used, as in starter, carter (‘one who carts’), harden; here the existence
of the simple words (start, cart, hard), whose [a] is not subject
to shortening, has led to the favoring of the normal, longer
variant. The word larder (not part of the colloquial vocabulary)
could be read with the shorter variant, but the agent-noun larder
(‘one who lards’) could be formed only with the longer type of
the [a]-phoneme. This distribution of the sub-phonemic variants
is quite like the results of analogic change, and, whatever its
origin, the distribution of this habit among speakers is doubtless
effected by a process of imitation which we could identify with
linguistic borrowing. If the difference between the two variants
should become distinctive, then the comparatist would say that
a sound-change had occurred, but he would find the results of
this sound-change overlaid, from the very start, by the effects
of borrowing and of analogic change.

We can often observe that a non-distinctive variant has become
entirely obsolete. In eighteenth-century English, forms like geese,
eight, goose, goat had long vowels of the types [iː, e:, uː, o:],
which since then have changed to the diphthongal types [ij, ej,
uw, ow]. This displacement has had no bearing on the structure
of the language; a transcription of present-day standard English
which used the symbols [iː, e:, uː, o:] would be perfectly adequate.
It is only the phonetician or acoustician who tells us that
there has been a displacement in the absolute physiologic and
acoustic configuration of these phonemes. Nevertheless, we can
see that the non-diphthongal variants, which at first were the
predominant ones, are today obsolete. The speaker of present-day
standard English who tries to speak a language like German
or French which has undiphthongized long vowels, has a hard
366time learning to produce these types. It is as hard for him to articulate
these acoustic types (which existed in English not so many
generations ago) as it is for the Frenchman or the German to
produce the English diphthongal types. The speaker learns only
with difficulty to produce speech-sounds that do not occur in
his native language, even though the historian, irrelevantly, may
assure him that an earlier stage of his language possessed these
very sounds.

We can speak of sound-change only when the displacement of
habit has led to some alteration in the structure of the language.
Most types of American English speak a low vowel [ɑ] in forms
like got, rod, not, where British English has kept an older mid-vowel
type [ɔ]. In some types of American standard English, this [ɑ]
is distinct from the [a] of forms like calm, far, pa — so that
bother does not rime with father, and bomb, is not homonymous
with balm: there has been no displacement of the phonemic
system. In other types of American standard English, however,
the two phonemes have coincided: got, rod, bother, bomb, calm,
far, pa, father, balm all have one and the same low vowel [a], and
we say, accordingly, that a sound-change has taken place. Some
speakers of this (as well as some of the other) type pronounce
bomb as [bom]: this form is due to some sort of linguistic borrowing
and accordingly cannot exhibit the normal correlation.

The initial clusters [kn-, gn-], as in knee, gnat, lost their stop
sound early in the eighteenth century: hereby knot and not, knight
and night, gnash and Nash became homonymous. English-speakers
of today learn only with difficulty to produce initial clusters like
these, as, say, in German Knie [kniː] ‘knee.’

In Dutch-German area, the Primitive Germanic phoneme [θ]
changed toward [ð] and then toward [d]; by the end of the Middle
Ages this [d] coincided, in the northern part of the area, with
Primitive Germanic [d]. Hence modern standard Dutch has initial
[d] uniformly, both in words like dag [dax] ‘day,’ doen [duːn]
‘do,’ droom [dro:m] ‘dream,’ where English has [d], and in words
like dik [dik] ‘thick,’ doom [do:rn] ‘thorn,’ drie [driː] ‘three,’ where
English has [θ]. The distinction has been entirely obliterated,
and could be re-introduced only by borrowing from a language
in which it has been preserved. Needless to say, the Dutchman
or North German has as hard a time learning to utter an English
[θ] as though this sound had never existed in his language.367

The favoring of variants which leads to sound-change is a
historical occurrence; once it is past, we have no guarantee of
its happening again. A later process may end by favoring the very
same acoustic types as were eliminated by an earlier change.
The Old and Middle English long vowels [iː, uː], as in [wiːn, huːs],
were eliminated, in the early modern period, by change toward
the diphthongal types of the present-day wine, house. At about
the same time, however, the Old and Middle English long mid
vowels, as in [ge:s, go:s], were being raised, so that eighteenth-century
English again had the types [iː, uː] in words like geese,
goose. The new [iː, uː] arrived too late to suffer the change to
[aj, aw] which had overtaken the Middle English high vowels.
Similarly, we must suppose that the pre-Greek speakers of the
generations that were weakening the phoneme [s] between vowels,
could learn only with difficulty to utter such a thing as a distinct
simple [s] in intervocalic position, but, after the change was over,
the simplification of long [ss] re-introduced this phonetic type, and
(doubtless independently of this) new combinations of the type
[ˈeste:sa] (§ 20.10) were again fully pronounceable. In this way,
we can often determine the succession (relative chronology) of
changes. Thus, it is clear that in pre-Germanic time, the Primitive
Indo-European [b, d, g] can have reached the types of Primitive
Germanic [p, t, k] only after Primitive Indo-European [p, t, k]
had already been changed somewhat in the direction of the types
of Primitive Germanic [f, θ, h] — for the actual Germanic forms
show that these two series of phonemes did not coincide (§ 20.2).368

Chapter 21
Types of Phonetic Change

21. 1. Phonetic change, as defined in the last chapter, is a change
in the habits of performing sound-producing movements. Strictly
speaking, a change of this kind has no importance so long as it
does not affect the phonemic system of the language; in fact,
even with perfect records at our command, we should probably
be unable to determine the exact point where a favoring of certain
variants, began to deserve the name of a historical change. At
the time when speakers of English began to favor the variants
with higher tongue-position of the vowels in words like gōs ‘goose’
and gēs ‘geese,’ the dislocation was entirely without significance.
The speakers had no way of comparing the acoustic qualities of
their vowels with the acoustic qualities of the vowels which their
predecessors, a few generations back, had spoken in the same linguistic
forms. When they heard a dialect which had not made
the change, they may have noticed a difference, but they could
have had no assurance as to how this difference had arisen. Phonetic
change acquires significance only if it results in a change of
the phonemic pattern. For instance, in the early modern period,
the Middle English vowel [ɛ:], as in sed [sɛ:d] ‘seed,’ was raised
until it coincided with the [e:] in ges [ge:s] ‘geese,’ and this coincidence
for all time changed the distribution of phonemes in the
forms of the language. Again, the Middle English short [e] in a
so-called “open” syllable — that is, before a single consonant followed
by another vowel, as in ete [ˈete] ‘eat’—was lengthened
and ultimately coincided with the long vowels just mentioned.
Accordingly, the phonemic structure of modern English is different
from that of medieval English. Our phoneme [ij] continues,
among others, these three older phonemes; we may note, especially,
that this coincidence has given rise to a number of

Old and Middle English [e:] has changed to modern [ij] in heel,
steel, geese, queen, green, meet (verb), need, keep.

Old and Middle English [ɛ:] has changed to modern [ij] in heal,
369meal (‘taking of food’), cheese, leave, clean, lean (adjective), street,
mead (‘meadow’), meet (adjective).

Old and Middle English [e] has changed to modern [ij] in steal,
meal (‘flour’), weave, lean (verb), quean, speak, meat, mete, eat,
mead (‘fermented drink’).

On the other hand, the restriction of this last change to a
limited phonetic position, has produced different phonemes in
forms that used to have the same phoneme: the old [e] was lengthened
in Middle English weve < weave, but not in Middle English
weft < weft. In the same way, a phonetic change which consisted
of shortening long vowels before certain consonant-clusters has
produced the difference of vowel between meadow (< Old English
[ˈmɛ:dwe]) and mead, or between kept (< Old English [ˈke:pte])
and keep.

A few hundred years ago, initial [k] was lost before [n]: the
result was a change in the phonemic system, which included such
features as the homonymy of knot and not, or of knight and nighi,
and the alternation of [n-] and [-kn-] in know, knowledge : acknowledge.

21. 2. The general direction of a great deal of sound-change
is toward a simplification of the movements which make up the
utterance of any given linguistic form. Thus, consonant-groups
are often simplified. The Old English initial clusters [hr, hl, hn,
kn, gn, wr] have lost their initial consonants, as in Old English
hring > ring, hlēapan > leap, hnecca > neck, cnēow > knee, gnagan
> gnaw, wringan > wring. The loss of the [h] in these groups
occurred in the later Middle Ages, that of the other consonants
in early modern time; we do not know what new factor intervened
at these times to destroy the clusters which for many centuries
had been spoken without change. The [h]-clusters are still spoken
in Icelandic; initial [kn] remains not only in the other Germanic
languages (as, Dutch knie [kniː], German Knie [kniː], Danish
[knɛ:ʔ], Swedish [kne:]), but also in the English dialects of the
Shetland and Orkney Islands and northeastern Scotland. The
[gn] persists almost as widely — in English, more widely; [wr-],
in the shape of [vr-], remains in Scandinavian, the northern part
of the Dutch-German area, including standard Dutch, and in
several scattered dialects of English. As long as we do not know
what factors led to these changes at one time and place but not
at another, we cannot claim to know the causes of the change —
370that is, to predict its occurrence. The greater simplicity of the
favored variants is a permanent factor; it can offer no possibilities
of correlation.

Simplification of final consonant-clusters is even more common.
A Primitive Indo-European *[pe:ts] ‘foot’ (nominative singular)
appears in Sanskrit as [pa:t] and in Latin as pes [pe:s]; a Primitive
Indo-European *[ˈbheronts] ‘bearing’ (nominative singular masculine)
appears in Sanskrit as [ˈbharan], and in Latin as ferens
[ˈferens], later [ˈfere:s]. It is this type of change which leads to
habits of permitted final (§ 8.4) and to morphologic alternations
of the type described in § 13.9. Thus, a Primitive Central Algonquian
*[axkehkwa] ‘kettle,’ plural *[axkehkwaki], reflected in
Fox [ahko:hkwa, ahko:hko:ki], loses its final vowel and part of
the consonant-cluster in Cree [askihk, askihkwak] and in Menomini
[ahkɛ:h, ahkɛ:hkuk], so that the plural-form in these languages
contains a consonant-cluster that cannot be determined
by inspection of the singular form. In English, final [ŋg] and
[mb] have lost their stop; hence the contrast of long : longer [lɔŋ —
ˈloŋgr̩], climb : clamber [klajm — ˈklɛmbr̩].

Sometimes even single final consonants are weakened or disappear.
In pre-Greek, final [t, d] were lost, as in Primitive Indo-European
*[tod] ‘that,’ Sanskrit [tat]: Greek [to]; final [m] became
[n], as in Primitive Indo-European *[ǰuˈgom] ‘yoke,’ Sanskrit
[juˈgam]: Greek [zuˈgon]. The same changes seem no have occurred
in pre-Germanic. Sometimes all final consonants are lost and
there results a phonetic pattern in which every word ends in a
vowel. This happened in pre-Slavic, witness forms like Old Bulgarian
[to] ‘that,’ [igo] ‘yoke.’ It is a change of this sort that
accounts for morphologic situations like that of Samoan (§ 13.9);
a Samoan form like [inu] ‘drink’ is the descendant of an older
*[inum], whose final consonant has been kept in Tagalog [iˈnum].

When changes of this sort appear at the beginning or, more often,
at the end of words, we have to suppose that the languages in which
they took place had, at the time, some phonetic marking of the
word-unit. If there were any forms in which the beginning or the
end of a word had not the characteristic initial or final pronunciation,
these forms would not suffer the change, and would survive
as sandhi-forms. Thus, in Middle English, final [n] was lost, as in
eten > ete ‘eat,’ but the article an before vowels must have been
pronounced as if it were part of the following word — that is,
371without the phonetic peculiarities of final position — so that the
[n] in this case was not lost (like a final [n]), but preserved (like a
medial [n]): a house but an arm. Latin vōs ‘ye’ gives French vous
[vu], but Latin phrase-types like vōs amātis ‘ye love’ are reflected
in the French sandhi-habit of saying vous aimez [vuz eme]. Latin
est ‘he is’ gave French est [e] ‘is,’ but the phrase-type of Latin
est ille? ‘is that one?’ appears in the French sandhi-form in est-il?
[ɛt i?] ‘is he?’ In the same way, a Primitive Indo-European
*[ˈbheronts] is reflected not only in Sanskrit [ˈbharan], above cited,
but also in the Sanskrit habit of adding a sandhi [s] when the next
word began with [t], as in [ˈbharans ˈtatra] ‘carrying there.’

21. 3. Simplification of consonant-clusters is a frequent result
of sound-change. Thus, a pre-Latin *[ˈfulgmen] ‘flash (of lightning)’
gives a Latin fulmen. Here the group [lgm] was simplified
by the change to [lm], but the group [lg], as in fulgur ‘flash,’ was
not changed, and neither was the group [gm], as in agmen ‘army.’
In describing such changes, we speak of the conditions as conditioning
(or causing factors) and say, for instance, that one
of these was absent in cases like fulgur and agmen, where the
[g], accordingly, was preserved. This form of speech is inaccurate,
since the change was really one of [lgm] to [lm], and cases like
fulgur, agmen are irrelevant, but it is often convenient to use these
terms. The result of a conditioned change is often a morphologic
alternation. Thus, in Latin, we have the suffix -men in agere ‘to
lead’: agmen ‘army’ but fulgere ‘to flash’: fulmen ‘flash (of lightning).’
Similarly, pre-Latin [rkn] became [rn]; beside pater ‘father’:
paternus ‘paternal,’ we have quercus ‘oak’ : quernus ‘oaken.’

Quite commonly, clusters change by way of assimilation: the
position of the vocal organs for the production of one phoneme is
altered to a position more like that of the other phoneme. The
commoner case is regressive assimilation, change of the prior

Thus, the voicing or unvoicing of a consonant is often altered
into agreement with that of a following consonant; the [s] of
goose and house has been voiced to [z] in the combinations gosling,
husband. This, again, may give rise to morphologic alternations.
In the history of Russian the loss of two short vowels (I shall
transcribe them as [i] and [u]) produced consonant-clusters; in
these clusters a stop or spirant was then assimilated, as to voicing,
to a following stop or spirant. The old forms can be seen in Old
372Bulgarian, which did not make the changes in question. Thus
*[ˈsvatiba] ‘marriage’ gives Russian [ˈsvadba]; compare Russian
[svat] ‘arranger of a marriage.’ Old Bulgarian [otube:žati] ‘to
run away’ appears in Russian as [odbeˈzat]; compare the simple Old
Bulgarian [otu] ‘from, away from’ : Russian [ot]. On the other
hand, Old Bulgarian [podukopati] ‘to undermine’ appears in
Russian as [potkoˈpat]; contrast Old Bulgarian [podu igo] ‘under
the yoke’ : Russian [ˈpod igo].

The assimilation may affect the action of the velum, tongue, or
lips. If some difference between the consonants is kept, the assimilation
is partial; thus in pre-Latin [pn] was assimilated to [mn], as
in Primitive Indo-European *[ˈswepnos] ‘sleep,’ Sanskrit [ˈsvapnah]
: Latin somnus. If the difference entirely disappears, the
assimilation is total, and the result is a long consonant, as in
Italian sonno [ˈsɔnno]. Similarly, Latin octō ‘eight’ > Italian otto
[ˈɔtto]; Latin ruptum ‘broken’ > Italian rotto [ˈrotto].

In progressive assimilation the latter consonant is altered. Thus,
pre-Latin *[kolnis] ‘hill’ gives Latin collis; compare Lithuanian
[ˈka:lnas] ‘mountain.’ Our word hill underwent the same change
[In] > [ll] in pre-Germanic; witness Primitive Indo-European
*[pl̩:ˈnos] ‘full,’ Sanskrit [puːrˈnah], Lithuanian [ˈpilnas] : Primitive
Germanic *[ˈfollaz], Gothic fulls, Old English full, or Primitive
Indo-European *[ˈwl̩:na:] ‘wool,’ Sanskrit [ˈuːrna:], Lithuanian
[ˈvilna] : Primitive Germanic *[ˈwollo:], Gothic wulla, Old English

21. 4. A great many other changes of consonants can be viewed
as assimilative in character. Thus, the unvoicing of final consonants,
which has occurred in the history of various languages,
can be viewed as a sort of regressive assimilation: the open position
of the vocal chords which follows upon the end of speech, is anticipated
during the utterance of the final consonant. Thus, many
dialects of the Dutch-German area, including the standard languages,
have unvoiced all final stops and spirants; the result is an
alternation of unvoiced finals with voiced medials (§ 13.9):

Old High German tag ‘day’ > New High German Tag [ta:k],
but, plural, taga ‘days’ > Tage [ˈta:ge], with unchanged [g];

Old High German bad ‘bath’ > New High German Bad [ba:t],
but, genitive case, bades > Bades [ˈba:des];

Old High German gab ‘(he) gave’ >New High German gab
[ga:p], but, plural, gābun ‘(they) gave’ > gaben [ˈga:ben].373

The voiced consonant may be preserved in sandhi — that is, in
traditional phrase-types where it did not come at the end of
speech. This does not happen in standard German; here the final-form
has been carried out for every word-unit. In Russian, however,
we have not only the final-form, by which an old [podu],
after loss of the vowel, became [pot], but also phrasal types like
[ˈpod igo] ‘under the yoke.’ There is a type of Dutch pronunciation
where an old hebbe ‘(I) have’ appears, after loss of the final vowel,
not only in the final-form with [-p], as in ik heb [ek ˈhep], but also
in the phrasal sandhi-type, heb ek? [ˈheb ek?] ‘have I?’ This is the
origin of reminiscent sandhi (§ 12.5).

A very common type of change is the weakening of consonants
between vowels or other open sounds. This, too, is akin to assimilation,
since, when the preceding and following sounds are open
and voiced, the less marked closure or the voicing of a stop or
spirant represents an economy of movement. The change which
gave rise to the American English voiced tongue-flip variety of
[t], as in water, butter, at all (§ 6.7), was surely of this sort. Latin
[p, t, k] between vowels are largely weakened in the Romance
languages: Latin rīpam ‘bank, shore,’ sētam ‘silk,’ focum ‘hearth’
appear in Spanish as riba, seda, fuego ‘fire,’ where the [b, d, g] are
largely spirant in character, and in French as rive, soie, feu [riːv,
swa, fø]. Some languages, such as pre-Greek, lose sounds like
[s, j, w] between vowels. The Polynesian languages and, to some
extent, the medieval Indo-Aryan languages, show a loss of the old
structure of medial consonants, much like that in the French forms
just cited. In the history of English, loss of [v] is notable, as in
Old English [ˈhɛvde, ˈhavok, ˈhla:vord, ˈhla:vdije, ˈhe:avod,
ˈnavoga:r] > modern had, hawk, lord, lady, head, auger; this change
seems to have occurred in the thirteenth century.

If the conditioning factors are removed by subsequent change,
the result is an irregular alternation. In this way, arose, for example,
the sandhi-alternation of initial consonants in Irish (§ 12.4).
In the history of this language, stops between vowels were weakened
to spirants, as in Primitive Indo-European *[ˈpibo:mi] ‘I
drink,’ Sanskrit [ˈpiba:mi]: Old Irish ebaim [ˈevim]. Apparently
the language at this stage gave little phonetic recognition to the
word-unit, and carried out this change in close-knit phrases,
changing, for instance, an *[eso bowes] ‘his cows’ (compare
Sanskrit [aˈsja ˈga:vah]) to what is now [a va:], in contrast with
374the absolute form [ba:] ‘cows.’ This type of sandhi is preserved in
a limited number of cases, as, in our instance, after the pronoun
[a] ‘his.’ In the same way, [s] between vowels was weakened to
[h] and then lost: a Primitive Indo-European *[ˈsweso:r] ‘sister,’
Sanskrit [ˈsvasa:], giving first, presumably, *[ˈsweho:r], and then
Old Irish siur. Final [s] similarly was lost: a Gallic tarbos ‘bull’
appears in Old Irish as tarb. We have to suppose, now, that the
change [s > h] between vowels took place also in close-knit phrases,
so that an *[esa:s o:wjo] ‘her egg’ (compare Sanskrit [aˈsja:h]
‘her,’ with [-h] from [-s]) resulted in a modern [a huv] ‘her egg,’
in contrast with the independent [uv] ‘egg’ — again, a habit preserved
only in certain combinations, as after the word for ‘her.’
Similarly, [m] was first changed to [n] and then lost at the end of
words, but between vowels was preserved; both treatments appear
in *[neme:tom] ‘holy place,’ Old Gallic [neme:ton], Old Irish
nemed. At the stage where [-m] had become [-n], an old *[sen-to:m
o:wjo:m] ‘of these eggs’ (compare the Greek genitive plural
[˅to:n]) gave what is now [na nuv], in contrast with the absolute
[uv] ‘egg.’ To a similar, but more complicated development we owe
the sandhi-alternant with initial [t], as in [an tuv] ‘the egg;’
ultimately this is due to the fact that the Primitive Indo-European
nominative-accusative singular neuter pronoun-forms ended in
[d], as Sanskrit [tat] ‘that,’ Latin id ‘it.’

We may interpret the pre-Germanic change discovered by
Verner (§§ 18.7; 20.8) as a weakening of unvoiced spirants [f, θ,
h, s] between musical sounds to voiced [v, ð, ɣ, z]; then the restriction
of the change to cases where the preceding vowel or
diphthong was unstressed is subject to a further interpretation
of the same sort: after a loudly stressed vowel there is a great
amount of breath stored up behind the vocal chords, so that their
opening for an unvoiced spirant is easier than their closure for a
voiced. We cannot view these interpretations as correlating
(“causal”) explanations, however, for enough languages keep
unvoiced spirants intact between vowels, while others change
them to voiced regardless of high stress on a preceding vowel.
Here, too, the conditioning factor was afterwards removed by other
changes: in an early pre-Germanic *[ˈwerθonon] ‘to become’ versus
*[wurðuˈme] ‘we became,’ the alternation [θ:ð] depended on
the place of the stress; later, when the stress had changed to the
first syllable of all words, the alternation in Primitive Germanic
375*[ˈwerθanan — ˈwurdume], Old English [ˈweorθan — ˈwurdon], was
an arbitrary irregularity, just as is the parallel was : were, from
Primitive Germanic *[ˈwase — ˈwe:zume], in modern English.
A similar change occurred much later in the history of English;
it accounts for such differences as luxury : luxurious [ˈlokšr̩ij
—logˈžuwrjos] in a common type of pronunciation, and for the
two treatments of French [s] in forms like possessor [poˈzesr̩].
This change involved the voicing of old [s] after an unstressed
vowel in suffixes, as in glasses, misses, Bess's; a few forms like
dice (plural of die) and pence show the preservation of [s] after
a stressed vowel. Immediately after this change the stressed
forms must have been off [of], with [wiθ], is [is], his [his], and the
atonic forms of [ov] and [wið, iz, hiz,] but this alternation has
been destroyed: off and of have been redistributed by analogic
change, [wi0] survives as a variant of [wið], and the [s]-forms of
is and his have fallen into disuse.

21. 5. Consonants are often assimilated to the tongue-position
of preceding or following vowels. The commonest case is the assimilation
especially of dentals and velars to a following front
vowel; this is known as palatalization. A change of this kind which
did not cause phonemic alterations, must have occurred not too
long ago in English, for phoneticians assure us that we make the
tongue-contact of [k, g] farther forward before a front vowel, as
in kin, keep, kept, give, geese, get, than before a back vowel, as in
cook, good. In pre-English there occurred a change of the same
sort which led to alteration of the phonemic structure. To begin
with, the palatalized form of [g] — presumably this phoneme had
a spirant character — coincided with another phoneme, [j]. The
change in phonemic distribution appears plainly when we compare
the cognate forms from North German (Old Saxon), where the
old phonemic distribution remained intact:

tableau North German | Pre-English | Old English | Modern English | gold | good | yield | yarn | yoke | year376

Another way in which the pre-English palatalization in time
affected the structure of the language, was by the obscuration of
the conditioning factor. The back vowels [o, u], which did not
affect a preceding velar, were changed, under certain conditions,
to front vowels [ø, y] and later to [e, i], which coincided with old
front vowels that had effected palatalization. Hence, in the
later stages of English, both palatalized and unpalatalized velars
occurred before front vowels.

Palatalized velars, before old front vowels:

tableau Pre-English | Old English | Modern English | cheese | chin | yield | yarn

Unpalatalized velars, before new front vowels:

tableau Pre-English | Old English | Modern English | keen | kin | geese | gild

A third factor of the same kind was the loss, by later sound-change,
of the conditioning feature, — that is, of the front vowel
[e, i, j] which had caused the palatalization:

Palatalized velars, followed, at the critical time, by a front

tableau Pre-English | Old English | Modern English | drench | stitch | singe | bridge

Unpalatalized velars, not followed by front vowel:

tableau Pre-English | Old English | Modern English | drink | stick | sing | frog377

The sound-change which we call palatalization changes consonants
at first to varieties which the phonetician calls palatalized;
the modern English forms in our preceding examples, with their
[č, ǰ, j], show us that these palatalized types may undergo further
changes. These, in fact, are extremely common, although their
direction varies. In the case of both velars and dentals, affricate
types [č, ǰ] and sibilant types, both abnormal [š, ž] and normal [s, z],
are fairly frequent. In modern English we have a development
of [tj > č, dj > ǰ, sj > š, zj > ž], as in virtue, Indian, session,
vision [ˈvr̩čuw, ˈinǰn̩,ˈ ˈsešn̩, ˈvižn̩]; more formal variants, such as
[vr̩tjuw, ˈindjn̩], have arisen by later changes. The Romance
languages exhibit a great variety of development of palatalized

tableau Latin | Italian | French | Spanish | ‘hundred’ | ‘nation’

Part of the French area has a palatalization of [k] before [a];
in the Middle Ages, when English borrowed many French words,
this had reached the stage of [č], so that a Latin type like cantare
[kanˈta:re] ‘to sing’ > Old French chanter [čanˈte:r] appears in
English as chant; similarly, Latin cathedram [ˈkatedram] appears
as chair; Latin catenam [kaˈte:nam] as chain; Latin cameram [ˈkameram]
as chamber. In modern standard French, further change
of this [č] has led to [š]: chanter, chaire, chaîne, chambre [šɑnte,
šɛ:r, šɛ:n, šɑnbr].

Palatalization has played a great part in the history of the
Slavic languages: it has occurred at different times with different
results, and has affected every type of consonant, including even

A case of palatalization whose causing factor was obscured by
later change, played an important part in the development of
Indo-European studies. In the Indo-Iranian languages a single
vowel-type [a] corresponds to the three types [a, e, o] of the other
Indo-European languages. Thus, Latin ager ‘field,’ equos ‘horse,’
octō ‘eight’ are cognate with Sanskrit [ˈaǰrah, ˈaçvah, ašˈta:w].
For a long time students believed that the Indo-Iranian languages
had here preserved the Primitive Indo-European state of affairs,
378and that the diverse vowels of the European languages were due
to later change, made during a common pre-European period.
Before the [a] of the Indo-Iranian languages, Primitive Indo-European,
velars [k, g] appeared sometimes unchanged and sometimes
as [č, ǰ]. In the 1870's several students independently saw
that these latter reflexes are probably due to palatalization, and,
in fact, correlate fairly well with the cases where the European
languages have [e]. Thus we find, with back vowels in the languages
of Europe and velar stops in Indo-Iranian, correspondences

Primitive Indo-European *[kwod], Latin quod [kwod] ‘what’:
Sanskrit kat- (as first member in compounds);

Primitive Indo-European *[gwo:ws], Old English cu [kuː] ‘cow’:
Sanskrit [ga:wh].

On the other hand, with the front vowel [e] in the languages of
Europe and affricates instead of velar stops in Indo-Iranian, we
find correspondences like

Primitive Indo-European *[kwe], Latin que [kwe] ‘and’ : Sanskrit

Primitive Indo-European *[gwe:nis], Gothic qens [kwe:ns] ‘wife’:
Sanskrit [-ǰa:nih] (final member in compounds).

From cases like these we conclude that the uniform [a] of Indo-Iranian
is due to a later development: in pre-Indo-Iranian there
must have been an [e] distinct from the other vowels, and this
[e] must have caused palatalization of preceding velar stops. Since
this [e], moreover, agrees with the [e] of the European languages,
the distinction must have existed in Primitive Indo-European, and
cannot be due to a joint innovation by the languages of Europe.
This discovery put an end to the notion of a common parent speech
intermediate between Primitive Indo-European and the European
(as opposed to the Indo-Iranian) languages.

21. 6. The weakening or loss of consonants is sometimes accompanied
by compensatory lengthening of a preceding vowel. The
Old English combination [ht], preserved to this day in northern
dialects, has lost the [h] and lengthened the preceding vowel in
most of the area. Thus, Old English niht [niht, nixt] ‘night,’ modern
Scotch [nixt, next], became [niːt], whence modern night [najt].
Loss of a sibilant before voiced non-syllabics with compensatory
lengthening of a vowel is quite common, as in pre-Latin *[ˈdis-lego:]
‘I pick out, I like’ > Latin dīlīgo (compare dis- in dispendō ‘I
379weigh out,’ and legō ‘I pick, gather’); early Latin cosmis ‘kind’
> Latin cōmis; pre-Latin *[ˈkaznos] ‘gray-haired’ > Latin cānus
(compare, in Paelignian, a neighboring Italic dialect, casnar ‘old
man’); Primitive Indo-European *[nisdos] ‘nest’ (compare English
nest) > Latin nīdus.

If the lost consonant is a nasal, the preceding vowel is often
nasalized, with or without compensatory lengthening and other
changes. This is the origin of the nasalized vowels of many languages,
as of French: Latin cantāre > French chanter [šɑnte],
Latin centum > French cent [sɑn], and so on. The morphology of
Old Germanic shows parallel forms with and without nasal, such
as Gothic [ˈbringan—ˈbra:hta] ‘bring, brought,’ [ˈθankjan —
ˈθa:hta] ‘think, thought.’ The forms without [n] all have an [h]
immediately following a long vowel. The suspicion that in these
forms an [n] has been lost with compensatory lengthening, is
confirmed by a few comparisons with other Indo-European languages,
such as Latin vincere ‘to conquer’ : Gothic [ˈwiːhan] ‘to
fight.’ Further, we have a twelfth-century Icelandic grammarian's
statement that in his language forms like [θe:l] ‘file’ (from *[ˈθinhlo:))
had a nasalized vowel. In Old English, the [a:] of the other
Germanic languages, in forms like these, is represented by [o:],
as in [ˈbro:hte] ‘brought,’ [ˈθo:hte] ‘thought.’ We have reason to
believe that this divergent vowel quality is a reflex of older nasalization,
because in other cases also, Old English shows us an
[o:] as a reflex of an earlier nasalized [a]. The loss of [n] before [h]
occurred in pre-Germanic; before the other unvoiced spirants
[f, s, θ] an [n] remained in most Germanic dialects, but was lost,
with compensatory lengthening, in English, Frisian, and some of
the adjacent dialects. In these cases, too, we find an [o:] in Old
English as the reflex of a lengthened and nasalized [a]. Thus, the
words five, us, mouth, soft, goose, other appear in the oldest German
documents as [finf, uns, mund, sanfto, gans, ˈander] (with [d] as
reflex of an old [θ]), but in Old English as [fiːf, uːs, muːθ, ˈso:fte,
go:s, ˈo:ðer].

When a consonant has been lost between vowels, the resulting
succession of vowels often suffers contraction into a single vowel or
diphthongal combination. Our earliest English records still show
us an [h] between vowels, but very soon afterward this h disappears
from the texts, and single vowels are written. Thus, the word toe
appears first as tahæ, presumably [ˈta:hɛ], but soon as ta [ta:];
380a pre-English type *[ˈθanho:n] ‘clay’ appears first as thohæ
[ˈθo:hɛ], then as [θo:]; Gothic [ˈahwa] ‘river’ (cognate with Latin
aqua ‘water’) is paralleled by Old English ea [e:a], from pre-English
*[ˈahwu]; Gothic [ˈsehwan] ‘to see’ is matched by Old
English seon [se:on].

21. 7. Vowels are often assimilated to vowels that precede or
follow in the next syllable. During the early Middle Ages, changes
of this kind occurred in several Germanic dialects. These changes
in the Germanic languages are known by the name of umlaut; somewhat
confusingly, this term is applied also to the resultant grammatical
alternations. The commonest type of umlaut is the partial
assimilation of a stressed back vowel to a following [i, j]. The
resulting alternations, after the loss of the conditioning [i, j], became
purely grammatical:

tableau Pre-English | Old English | Modern English | gold | gild | mouse | mice | foot | feet | goose | geese | drank | drench 129

Old Norse had also other types of umlaut, such as assimilation
of [a] toward the back-vowel quality of a following [u], as in
*[ˈsaku] ‘accusation’ (compare Old English sacu ‘dispute’) > Old
Norse [sɔk]. Similar changes, supplemented, no doubt by regularizing
new-formations, must have led to the vowel-harmony that
prevails in Turco-Tartar and some other languages (§ 11.7).

The effect of simplification appears most plainly in shortening
and loss of vowels. In the final syllables of words, and especially
in final position, this occurs in all manner of languages. Among the
Central Algonquian languages, Fox alone has kept the final vowels:
Primitive Central Algonquian *[eleniwa] ‘man’ > Fox [neniwa],
Ojibwa [inini], Menomini [enɛ:niw], Plains Cree [ijiniw]. Certain
381types of two-syllable words are exempt from this shortening:
*[ehkwa] ‘louse’ > Fox [ehkwa], Ojibwa [ihkwa], Menomini
[ehkuah], Cree [ihkwa].

Languages with strong word-stress often weaken or lose their
unstressed vowels. The loss of final vowels, as in Old English
(ic) singe > (I) sing, is known as apocope; that of medial vowels,
as in Old English stānas > stones [stownz], as syncope. The contrast
between the long forms of Primitive Germanic, the shorter
forms of Old English, and the greatly reduced words of modern
English, is due to a succession of such changes. Thus, a Primitive
Indo-European *[ˈbheronom] ‘act of bearing,’ Sanskrit [ˈbharanam],
Primitive Germanic *[ˈberanan], gives Old English beran,
Middle English bere, and then modern (to) bear. The habit of
treating certain words in the phrase as if they were part of the
preceding or following word, was inherited from Primitive Indo-European;
when, in pre-Germanic time, a single high stress was
placed on each word, these atonic forms received none; later, the
weakening of unstressed vowels led to sandhi-variants, stressed
and unstressed, of such words. Weakenings of this kind have
occurred over and over again in the history of English, but the
resultant alternations have been largely removed by re-formations
which consisted either of using the full forms in unstressed positions,
or of using the weakened forms in stressed positions. Our
on, for instance, was in the medieval period the unweakened form;
the weakened form of this word was a, as in away, from Old
English on weg [on ˈwej]; this weakened form survives only in a
limited number of combinations, such as away, ashore, aground,
aloft, and the unweakened on is now used in atonic position, as in
on the table, but has here been subjected to a new weakening, which
has resulted in unstressed [on] beside stressed [ɑn], as in go on
[gow ˈɑn]. In contrast with this, our pronoun I, which we use in
both stressed and unstressed positions, reflects an old unstressed
form, in which the final consonant of Old English ic has been lost;
the old stressed form survives in the [ič] ‘I’ of a few local dialects.
These changes have left their mark in the unstressed sandhi-variants
of many words, such as is, but [z] in he's here; will, but
[l] in I'll go; not, but [n̩t] in isn't; and in the weakened forms of
some unstressed compound members: man, but [-mn̩] in gentleman;
swain but [-sn̩] in boatswain. The same factor accounts for the
shortness of French words compared to Latin; as in centum > cent
382[sɑn]; since the time of these shortenings, however, French has lost
the strong word-stress and ceased shortening its forms.

If a language goes through this kind of change at a time when
morphologically related forms stress different syllables, the result
may be an extremely irregular morphology. We can see the
beginnings of this in our foreign-learned vocabulary, which stresses
different syllables in different derivatives: angel [ˈejnǰl̩], but
angelic [ɛnˈǰelik]. In Primitive Germanic the prefixes were unstressed
in verb-forms but stressed in most other words; the
weakenings that ensued broke up some morphologic sets, such as

pre-English *[bi-ˈha:tan] ‘to threaten’ > Old English behatan
[beˈha:tan], but

pre-English *[ˈbi-ha:t] ‘a threat’ > Old English beot [be:ot].

A similar process rendered the morphology and, as to sandhi,
the syntax of Old Irish extremely irregular:

pre-Irish *[ˈbereti] ‘he bears’ > Old Irish berid [ˈberið];

pre-Irish *[eks ˈberet] ‘he bears out, brings forth’ > Old Irish
asbeir [asˈber] ‘he says;’

pre-Irish *[ne esti ˈeks beret] ‘not it-is that-he-forth-brings’
(that is, ‘he does not bring forth’) > Old Irish nī epir [niː ˈepir]
‘he does not say.’

21. 8. Some changes which superficially do not seem like weakenings
or abbreviations of movement, may yet involve a simplification.
In a good many languages we find an intermediate consonant
arising in a cluster. A Primitive Indo-European [sr] appears
as [str] in Germanic and in Slavic; thus, Primitive Indo-European
*[srow-] (compare Sanskrit [ˈsravati] ‘it flows’) is reflected in
Primitive Germanic *[ˈstrawmaz] ‘stream,’ Old Norse [strawmr],
Old English [stre:am], and in Old Bulgarian [struja] ‘stream.’
English, at more than one time, has inserted a [d] in the groups
[nr, nl] and a [b] in the groups [mr, ml]: Old English [ˈθunrian]
> (to) thunder; Old English [ˈalre] (accusative case) > alder;
Gothic has [ˈtimrjan] ‘to construct’ as well as [ˈtimbrjan], but
Old English has only [ˈtimbrian] and [jeˈtimbre] ‘carpentry-work,’
whence modern timber; Old English [ˈθymle] > thimble. These
changes involve no additional movement, but merely replace
simultaneous movements by successive. To pass from [n] to [r],
for instance, the speaker must simultaneously raise his velum
and move his tongue from the closure position to the trill position:383

image [n] | [r] | velum lowered | velum raised | dental closure | trill position

If, with a less delicate co-ordination, the velum is raised before
the change of tongue-position, there results a moment of unnasalized
closure, equivalent to the phoneme [d]:

image [n] | [d] | [r] | velum lowered | velum raised | dental closure | trill position

The second of these performances is evidently easier than the

In other cases, too, an apparent lengthening of a form may be
viewed as lessening the difficulty of utterance. When a relatively
sonorous phoneme is non-syllabic, it often acquires syllabic function;
this change is known by the Sanskrit name of samprasarana.
Thus, in sub-standard English, elm [elm] has changed to [ˈelm̩].
This is often followed by another change, known as anaptyxis,
the rise of a vowel beside the sonant, which becomes non-syllabic.
Primitive Indo-European *[agros] ‘field’ gives pre-Latin *[agr];
in this the [r] must have become syllabic, and then an anaptyctic
vowel must have arisen, for in the historical Latin form ager
[ˈager] the e represents a fully formed vowel. Similarly, Primitive
Germanic forms like *[ˈakraz] ‘field,’ *[ˈfoglaz] ‘bird,’ *[ˈtajknan]
‘sign,’ *[ˈmajθmaz] ‘precious object’ lost their unstressed
vowels in all the old Germanic dialects. The Gothic forms [akrs,
fugls, tajkn, majθms] may have been monosyllabic or may have
had syllabic sonants; anaptyxis has taken place in the Old English
forms [ˈɛker, ˈfugol, ˈta:ken, ˈma:ðom], though even here
spellings like fugl are not uncommon.

Another change which may be regarded as a simplification
occurs in the history of some stress-using languages: the quantities
of stressed vowels are regulated according to the character
of the following phonemes. Generally, long vowels remain long
and short vowels are lengthened in “open” syllables, that is, before
a single consonant that is followed by another vowel; in other
positions, long vowels are shortened and short ones kept short.
Thus, Middle English long vowels remained long in forms like
clene [ˈklɛ:ne] > clean, kepe [ˈke:pe] > keep, mone [ˈmo:ne] >
moon, but were shortened in forms like clense > cleanse, kepte
> kept, mon(en)dai > Monday: and short vowels were lengthened
384in forms like weve [ˈweve] > weave, stele [ˈstele] > steal,
nose [ˈnose] > nose, but stayed short in forms like weft, stelth >
stealth, nos(e)thirl > nostril. In some languages, such as Menomini,
we find a very complicated regulation of long and short vowels
according to the preceding and following consonants and according
to the number of syllables intervening after the last preceding
long vowel.

The complete loss of quantitative differences, which occurred,
for instance, in medieval Greek and in some of the modern Slavic
languages, makes articulation more uniform. The same can be
said of the abandonment of distinctions of syllable-pitch, which
has occurred in these same languages; similarly, the removal of
word-accent uniformly to some one position such as the first
syllable, in pre-Germanic and in Bohemian, or the next-to-last,
in Polish, probably involves a facilitation.

In the same sense, the loss of a phonemic unit may be viewed
as a simplification. Except for English and Icelandic, the Germanic
languages have lost the phoneme [θ] and its voiced development
[ð]; the reflexes coincide in Frisian and in Scandinavian largely
with [t], as in Swedish torn [to:rn] : thorn, with the same initial
as tio [˅tiːe] : ten, and in the northern part of the Dutch-German
area with [d], as in Dutch doorn [do:rn] : thorn, with the same initial
as doen [duːn] : do. Old English [h] before a consonant, as in
niht ‘night,’ or in final position, as in seah ‘(I) saw,’ was acoustically
doubtless an unvoiced velar or palatal spirant; in most of
the English area this sound has been lost or has coincided with
other phonemes.

21. 9. Although many sound-changes shorten linguistic forms,
simplify the phonetic system, or in some other way lessen the labor
of utterance, yet no student has succeeded in establishing
a correlation between sound-change and any antecedent
phenomenon: the causes of sound-change are unknown. When
we find a large-scale shortening and loss of vowels, we feel safe
in assuming that the language had a strong word-stress, but many
languages with strong word-stress do not weaken the unstressed
vowels; examples are Italian, Spanish, Bohemian, Polish. The
English change of [kn-, gn-] to [n-] seems natural, after it has
occurred, but why did it not occur before the eighteenth century,
and why has it not occurred in the other Germanic

Every conceivable cause has been alleged: “race,” climate,
topographic conditions, diet, occupation and general mode of
life, and so on. Wundt attributed sound-change to increase in the
rapidity of speech, and this, in turn, to the community's advance
in culture and general intelligence. It is safe to say that we speak
as rapidly and with as little effort as possible, approaching always
the limit where our interlocutors ask us to repeat our utterance,
and that a great deal of sound-change is in some way connected
with this factor. No permanent factor, however, can account for
specific changes which occur at one time and place and not at
another. The same consideration holds good against the theory
that sound-change arises from imperfections in children's learning
of language. On the other hand, temporary operation of factors
like the above, such as change of habitat, occupation, or diet, is
ruled out by the fact that sound-changes occur too often and
exhibit too great a variety.

The substratum theory attributes sound-change to transference
of language: a community which adopts a new language will
speak it imperfectly and with the phonetics of its mother-tongue.
The transference of language will concern us later; in the present
connection it is important to see that the substratum theory can
account for changes only during the time when the language
is spoken by persons who have acquired it as a second language.
There is no sense in the mystical version of the substratum theory,
which attributes changes, say, in modern Germanic languages,
to a “Celtic substratum” — that is, to the fact that many centuries
ago, some adult Celtic-speakers acquired Germanic speech.
Moreover, the Celtic speech which preceded Germanic in southern
Germany, the Netherlands, and England, was itself an invading
language: the theory directs us back into time, from “race” to
“race,” to account for vague “tendencies” that manifest themselves
in the actual historical occurrence of sound-change.

Aside from their failure to establish correlations, theories of
this kind are confuted by the fact that when sound-change has
removed some phonetic feature, later sound-change may result
in the