Joseph H. Greenberg-Universals of Language - Revised 2nd Edition-The MIT Press (1966)

,
UNIVERSALS
of LANGUAGE
SECOND EDITION
EDITED BY JOSEPH H. GREENBERG

PROFESSOR OF ANTHROPOLOGY
STANFORD UNIVERSITY
REPORT OF A CONFERENCE HELD AT

DOBBS FERRY, NEW YORK
APRIL 13-15, 1961
.1111111
THE M.I.T. PRESS

MASSACHUSETIS INSTITUTE OF TECHNOLOGY
CAMBRIDGE, MASSACHUSETIS, AND LONDON, ENGLAND
PREFACE
r--~~~::"-~:~ ......
::-;,:~,\~Wt'RSIVAl! C11;;'l<::::-~
/f.I...\} .
!/~'
II,
r{"-
\\~
.
cI
Jt ".,.
~.-'-
'"
8 3 3 5' 6
"
/-f I It r ,.
,!
~r" ~
t TF.
~.\\
".~_\,
"-1'
~-.. ,IJ
r$.."1:."
,I
Copyright
1963.
1966
by
. The Massachusetts Institute of Technology

Second Edition
All Rights Reserved. This book may not be

reproduced, in whole or in part, in any form
(except by reviewers for the public press),
without written permission from the publishers.
Second printing, first paperbaCk edition, March 1966

Third printing. August 1968
Fourth printing, January 1973
Fifth printing, March 1976
ISBN 0 262 57008 4 (paper)
ISBN 0 262 07020 0 (hard)
Library of Congress Catalog Card Number: 62-22020
Printed in the United States of America
The Conference on Language Universals was held at Gould

House, Dobbs Ferry, New York, April 13-15, 1961, under the
sponsorship of the Linguistics and Psychology Committee of the
Social Science Research Council with a grant from the National
Science Foundation. Although the topic of universals of language
was one of the first to receive interdisciplinary interest from
linguists and psychologists in the course of collaboration under
the aegis of the S.S.R.C. Committee, the immediate stimulus
for the Conference came during the academic year 1958-1959
from Joseph B. Casagrande, at that time a staff member of the
Council. He suggested that the three members of the Committee
who were resident Fellows that year at the Ford Center for Ad
vanced Studies at Stanford, California - Joseph H. Greenberg,
James J. Jenkins, and Charles E. Osgood - prepare a memoran
dum on the subject of universals in language which might serve
as a basis for theoretical investigation in this area, and for the
planning of a Conference. This document, "Memorandum Con
cerning Language Universals," was subsequently distributed in
slightly'revised form to those invited to the Conference and was
itself one of the subjects of discussion at the meeting. It is printed
on page xv of this book.
The original plan for papers for the Conference was based on
a cross-cutting division of the field of universals into those of
phonology, grammar, and semantics on the one hand, and syn
chronic and diachronic on the other. While it was realized that
such a scheme might cause neglect of topics which did not readily
fit into such a pattern of organization, it was felt that in view of
the exploratory nature of the meeting no great harm would re
v
vi
II
I
PREFACE
suIt. Moreover, no viable alternative had presented itself. As

it turned out. various modifications in individual instances largely
obscured this ground plan, and this was perhaps just as well.
In addition to these invited papers, prepared and circulated in
advance. final oral summaries from the viewpoints of linguistics,
cultural anthropology, and psychology were presented by Roman
Jakobson, Joseph B. Casagrande, and Charles E. Osgood, re
spectively. These summaries, as subsequently edited by the au
thors, appear as the three final chapters of this book.
The results of the Conference as reported in this book do not
contain either verbatim or edited reports of the highly stimulat
ing and productive discussions which took place. However, the
recordings of these discussions and subsequent written comments
and criticisms were circulated and have been incorporated in
essential ways.
In addition to individual bibliographies, a general bibliography
of language universals has been compiled as the final section of
this work. This bibliography is obviously not exhaustive. It

necessarily overlaps to some extent with bibliographies of the
individual chapters but contains many items not cited elsewhere
in the present work. I have sought to include specific proposals
regarding universals, discussions of the nature of universals,
typological analyses from which universals might be derived
through empirical investigation, and, in a few cases, analytical
discussions which seem suggestive from this point of view. I

am aware of the looseness of these criteria. The present brief
compilation is intended merely as a general orientation for the non
professional to the problems discussed in the book. I am grateful
to Uriel Weinreich for advice and assistance in its preparation.

I wish to thank all those who participated so enthusiastically
and effectively in the work of the Conference. A complete list
of participants is included on page ix of this book. In particular,
I wish to thank Dr. Francis H. Palmer of the S.S.R.C. staff,
to whom fell the task of final preparation of the meeting and the
further responsibilities arising from it including aid and advice
in the editing of this volume.
JOSEPH
H.
GRBENBERG
CONTENTS
List of Participants
ix
Introduction
xi
Memorandum Concerning Language Universals

Joseph H. Greenberg, Charles E. Osgood, and
James J. Jenkins
xv
1.
2.
3.
4.
S.
The Problem of Universals in Language

Charles F. Hockett
Are There Universals of Linguistic Change?

Henry M. Hoenigswald
30
Assumptions about Nasals: A Sample Study in

Phonological Universals
Charles A. Ferguson
53
Phoneme Distribution and Language Universals

Sol Saporta
61
Some Universals of Grammar with Particular

Reference to the Order of Meaningful Elements
Joseph H. Greenberg
73
vii
viii
CONTENTS
6. A Search for Universals in Indo-European Diachronic

Morphology
Warren Cowgill
IJ4
LIST OF PARTICIPANTS
7. On the Semantic Structure of Language

Uriel Weinreich
142
8. Semantic Universals
Stephen Ullmann
217
9. Implications of Language Universals for Linguistics

Roman Jakobson
263
The participants in the Conference, in addition to Dr. Francis H.
Palmer as S.S.R.C. staff member, were the following:
10. Language Universals in Anthropological Perspective

Joseph B. Casagrande
279
11. Language Universals and Psycholinguistics

Charles E. Osgood
299
Bibliography
323
Index
329
Dr.loseph H. Greenberg
Department of Anthropology
Stanford University
Dr.lohn B. Carroll
Graduate School of Education
Harvard University
Dr. Joseph B. Casagrande
University of Illinois
Dr. Einar Haugen

Department of Scandinavian
Languages
University of Wisconsin
Dr. Harold C. Conklin

Yale University
"'Dr. Charles F. Hockett

Department of Linguistics
CornelI University
Dr. FrankUn S. Cooper

Haskins Laboratories
New York 17, New York
Dr. Henry M. Hoeoigswald

University of Pennsylvania
Dr. Warren C. Cowgill

Linguistics Department
Yale University
Dr. Charles A. Ferguson
Center for Applied Linguistics
Washington 6, D.C.
Dr. Fred W. Householder,lr.

rndiana University
Dr. William I. Gedney

Department of English
Language and Literature
University of Michigan
.Dr. Dell H. Hymes

University of California
"'Professor Charles F. Hockett of Cornell University could not attend the

meeting, but in addition to submitting a paper, he cooperated fully in the
activities grOwing out of the Conference.
ix
x
Dr. Roman lakobsoD
Department of Slavic Languages
Harvard University and
Massachusetts Institute of
Technology
Dr. lames 1.leoJdos
Department of Psychology
University of Minnesota
Dr. Sydney M. Lamb
Computer Center
University of california
Dr. Wallace E. Lambert

Department of Psychology
McGill University
Dr. Eric H. Leooeberg
Speech Research Laboratory
The Children's Hospital Medical
Center
Boston 15, Massachusetts
Dr. Leigh lJsker
LIST OF PARTICIPANTS
Dr. George P. Murdock
University of Pittsburgh
Dr. Cbarles E. Osgood
Institute of Communicatioos
Research
INTRODUCTION
Dr. Herbert H. Paper

Department of Near Eastern
Languages
University of Michigan
Dr. Sol Saporta
Department of Spanish and
Portuguese
University of Washington
Dr. Tbomas A. Sebeok
Center for Anthropology,
Folklore and Linguistics
Indiana University
Dr. StepheD UIlmaDD

Department of French
University of Leeds
Dr. Charles F. VoegeUn
Ind~na University
Dr.lobn Lob:
Department of U ralic and Altaic
Languages
Columbia University
Dr. Uriel Welnrelcb

Columbia University
Dr. Floyd G. Louosbury

Yale University
Dr. Rulon S. Wells

Department of Philosophy
Yal~ University
Since a number of the papers in the present volume, including

the three final statements and the Memorandum, are largely con
cerned with the basic theoretic and methodological issues in
volved in the investigation of language universals, a brief state
ment here seems sufficient to orient the reader to the papers.
These remarks are intended to summarize some of the salient
points which emerged from the papers and the discussion. It
should be understood that while all, or almost all, of the partici
pants in the Conference might concur in what is said here, this
introduction represents the personal reactions of the editor to
issues discussed at the meeting.
In view of the present level of methodological sophistication
of both synchronic and diachronic linguistics and the truly enor
mous mass of empirical data on languages of the world now at
our disposal, the time appears ripe for generalizing efforts on
a wide scale. Indeed, this is imperative for linguistics both to
fulfill its own promise as a science and to make the contributions
to the formulation of a general science of human behavior which
its sister disciplines may legitimately expect.
Such attempts should not be identified with earlier approaches
based on categories formulated a priori from supposed necessary
categories of thought derived from normative logic. One of the
recurrent themes of the meeting was, indeed, reference to Bloom
field's weB-known dictum in his classic work Language that
"the only useful generalizations about language are inductive genXl
'
r,
xii
INTRODUCTION
eralizations." However, it seemed also to be generally agreed

that the method of science is both inductive and deductive. The
formulation of generalizations attained by inductive examination
leads to higher-level hypotheses from which, in turn, further gen
eralizations may be deduced. The latter must then be put to the
test of empirical validation.
Such principles derived from generalizations concerning lin
guistic change and linguistic structure reflect important and
fundamental aspects of human behavior. They cannot be fully
understood without to some extent abandoning the traditional
self-sufficiency of linguistics in favor of fundamental collabora
tion with psychology and the social sciences. Such a statement
should not be construed as a criticism of the great and continu
ing value of the standard synchronic and diachronic procedures.
Indeed, as will be evident in much of what follows, they form
the indispensable bases for arriving at generalizations about
language.
There was general agreement that it was necessary and com
pletely legitimate to include as universals in addition to state
ments of the simple type "all languages have a given feature x,"
likewise implicational relations, universal frequency distribu
tions, statistically better than chance correlations, and other
logic types as set forth in greater detail in the Memorandum.
From a purely logical point of view this might be summarized as
follows. All statements of the form (x) x E L ) ... , that is,
"for all x, if x is a language, then ... ," are permitted.
Finally, two matters of particular concern to linguists may be
mentioned: the question of typology and the question of universals
in relation to the two major divisions of screntific linguistics
(synchronic and diachronic studies). The usefulness of typologies
in the present connection may be illustrated from my own paper
on which a typology based On the order of elements in certain
major constructions appears as a virtually indispensable tool in
the search for cross-linguistic regularities in this aspect of lan
guage. It is perhaps not overstating the case to say that one of
the values of this Conference was the realization that typological
classification finds its sought-for justification in the investigation
xiii
INTRODUCTION
of universals. This also means that the proposed research on

universals finds a very real and useful foundation in earlier
linguistic work on typologies; for example, in the grammatical
typologies of the nineteenth-century pioneers, in those of Sapir
and Bally, and in the phonological endeavors of Trubetskoy,
Jakobson, Hockett, Menzerath, and Voegelin, to mention but a
few.
Another motif of this Conference was the interrelationship of
synchronic and diachronic approaches both of which are seen to
complement each other in that neither can be fully understood
without the other. There is perhaps a lesson here for cultural
anthropology where there has been a tendency to identify the
search for laws with an ahistorical functionalist orientation and
to oppose to it a historical particularistic approach. In fact,
generalizations may apply equally to diachronic processes and
synchronic states. The possibility of universals of change, more
over, would seem to be of particular relevance to psycholinguistics
since change of habits over time is the very stuff of learning, which
is such a central concern of contemporary psychology.
Perhaps enough has been said to indicate some of the more
significant aspects of the study of language universals exemplified
in the papers and discussions of the Conference. It is hoped that
the report of the Conference will serve to amplify and elucidate
these brief introductory remarks.
JOSEPH
New York, New York

October 1962
H. GREENBERG
MEMORANDUM CONCERNING LANGUAGE
UNIVERSALS
presented to the Conference on Language Universals,
Gould House, Dobbs Ferry, N. Y., April 13-15, 1961
1. Introduction
I!:
r;
Ii
Underlying the endless and fascinating idiosyncrasies of the

world's languages there are uniformities of universal scope.
Amid infinite diversity, all languages are, as it were, cut from
the same pattern. Some interlinguistic similarities and identities
have been formalized, others not, but working linguists are in
many cases aware of them in some sense and use them as guides in
their analyses of new languages. This is an important but limited
and incomplete use of these consistencies. Language universals
are by their very nature summary statements about characteristics
or tendencies shared by all human speakers. As such they consti
tute the most general laws of a science of linguistics (as contrasted
with a method and a set of specific descriptive results). Further,
since language is at once both an aspect of individual behavior
and an aspect of human culture, its universals provide both the
major point of contact with underlying psychological principles
(psycholinguistics) and the major source of implications for
human culture in general (ethnolinguistics).
It is our belief that coordinated efforts beyond the scope of
individual researchers will be necessary to establish on firm
grounds the actual facts concerning universals in language. Thus,
the illustrations cited later in this Memorandum must be taken
cum grano salis as based on the specific knowledge of the writers
xv
xvi
MEMORANDUM
which, however wide it might be, could not in the nature of things
be exhaustive. Organization of some central source of data, some
thing like a cross-cultural file for a large and representative
sample of world languages would vastly facilitate the establish
ment of well-grounded universals and their continued study by
scholars. As a first step, it is proposed that the Committee on
Linguistics and Psychology of the Social Science Research Council
arrange for a Work Conference on Language Universals. This
Memorandum, which has grown out of discussions held at the
Center for Advanced Study in the Behavioral Sciences during
1958-1959, is offered to stimulate activity leading to such a
conference and to suggest the kinds of topics which might appro
priately be discussed.
2. Examples of Universals
Before going further, it is perhaps wise to describe a few ex
amples of language universals which will illustrate some of the
scope and diversity involved in the types of similarities seen
between language systems.
First, we may take an example from phonology. The phonemes,
or individual sound units, may be looked upon as consisting of
the simultaneous occurrence of several elements called features.
For example, in English the phoneme Ibl is characterized by
voicing, stop articulation (that is, it involves a complete closure
as contrasted with various types of fricatives), and it is oral, that
is, nonnasaI. There is another phoneme Ipi in English which shares
all of these characteristics except voicing. In general, the features
of a particular phoneme are not unique, and the entire set consists,
of varying combinations of the same small inventory of features.
More often than not, there is a parallelism or symmetry in the
combinations observed. This leads to certain expectations on the
part of the investigator. For example, in the investigation of a
hitherto unstudied language in Nigeria, a phonemic contrast was
found between the two velar stop consonants /k/ and Ik'I, the former
unglottalized and the latter glottalized, as well as a pair of dentals
ItI and 11'/. Since the third unvoiced stop consonant Ipl was also
MEMORANDUM
xvii
found, the linguist at this point formed the hypothesis that a

glottalized counterpart Ip'I was also likely to occur even though
it had not yet appeared in a fairly considerable body of linguistic
material. Ultimately it was found to occur in a very small number
of words. This expectation might, of course, have been disap
pointed, but investigators do form such hypotheses and find that
the alertness engendered pays off in a majority of cases.
The tendency toward symmetry in the sound system of lan
guages described here has, ofcourse, psycholinguistic implications.
The articulatory habits of speakers involved in the production of
the phonemes consist of varied combinations of certain basic
habits, those employed in the production of the features. This
appears, for example, in language acquisition by the child. At
the point in the development of the English-speaking child that
he acq uires the distinction between band p based on voicing
versus nonvoicing, he simultaneously makes the distinctions
between d and t, g and k, and other similar pairs. In other words,
he has acquired the feature, voicing versus nonvoicing, as a unit
habit of motor differentiation. Such facts have an obvious im
portance for learning theory in psychology.
A quite different sort of universal may also be illustrated within
the domain of phonology. As stated earlier, distinctive features
are combined to generate the phonemes employed in any given
language. It is of some linguistic interest and great psycho lin
guistic interest to examine the relation between the number of
distinctive features required to generate the number of different
phonemes employed by the language and the number of distinctive
features a~tual1y in use. A maximally efficient code, in the infor
mation theory sense, would employ just the number of features
necessary to distinguish its phonemes; for example, the 32 pho
nemes of English would require only five distinctive binary
features (that is, the features could be combined in two to the
fifth power different combinations, or 32 combinations). However,
in English nine binary features are actually employed. The effi
ciency of English in respect to phonology is therefore about five
ninths, or 56%. Investigation of several languages suggests the
generalization that the phonetic efficiency of languages is distrib
xviii
MEMORANDUM
uted roughly around the 50% point. A study of one language

(Spanish) as it has changed over time reinforced this generalization
by revealing that the efficiency of that language oscillates around
the 50% value over time.
It appears that there are sets of pressures bearing on any pho
netic system which cause it to maintain some optimal efficiency
value. If the language becomes too inefficient, that is, has too
many features overdetermining the phonemes, it becomes pos
sible to neglect some of them and still be understood. We presume
that such lapses become more frequent and the sound system
begins to change toward simplicity. On the .other hand, if the
system is too efficient, mishearing and misperceptions should
become frequent, and we assume that the speakers are led (or
driven) to make additional distinctions to maintain clarity. It is
obvious that this "explanation" generates a complex statistical
function, but one that presumably reftects universal processes in the
total dynamics of communication between speakers and listeners.
3. The Nature of Universals
The examples just cited illustrate that the term "universal" is

used here in a somewhat extended sense. We have not limited
ourselves to statements of the type that all languages have vowels;
all languages have phonemes; allianguag~ sound systems may be
resolved into distinctive features, etc. We feel that it is important
to include generalizations which tend to hold true in more than
a chance number of comparisons (such as symmetry of sound
systems) or which state tendencies to approach statistical limits
across languages or in one language over time. We are convinced
that the wider use of this concept will prove to be most fruitful
from the psycholinguistic viewpoint. All phenomena which occur
with significantly more than chance frequency in languages in
general are of potential psychological interest.
With this expanded view of universals, confusion may be most
easily avoided by pointing out that types of universals may be
differentiated both with respect to logical structure and with
respect to substantive content.
MEMORANDUM
XIX
4. Logical Structure of Universals

From a stictly logical point of view, it is possible to define
universals as any statements about language which include all
languages in their scope, technically all statements of the form
at 'IS, "F or a II x, 1'f x IS
en ....
"( x ) x E L ::::l , "th
. a Ianguage, th
"
These statements fall into various logical subtypes. Such an
analysis is useful since in addition to specifying clearly what is
to be considered a universal, the distinct subtypes do to some
extent present distinguishably different problems from other
points of view. We have considered and will present here six types
of universals. The first three may be considered as universals
which concern existence (that is, "X does or does not exist") and
the last three as universals which concern probabilities (that is,
"X (or some value of X) is more probable than Y (or some other
value of X)").
4.1. Unrestricted universals
These are characteristics possessed by all languages which are

not merely definitional; that is, they are such that if a symbolic
system did not possess them, we would still call it a language.
Under this heading would be included not only such obvious
universals as, for example, that all languages have vowels, but also
those involving numerical limits, for example, that for all
languages the number of phonemes is not fewer than 10 or more
than 70, or that every language has at least two vowels. Also
included are universally valid statements about the relative text
or lexicon frequency of linguistic elements.
4.2. Universal implications
These always involve the relationship between two characteris

tics. It is asserted universally that if a language has a certain
characteristic, (cp), it also has some other particular characteristic
but not vice versa. That is, the presence of the second (1jJ) docs
not imply the presence of the first (cp). For example, if a language.
has a category of dual, it also has a category of plural but not
necessarily vice versa. Hereafter we express such relationships
between predicates by an arrow, for example, dual -- plural.
xx
MEMORANDUM
Such implications are fairly numerous, particularly in the phono

logic aspect of languages.
4.3. Restricted equivalence
This is the case of mutual implication between characteristics
which are not universaL That is, if any language has a particular
nonuniversal characteristic, rp, it also has 4 and vice versa. For
example, if a language has a lateral click, it always has a dental
click and vice versa. In this example, unfortunately, all the
languages are from a restricted area in South Africa, and the
equivalence is really a single case. Equivalences of more fre
quently appearing logically independent characteristics are diffi
cult to find. They would be of great interest as indicating impor
tant necessary connections between empirically diverse properties
of language.
4.4. Statistical universals
These are defined as follows: For any language a certain char
acteristic (rp) has a greater probability than some other (frequently
its own negative). This includes "near universals" in extreme cases.
Only Quileute and a few neighboring Salishan languages among
all the languages of the world lack nasal consonants. Hence we
may say that, universally, the probability of a language having
at least one nasal consonant (rp) is greater (in this instance far
greater) than that it will lack nasal consonants (not rp). We may
extend this type to include cases of more than one alternative.
For example, of the three devices of suffixing, prefixing, and
infixing, the probabilities are not random and in fact are here
stated in decreasing order. In this case the alternatives are
not mutually exclusive, for example, a language can have both
prefixes and suffixes.
4.5. Statistical correlations
This differs from the preceding in a manner parallel to that in
which universal implications differ from unrestricted universals.
In this instance also we are interested in the relation of several
characteristics. By a statistical correlation we mean, then, that
universally, if a language has a particular characteristic (rp) it has
MEMORANDUM
xxi
a significantly greater probability of possessing some other charac

teristic (4) than if it does not possess (rp).
The following is a probable example. Languages with gender
distinctions in the second person singular are rarer than in the
third person. Usually a language with gender distinction in the
second person singular also has this distinction in the third person
singular but not vice versa. If this were without exception, we
would have the implication: Second person singular pronominal
gender -. third person singular pronominal gender. There are
apparently, however, a few languages in central Nigeria which
have the distinction in the second person, but not in the third.
The proviso here is that these languages have not been well
studied. If the exceptions are genuine, then we have the following
statistical correlation: If a language has pronominal gender in
the second person singular, it has a greater probability (much
greater in this case) of having this distinction in the third person
singular than of not having it.
4.6. Universal frequency distributions
Finally we have instances where a certain measurement, for

example, redundancy in information theory, as mentioned earlier,
may be applied to any language. When this is so, it is possible that
the results of each measurement over an adequate sample of
languages will show a characteristic mean and standard deviation.
Means, standard deviations, or other statistical measures derived
from such distibutions may be considered as universal facts about
languages.
5. Substantive Classes of Universals
A second basis of classification which obviously crosscuts the

division by logical type is that which operates with the aspect of
language involved. While a variety of alternative categories is
possible~ in general, this principle of division will give us four
types: phonological, grammatical, semantic, and symbolic. In this
classification, the first three involve either form without meaning
or meaning without form, whereas the last, which is concerned
P
~f'
xxii
MEMORANDUM
with sound symbolism, involves the connection between the two.

For example, the near universality of nasals is a phonologic
universal in whose statement we are not concerned with the
meanings of the linguistic forms in which the nasals do or do not
figure. The grammatical statement that suffixing is more frequent
than infixing is not concerned, on the other hand, with the
particular sounds utilized in suffixing. Again, the semantic univer
sal that all languages have some metaphorically transferred
meanings is not concerned with the particular sounds of the forms
in which they occur. On the other hand, a statistical symbolic
universal such as "there is a high probabi)ity that a word desig
nating the female parent will have a nasal consonant" involves
both sound and meaning.
6, Domain of the Universals
All the examples thus far cited in this M,emorandum have been
synchronic; that is, the statements refer to universally discover
able regularities arrived at by observing the characteristics of
language states rather than of language changes. The definition
of universals, moreover, and the further classifications of their
occurrence into phonologic, grammatical, semantic, and symbolic
have all been framed with a view to synchronic universals.
However, we feel it is essential to extend the consideration of
universals to diachronic facts of language. From the present
point of view, it would be unwise to exclude these from considera
tion, in spite of the important differences to be noted, since
universals of change have important psycholinguistic implications.
From the general linguistic point of view. some universals are
most easily understood as the outcome of dynamic processes,
for example, semantic metaphor as the result 0/ metaphorical
semantic change, or again the universal, or almost universal,
existence of variant forms of meaningful units (that is, morpho
phonemic alterations) as the result of the diachronic process of
regular conditioned sound change. From the psychological point
of view, such universals may serve to focus attention on phenom
ena, which may be brought under exper~mental control in the
MEMORANDUM
xxiii
laboratory for study (e.g., the historical instability of liquids and

nasals suggests both articulatory and auditory studies of interest
in motor skills and perception).
Diachronic universals do differ in several fundamental ways in
regard to bases of classification mentioned earlier. To begin with,
although there are important universal hypotheses concerning
change such as "aU languages change" or "the rate of replacement
of fundamental vocabulary is constant over time," the particular
substantive diachronic universals are probabilistic. We can never
say with certainty that a particular class of changes will always
occur. The varied development for distinct but related languages
from the same basis is enough to show this. Further, the logical
form for universals presented earlier requires significant modi
fication. Whereas for synchronic universals we always start with
"For all x if x is a language (Le., a single synchronic state),
then ... ," in the case of diachronic rules the reference to two
synchronic states is essentially with the further proviso that one
be the historical continuation of the other. It is common usage
to say that these are the same language unless the chronological
distance is great, that is, Latin and French. Logically, then,
diachronic universals are of the form "For all (x) and all (y)
where (x) is an earlier and (y) a later stage of the same language ... ".
Further, for diachronic change, the division into phonologic,
semantic, and grammatical processes holds, but symbolism is not
a type of change, although changes can result in forms which are
more or less similar to universal sound symbolic norms.
Synchronic and diachronic regularities are obviously interre
lated. The most general statement of this interrelationship is in
the form of limitations, namely, that no synchronic state can exist
which is not the outcome of possible diachronic processes
(except perhaps de novo for artificial and pidgin languages) and
no diachronic process can be posited which could lead to a
synchronic state which violates a universally valid synchronic
norm. It is important to note that, just as was indicated earlier
that some synchronic universals are most easily understood as
the outcomes of certain widespread processt?s, so specific dia
chronic changes cannot be understood without reference to the
-r .
r
xxiv
MEMORANDUM
network of synchronic relations within the language at the time

of the change. This is the basic contribution of structural lin
guistics to the study of linguistic change. Diachronic universals
are probabilistic precisely because simultaneously with the
universal tendencies toward changes of one kind as against other
possibilities there are significant variables in the language structure
itself, and every language structure is unique in some way.
An example of a diachronic process with important psycho
logical implications is the tendency found in the most diverse
languages for unvoiced consonants between vowels to become
voiced. The psychologist has a background of experimental data
dealing with the processes of anticipation (performing an act or
portion of an act before it is wholly appropriate) and persevera
tion (continuing a behavioral element beyond the time it is wholly
appropriate). He expects adjacent phonemes to influence one
another
the commonly observed phenomenon of conditioned
allophonic variation. Given a sequence of vowel, consonant, vowel
he must predict on the grounds of both anticipation and perseve
ration that there will be a strong tendency for the consonant to be
voiced rather than unvoiced since both the preceding and follow
ing elements are voiced. The psychologist would select the vowel
unvoiced consonant-vowel sequence as a "weak" spot in the
language and one where change is more likely than either conso
nant-vowel or vowel-consonant alone. This prediction, of course,
has two aspects: first, that diachronically unvoiced consonants
between vowels will tend to become voiced and, second (all other
things being equal), in a language at any given time there will tend
to be more vowel-voiced consonant-vowel combinations than
vowel-unvoiced consonant-vowel combinations. The verification
of these findings also suggests to the psycholinguist methods for
working with the phenomena of anticipation and perseveration of
sound pattern in the laboratory setting.
7. Interrelations of Language Universals
In addition to its importance for the interdisciplinary field of

psycholinguistics and psychology proper, this study of language
MEMORANDUM
xxv
universals is intimately connected with the establishment of

scientific laws in the linguistic aspects of human behavior. It is
thus of general significance for the development of the behavioral
sciences. The study of universals leads to a whole series of empir
ical generalizations about language behavior, some as yet tentative,
some well established. These are the potential material for a
deductive structure or scientific laws. Some, indeed, probably
most, of these have the status of empirical generalizations which
cannot at our present state of knowledge be related to other
generalizations or deduced from laws of more general import.
For example, it seems well established that every language has
syllables of the form ev (consonant followed by vowel) in
addition to whatever other type it may possess. We cannot say
why this should be so, on the basis of general laws of wider scope.
For this reason it has a certain fragility. We would be quite
astonished if someone discovered a languagewhkh did not have
this kind of syllable, but we cannot give any reason why this
should not be found.
It is clear, however, that some universals having to do with
the same aspect of language are interconnected. For example,
we have chains of implications in this very area of syllabic
structure.
Thus, eeev - ccv - ev, where V may in any case be
preceded by sequences of e, and veee - vee - ve - V,
where V may be followed by sequences of e. In this instance we
can deduce all of these from the general statement that if syllables
containing sequences of n consonants in a language are to be found
as syllabic types, then seq uences of n I consonants are also to be
found in the corresponding position (prevocalic or postvocalic)
except that ev - V does not hold. The possibility of deducing
these five universal implications (and it probably holds for still
larger consonantal sequences) gives a degree of certitude to the
individual statements that they would not otherwise possess.
General statements of this kind may be called internal since
they contain predicates of the same kind as the individual uni
versals that they explain. In other cases, we have external deduc
tions, as in some of the examples discussed earlier, where psycho
f
):
xxvi
MEMORANDUM
logical principles are adduced which do not specifically involve

linguistic predicates and which serve as explanatory principles for
a much wider variety of phenomena, for example, the behavior of
rats in mazes. These wider principles need not always be psycho
logical in the narrower sense. For example, they may be cultural
with a social-psychological aspect as when we consider the
prestige and power relations of two linguistic communities as a
variable in accounting for tendencies of universal scope involving
the effects of one language on another.
8. Present Needs
The importance of the study of languange universals to both

the burgeoning field of psycholinguistics and the development of
linguistics as a behavioral science has, we believe, been sufficiently
indicated. It has been further suggested that important consequences
for several others of the behavioral sciences may be involved. It
remains to be considered whether coordinated efforts outside the
scope of the individual researchers can be useful for the develop
ment of this area of study. The first step methodologically is
obviously to establish on firm grounds the actual facts concerning
the universals of language. For some of the more elaborate
hypotheses concerning, for example, semantic universals, it is
clear that there is no substitute for special individual research
projects aimed at particular problems and involving fieldwork
(so, for example, the Southwest Project in Comparative Psycho
linguistics). For many types of universals, however, particularly
synchronic phonologic and grammatical universals, the organiza
tion of something of the order of cross-cultural files for a large
sample of languages would vastly facilitate the establishment of
factually well-grounded universals concerning language. The
area of sound symbolism might be selectively indexed since an
exhaustive body of data would obviously include all the mor
phemes of all the languages of the world.
Such a project would obviously require careful planning. The
categories to be selected, the manner of selecting, recording, and
indexing the data, the question as to how the results could be
xxvii
MEMORANDUM
made available generally to interested scholars, problems of

organization and financial support would all have to be con
sidered. It is, therefore, suggested that a work conference on the
subject of language universals be organized to include linguists,
psychologists, and anthropologists interested in this area under
the sponsorship of the Council in order to consider both the
theoretical problems of universals and the possible organization
of such a project as that mentioned earlier. In addition to the
specific problems of such a project, such a meeting might well
stimulate individual scholars in carrying on their research in this
area.
JOSEPH H. GREENBERG
CHARLES E. OSGOOD
JAMES J. JENKINS
CHAPTER 1
THE PROBLEM OF UNIVERSALS IN LANGUAGE

CHARLES F. HOCKEIT
Cornell University
1. Introduction
A language universal is a feature or property shared by all
languages, or by all language. The assertion of a (putative)
language universal is a generalization about language.
"The only useful generalizations about language are inductive
generalizations" (Bloomfield, 1933, p. 20). This admonition is
clearly important, in the sense that we do not want to invent lan
guage universals, but to discover them. How to discover them is
not so obvious. It would be fair to claim that the search is coter
minous with the whole enterprise of linguistics in at least two
ways. The first way in which this claim is true is heuristic: we can
never be sure, in any sort of linguistic study, that it will not reveal
something of importance for the search. The second way in which
the claim is plausible, if not automatically true, appears when we
entertain one of the various possible definitions of linguistics as a
branch of science: that branch devoted to the discovery of the
place of human language in the universe. This definition leaves the
field vague to the extent that the problem of linguistics remains
unsolved. Only if, as is highly improbable, the problem were com
pletely answered should we know exactly what linguistics is-and
In preparing this article (originally written early in 1961) for the second
printing of the volume, I have confined myself to the correction of misprints
and of a few infelicities of expression. I would now (1965) state many of the
points differently, and withdraw several of them altogether.
~.
t':'
Charles F. Hockett
at that same millennia I moment there would cease to be any

justification for the field. It is hard to discern any clear difference
between "the search for language universals" and "the discovery
of the place of human language in the universe." They seem
rather to be, respectively, a new-fangled and old-fashioned way
of describing the same thing.
But, however described, the problem is important; and it is
fitting that from time to time we set aside our sundry narrower
professional concerns and take stock. What are we really sure of
for all languages? What are the outstanding gaps? Can we point
to specific investigations of probably crucial importance? What
are the most important differences of expert opinion, and how are
they to be resolved?
In the present paper the writer will touch on five matters. The
balance of the Introduction sets forth a number of assumptions,
warnings, and pitfalls; these might be regarded as an expansion,
perhaps even as a clarification, of Bloomfield's terse remark
quoted earlier. Section 2 summarizes some features found (if the
writer is right) in all human languages but lacking in one or
another system of nonhuman animal communication. Section 3
proposes a set of features as criterial for language; that is, if a
communicative system has all the features of the set, it is proposed
that we call it a language. Sections 4 and 5 list a very few proper
ties, respectively phonological and grammatical, that seem to be
shared by all human languages but that are not obviously neces
sitated by the presence of the features of the criterial set.
1.1. The assertion of a language universal must be founded on
extrapolation as well as on empirical evidence.
Of course this is true in the trivial sense that we do not want

to delay generalizing until we have full information on all the
languages of the world. We should rather formulate generaliza
tions as hypotheses, to be tested as new empirical informa
tion becomes available. But there is a deeper implication. If
we had full information on all languages now spoken, there
would remain languages recently extinct on which the information
was inadequate. There is no point in imagining that we have
adequate information also on these extinct languages, because

that would be imagining the impossible. The universe seems to be
so constructed that complete factual information is unattainable,
at least in the sense that there are past events that have left only
incomplete records. Surely we seek constantly to widen the empir
ical base for our generalizations; equally surely, we always want
our generalizations to subsume some of the unobserved, and even
some of the unobservable, along with all of the observed.
1.2. The assertion oflanguage universals is a matter ofdefinition
as well as of empirical evidence and of extrapolation.
If the next "language" on which information becomes available
were to lack some feature we have believed universal, we could
deny that it was a language and thus save the generalization (cf.
Kemeny, 1959, pp. 97-98). Triviality from this source can be
avoided by various procedures, but they all involve making
decisions in advance-and such decisions are definitions. We can
decide that any system manifesting a certain explicitly listed set of
features (the defining set) is to be called a language. The univer
sality of the particular features we have chosen is then tautologous.
Of course, the list itself can be revised, for each successive round
of the search for universals.
1.3. A feature can be widespread or even universal without being
important.
This is most easily shown by a trick. Suppose that all the lan
guages of the world except English were to become extinct. There~
after, any assertion true of English would also assert a (synchronic)
language universal. Since languageS no longer spoken may have
lacked features we believe universal or widespread among those
now spoken, mere frequency can hardly be a measure of
importance.
1.4. The distinction between the universal and the merely wide
spread is not necessarily relevant.
The reasoning is as for 1.3. Probably we all feel that the univer
sality of certain features might be characterized as "accidental"
Charles F. Hockett
-they might just as well have turned out to be merely widespread.

This does not tell us how to distinguish between the "accidentally"
and the "essentially" universal. On the other hand, that which is
empirically known to be merely widespread is thereby disqual
ified as an "essential" universal-though careful study may show
that it is symptomatic of one.
1.5. The search for universals cannot be usefully separated from
the search for a meaningful taxonomy of languages.
(Here "taxonomy" refers to what might also be called "typol
ogy," not to genetic classification.) Suppose that some feature,
believed to be important and universal, turns out to be lacking
in a newly discovered language. The feature may still be im
portant. To the extent that it is, its absence in the new language
is a typological fact of importance about the language.
Conversely, if some feature is indeed universal, then it is
taxonomically irrelevant.
Here is an example that illustrates both 1.4 and 1.5. It was at
one time assumed that all languages distinguish between nouns
and'verbs-by some suitable and sufficiently formal definition of
those terms. One form of this assumption is that all languages
have two distinct types of stems (in addition, possibly, to various
other types), which by virtue of their behavior in inflection (if
any) and in syntax can appropriately be labeled nouns and verbs.
In this form, the generalization is rendered invalid by Nootka,
where all inflectable stems have the same set of inflectional
possibilities. The distinction between noun and verb at the level
of stems is sufficiently widespread that its absence in Nootka is
certainly worthy of typological note (1.5). But it turns out that
even in Nootka something very much like the noun-verb contrast
appears at the level of whole inflected words. Therefore, although
Nootka forces the abandonment of the generalization in one form,
it may still be that a modified form can be retained (1.4).
The Port Royal Grammar constituted both a putative descrip
tion of language universals and the basis of a taxonomy. The
underlying assumption was that every language must provide, by
one means or another, for all points in the grammatico-logical
scheme described in the Grammar. Latin, of course, stood at the

origin in this particular coordinate system. Any other language
could be characterized typologically by listing the ways in which
its machinery for satisfying the universal scheme deviated from
that of Latin. This classical view in general grammar and in
taxonomy has been set aside not because it is false in some logical
sense but because it has proved clumsy for many languages: it
tends to conceal differences that we have come to believe are
important, and to reveal some that we now think are trivial.
1.6. Widespread (or universal) features are most apt to be
important if they recur against a background of diversity.
1.7. Widespread (or universal) features are the more apt to be
important the less readily they diffuse from one language to
another.
Given a taxonomy, if we find that languages of the most
diverse types nonetheless manifest some feature in common, that
feature may be important. It is not apt to be, however, if it is an
easily diffusible item. Thus the fact that many languages all over
the world have phonetically similar words for 'mama' is more
significant than a similarly widespread general phonetic shape
for 'tea'. (On the former, see now Jakobson, 1961.)
In allowing for diffusion, we must also take into consideration
that even features that do not diffuse readily may spread from
one language to others when the speakers of the languages go
through a long period of intimate contact. This fact, if no other,
would seem to render suspect any generalizations based solely
on the languages of Western Europe. And it is true that some
such generalizations are refuted by the merest glance at an ap
propriate non-European language. But contrastive study based
exclusively on European languages also has a merit: our knowl
edge of those languages is currently deeper and more detailed than
our knowledge of languages elsewhere, so that generalizing
hypotheses can also be deeper. They may be due for a longer wait
before an appropriately broad survey can confirm or confute
them, but they are valuable nonetheless.
Charles F. Hockett
1.8. Universal features are important if their presence in a

system can be shown nor to be entailed by the presence of
features of the "defining set"; or if they are entailed thereby
but not obviously so.
The notion of a "defining set" was introduced in 1.2. For

examples illustrating the present point, see 5.7 and 5.6.
The second part of the point may need some justification. Map
makers have found empirically that they never need more than
four colors in order to guarantee that any two continuous regions
that share a boundary (not merely a point) shall be assigned
different colors. This is presumably a topological property of
planes and of spherical surfaces, yet it follows so unobviously
from the mathematical definitions of those surfaces that no mathe
matician has yet succeeded in proving the implication formally.
If a proof-or, indeed, a demonstration that five colors are
needed rather than four-is attained, the glory of the achievement
will not be diminished in the slightest by the fact that the con
clusion is implied by the premises.
1.9. A universal feature is more apt to be important if there are
communicative systems, especially nonhuman ones, that do
not share it.
t..
I
I,
ft may seem peculiar at first to propose that we can learn more

about human language by studying the communicative systems of
other animals; but a moment's reflection is enough to show that
we can only know what a thing is by also knowing what it is not.
As long as we confine our investigations to human language, we
constantly run the risk of mistaking an "accidental" universal for
an "essential" one-and we bypass the task of clearly defining
the universe within which our generalizations are intended to
apply. Suppose, on the other hand, that after discovering that a
particular feature recurs in every language on which we have
information, we find it lacking in some animal communicative
system. In some cases, this might lead us to add the feature to
our defining set for language. In any case, this seems to be one
way of trying to avoid triviality in the assembling of our defining
set.
The point Just proposed threatens a very lengthy program of

investigation of the communicative behavior of other animals,
since zoologists recognize approximately one million living
species and one can never be sure just where in this vast collection
some relevant property (or its notable absence) may appear. Who
would have thought, fifteen years ago, that we would learn
something crucial in linguistics from bees!
It might be suggested that we bypass the whole task by an ap
propriate definition. We could simply assert that a communicative
system is not a language unless it is manifested by human beings.
Good enough; but we must now ask our confreres in anthro
pology and biology to identify for us the class of human beings.
A serious reply is apt to include the remark "human beings are
hominoids that talk"-and the circle has been closed and nothing
achieved. We had better define language without reference to hu
man beings. Then, if it appears that-on our planet-only human
beings talk, this becomes a significant empirical generalization.
The comparison with nonhuman communication can be
revealing in another way. We have already noted that many
languages have a nursery word like mama. If we ask whether the
gibbon-call system has this feature, we find it embarrassing
merely to give the technically correct negative answer. The
question has been put badly. One hesitates to speak of "words"
in discussing gibbon calls. Thus we are led to examine more
closely what we mean in speaking of "words" in various lan
guages, and just why we are uncomfortable using the word
"word" for gibbon calls; and such lines of inquiry may yield a
more meaningful comparative question and a significant general
ization about language.
1.10. The problem of language universals is not independent of
our choice of assumptions and methodology in analyzing
single languages.
This is a terribly unstartling proposal, yet important. We must
generalize from our information about specific languages; we
must collect information about a specific language in terms of
some general frame of reference. The latter includes notions as
r
f
Charles F. Hockett
to what language must be as well as points of methodology. The

study of individual languages and the search for universals thus
stand in a dialectic complementation that can equally well
propagate error or truth.
To force such a system of investigation toward truth, the point
of entry is our way of manipulating data on specific languages;
and the procedure is the familiar one of contra positive assump
tion. Whatever one's favorite notion about language design, one
asks "if I assume that this particular language does not conform
to my pet scheme, can I describe it satisfactorily?" A single
success refutes or requires the revision of one's pet hypothesis. A
failure, on the other hand, merely means that the hypothesis is
still tentatively usable. Hypotheses, about language universals or
anything else, are by definition proposals to be knocked down,
not beliefs to be defended.
2. The Search for Universals through Comparison with Animal
Systems
The design features listed here are found in every language on

which we have reliable information, and each seems to be lacking
in at least one known animal communicative system (cf. 1.9).
They are not all logically independent, and do not necessarily all
belong to our defining list for language-a point to be taken up
separately, in Section 3.
All but the last three of these features have been presented in
detail elsewhere (Hockett, 1960). Exact repetition would be inap
propriate here, and the writer shrinks from the task of replacing
the earlier treatment by a newly formulated one of comparable
detail. Therefore the reader is requested to accept the present
listing as the briefest sort of synopsis, and to turn to the reference
just given for fuller information.
2.1. Vocal-Auditory Channel. The channel for all linguistic
communication is vocal-auditory.
Some animals have communication that is auditory but not

vocal (e.g., crickets); some have systems with totally different
channels (bee-dancing is kinetic-tactile-chemical).
The phrasing of this first design-feature excludes written lan
guages from the category "human language" just as it excludes
African drum signals. The exclusion is intentional; the grounds
for it will be discussed later (Section 3).
<.
2.2. Broadcast Transmission and Directional Reception. All
linguistic signals are transmitted broadcast and are received
directionally.
These properties are the consequences of the nature of sound,
of binaural hearing, and of motHity, and are thus implied by 2.1.
"Tight-beam" transmission is rare in the animal world, but occurs
in the nerve-nets of coelenterate colonies. Directional reception is
the general rule, barring occasional masking. An example of the
latter is that in a field full of crickets locating anyone cricket from
its call is difficult, even for another cricket.
2.3. Rapid Fading. AI/linguistic signals are evanescent.
To hear what someone says, one must be within earshot at the

right time. Spoors and trails fade more slowly. The property of
fading is also a consequence of 2.1.
2.4. Interchangeability. Adult members of any speech com

munity are interchangeably transmitters and receivers of
linguistic signals.
Among some species of crickets, only the males chirp, though

both males and females respond to the chirping of others.
2.5. Complete Feedback. The transmitter of a linguistic signal
himself receives the message.
There are pathological exceptions (as, also, to 2.4). In certain

varieties of kinetic-visual communication, as in the courtship
dance of sticklebacks, the transmitter cannot always perceive some
of the crucial features of the signal being emitted.
;1
10
Charles F. Hockett

~
2.6. Specialization. The direct-energetic consequences of lin
guistic signals are usually biologically trivial; only the

triggering effects are important.
Even the sound of a heated conversation does not raise the
temperature of a room enough to benefit those in it. A male
stickleback will not court a female unless her abdomen is distended
with roe; the distension is thus an essential part of her signal to
the male; the direct consequences of the distension are of obvious
biological relevance.
II
arbitrary. But in a discrete semantic system there is no necessary

implication as to iconicity or arbitrariness; therefore, for language,
2.8 is independent of 2.7 and 2.9.
2.10. Displacement. Linguistic messages may refer to things

remote in time or space, or both, from the site of the
communication.
"Remote" means out of the perceptual field of the communi
cators. Gibbon calls are never displaced. Bee dances always are.
Utterances in a language are freely displaced or not.
2.7. Semanticity. Linguistic signals function in correlating and
organizing the life of a community because there are asso
2.11. Openness. New linguistic messages are coined freely and

ciative ties between signal elements andfeatures in the world;
easily.
in short, some linguistic forms have denotations.
We
can
transmit messages (produce sentences) that have never
The distension by roe of the belly of the female stickleback is
been transmitted before, and be understood. Bees do this; gibbons

part of an effective signal, but does not "stand for" something else.
do not.
2.8. Arbitrariness. The relation between a meaningful element
Actually, this property reflects two partially separate facts

in language and its denotation is independent ofany physical
about language that deserve individual mention:

or geometrical resemblance between the two.
2.11.1. In a language, new messages are freely coined by blend

Or, as we say, the semantic relation is arbitrary rather than
ing, analogizing from, or transforming old ones.
iconic. There are marginal exceptions, including traces of ono
This says that every language has grammatical patterning.
matopoeia. In bee-dancing, the way in which the direction toward
2.11.2. In a language, either new or old elements are freely
the target site is mapped into a direction of dancing is iconic. The
assigned new semantic loads by circumstances and con
relation between a landscape painting and a landscape is iconic;
text.
the relation between the word landscape and a landscape is
This says that in every language new idioms constantly come into
arbitrary.
existence.
The openness of bee-dancing might be described as due to a
2.9. Discreteness. The possible messages in any language
very special sort of "grammatical patterning"; surely there is no

constitute a discrete repertory rather than a continuous one.
evidence that bees create new idioms.

Any utterance in a language must differ from any other utter
ance of the same length by at least a whole phonological feature.
2.12. Tradition. The conventions of a language are passed down
Utterances cannot be indefinitely similar to one another. Bee
by teaching and learning, not through the germ plasm.
dances can be: the repertory of possible dances constitutes a
Genes supply potentiality and probably a generalized drive,
twofold continuum.
since nonhuman animals cannot learn a (human) language and
In a continuous semantic system (one with property 2.7 but
humans can hardly be prevented from acquiring one. Bee-dancing
with the converse of2.9), the semantics must be iconic rather than
is probably genetic.
t:
12
Charles F. Hockett
13
2.13. Duality (of Patterning). Every language has both a cene

can hardly be generated, though false ones can: a gibbon could,
malic subsystem and a plerematic subsystem.
in theory, emit the food call when no food had been discovered.
Perhaps, however, one can imagine a system with these three
More commonly, we speak rather of the phonological and
underlying properties used by a species (or a collection of ma
grammatical (or grammatico-lexical) subsystems of a language.
chines)
that never lied.
The unusual terms, borrowed from Hjelmslev, are more appro
It
ought
to be noted that without the property here labeled
priate for the discussion of communication in general, since they
"prevarication"
the formulation of hypotheses is impossible.
circumvent the unwanted connotation that the physical channel of
a system with duality must necessarily be sound waves.
2.15. Reflexiveness. In a language, one can communicate about
By virtue of duality of patterning, an enormous number of
communication.
minimum semantically functional elements (pleremes, morphemes)
Bees dance about sites, but they cannot dance about dancing.
can be and are mapped into arrangements of a conveniently
This
property, also, is presumably derivative, resting largely on
small number of minimum meaningless but message-differen
2.11.2.
tiating elements (cenemes, phonological components). No animal
A tempting alternative to this property is "universality": in a
system known to the writer shows a significant duality.
language
one can communicate about anything. Reflexiveness
Some contemporary investigators strongly suspect that a human
would
obviously
follow from universality. The difficulty is an
language involves not just two, but at least three, major
empirical
one:
if
there are indeed things that we cannot com
subsystems: for example, "phonemic," "morphemic," and
municate about, the fact that we cannot communicate about them
"sememic." For our present purposes this possibility can be
may prevent us from recognizing that they exist. Anyway, the
setaside with the remark that a system with "triality" of patterning
idiom-forming mechanism of openness (2.11.2) guarantees that
would a fortiori have our property of "duality." The essential
we can come to communicate via language about anything that
contrast is between one and more than one subsystem.
we are capable of experiencing.
2.14. Prevarication. Linguistic messages can be false, and they
2.16. Learnability. A speaker of a language can learn another
can be meaningless in the logician's sense.

language.
I can assert that it is ten miles from the earth to the moon, or
In a science-fiction story (wisely rejected by all editors), the
that the interior of all opaque solids is green until exposed to
writer once invented a nonterrestrial species that had a com
light. Lying seems extremely rare among animals.
municative system like human language in all respects except
This feature is not independent. It would seem to rest on se
that
its conventions were transmitted entirely through the germ
manticity (2.7), displacement (2.10), and openness (2.11). Without
plasm.
The members of this species could learn a new language,
semanticity, a message cannot be tested for meaningfulness and
but
only
with terrible effort. On earth, at least, it seems likely that
validity. Without displacement, the situation referred to by a
the
relative
ease with which humans can learn other languages rests
message must always be the immediate context, so that a lie is
on design feature 2.12.
instantly given away. Without openness, meaningless messages
There is probably more of this sort of flexibility of readaptation
... George L. Trager and Sydney M. Lamb have been exploring the
among animals than we give them credit for; but some systems,
"triality" notion (or even more complex proposals), as yet without published
at least, lack the feature altogether (bee-dancing, stickleback
accounts to which reference can be made. The present writer's most thorough
courtship).
going discussion of duality is Hockett, 1961.
\,
14
Charles F. Hockett
3. Definition and Basic Hypotheses

The design features just presented are admittedly diverse. The
list was originally assembled not in a search for language univer
sals, but rather through a series of comparisons of human speech
with the communicative behavior of certain other animals. It
includes any point that such a comparison suggested. Thus it
comes about that some of the points apply directly to a language
as an "abstract" system (though "abstract" means different things
to different investigators); others rather to the organisms that
use the system; still others to how the organisms use or
acquire the system. This is also why some of the points mention
physics or biology (a most un-"abstract" policy), while others
do not.
In reweighing the sixteen design features for our present pur
poses, the first decision we must make concerns writing. Shall we
attempt a defining set of properties that subsumes writing systems,
or some of them, as well as "spoken languages"? Or shall we class
writing systems with drum signals and other clearly secondary
and derivative phenomena, as something apart from "language"?
Either choice can be justified. In .the long run we should
probably do both. But in this paper I shall exclude writing. The
reasons are as follows:
1;:
1. Spoken language is part of the '''common denominator of

cultures," and its antiquity is undisputed. Any generalization
about spoken language is also a hypothesis about human cultural
universals (Murdock, 1945). Writing is a recent invention, and
has not yet spread to all human communities. Although this in
itself does not preclude an attempt to determine what all spoken
and all written languages have in common, it seems reasonable to
break the total task up in a way that correlates with cultural
universality and its absence.
2. One crucial design feature of writing systems is relative
permanence, the exact opposite of the rapid fading (2.3) charac
teristic of spoken language. If we try to characterize spoken and
written language at the same time, we have to omit both rapid
fading and relative permanence. But the relative permanence of
r
r
15
writing is an important source of its enormous power; and the

rapid fading of speech (and of its prehuman precursors) was a
crucial factor conditioning the evolution of human communi
cation of all varieties. The joint consideration of spoken and
written language can thus best follow the detailed consideration
of the two taken separately.
3. Writing systems are quite varied in their designs, so that it is
difficult to be sure just what features are common to all. Do
writing systems have duality (2.13)? In one view, only a few do.
The Ogam script, for example, had cenemes consisting of certain
elementary strokes, and pleremes represented by certain arrange
ments of those strokes; the denotations of the pleremes were the
phonemes of Old Irish. In this view, English writing does not have
duality of patterning, because our pleremes (letters) are not built
out of a small stock of simpler cenemes. If we shift ground and
say that a writing system has duality in that it shares (essentially)
a plerematic subsystem with the correlated spoken language,
manifested cenematically in "phonic substance" in speech but in
"graphic substance" in writing, then how do we distinguish
between the Ogam script and contemporary English writing, or
between the latter and Chinese '!
Clearly, these questions can all be answered. The writer claims
the privilege of not attempting the answers here.
Having made this decision, we can consider the following de
fining set for language: openness (2.11), displacement (2.10),
duality (2.13), arbitrariness (2.8), discreteness (2.9), interchange
ability (2.4), complete feedback (2.5), specialization (2.6), rapid
fading (2.3), and broadcast transmission with directional reception
(2.2). Any system that has these ten properties will here be called
a language; any language manifested by our own species will be
called a human language. Every language also has semanticity
(2.7), since the contrast between arbitrariness (2.8, included
in the defining set) and iconicity is meaningless without it. Pre
sumably, but not so clearly, every language has prevarication
(2.14) and reflexiveness (2.15); at least, every human language
does.
16
Charles F. Hockett
To show the importance of the features of the defining set, we

can think of human language as we know it and consider the con
sequences of suppressing, in turn, each feature.
A language deprived of openness would generate only a finite
number of whole messages. Lying might be possible, but hy
pothesis-formation would not.
A language deprived of displacement would not allow its users
to communicate about the past or the future. Planning would be
impossible. Fictions-hence speculation, literature, science
would be precluded.
A language deprived of duality would be extremely cumber
some, since each plereme would have to differ holistically from
each other. It is hard to imagine any species remotely like our own
being able to handle-or, at least, to evolve-such a system.
However, perhaps duality is simply the mammalian way of
achieving a system with all other relevant properties. Some extra
terrestrial species might do differently.
A system without arbitrariness either lacks semanticity alto
gether or else has iconic semantics. The former possibility is most
unlike language. A system with iconic semantics is constrained
to use about things and situations that can be imitated, pictured,
or diagrammed. Swift's account of Gulliver's encounter with
the Laputans should be enough to show the crucial importance
of arbitrariness.
The alternative to discreteness is .continuous repertories of
signals, as among the bees. But a continuous semantic system
necessarily has iconic semantics (Hockett, 1960, p. 413).
It is interchangeability that enables a human to "internalize"
the roles of others and to carryon conversations with himself, thus
carrying over to the situations in which he is temporarily alone
the problem-solving powers of language.
Complete feedback also seems essential to the use of language
just described.
Specialization is such a general property of communicative
systems (human and animal) that some investigators hesitate to
use either the term "system" or the term "communication" of
types of behavior from which it is absent. In any case, special-
II
i
17
ization renders possible communication at a power-level (in the

literal physical sense) that is convenient for the species involved.
One does not have to increase the power-level to communicate
about large-scale matters, or to reduce it when concerned with
minutiae.
Rapid fading means, positively, that messages already trans
mitted do not clutter up the channel and impede the transmission
of new ones (as happens sometimes when one has a blackboard
but no eraser). Thus, emergency signals can get through. On the
other hand, it implies that the import of a message has to be
stored internally in the receiver if it is to be stored anywhere at
all. The "attention span" required of human hearers to take in a
long and involved sentence is considerable, when measured on the
general animal scale. The evolution of the capacity for such an
attention span has surely been conditioned by the rapid fading
property of vocal-auditory communication, and is related to the
development of displacement, as well as to such nonlinguistic
matters as tool-carrying and tool-making. Rapid fading is not an
"incidental" property of human language. When its undesirable
implications were overcome, by the development of writing, a
major revolution had occurred.
Broadcast transmission and directional reception also carry
both advantages and disadvantages. A warning cry may tell all
one's fellows something of the location of the danger, but also,
if the danger is a predator, it tells the predator where one is.
If we think only of the modern "civilized" world, in times of
peace, rapid fading, broadcast transmission, and directional
reception may seem relatively unimportant. But if we think of
the living conditions prevalent during the bulk of human history,
we see that these properties are not lightly to be regarded as
secondary. They are part of our heritage from prehuman times;
they have conditioned our own evolution and that of language;
and they are still with us, their potentially deleterious effects
canceled out only under special technological circumstances.
There is, nonetheless, a sense in which openness, displacement,
and duality (together with traditional transmission, which does
not appear on the defining list) can be regarded as the crucial or
18
Charles F. Hockett
nuclear or central properties of human language. From an

examination of what is known of the vocal-auditory communi
cative systems of contemporary nonhuman Hominoids, it seems
that the vocal-auditory system of the proto-Hominoids must, at
least, have lacked these three or four features. These three or
four, then, are human or Hominid innovations. Otherwise, human
language is not truly distinguishable from Hominoid com
munication in general.
Now we are ready for some generalizations that go beyond the
defining set.
3.1. Every human community has a language.
Surely no one will couQter with the instance of a Trappist

monastery: there would be no need for a rule against talking if
talking were not a possibility.
3.2. No species except our own has a language.
This may be disproved at any time by new zoological discov

eries. No guess either way is implied about extinct species and
genera of the Hominids (Homo neanderthalensis, Pithecanthropus,
Australopithecus).
3.3. Every human communicative system usually called a
(spoken) language is a language our sense.
The writer is disturbed by the possibiHty that a few human
systems not ordinarily called "spoken language," and that we do
not wish to include, may also fit the definition; for example,
Mazateco whistle-talk (Cowan, 1948). The derivative status of
such a system is obvious, but it is not clear just how to provide
formally for its exclusion.
i"
3.4. Every human language has the vocal-auditory channel (2.1).

This feature was excluded from the defining list because it
seems that its implications (broadcast transmission, directional
reception, rapid fading) are structurally more important, and
one can imagine other channels-say light, or heat-waves-that
19
would yield the same implications. Therefore this assertion is not

trivial.
3.5. Every human language has tradition (2.12).
If we design and build a collection of machines that communi
cate among themselves with a language, this property will be
lacking.
3.6. Every human language has learnability (2.16).

Probably this is a corollary of the preceding.
3.7. Every human language has both an intonational system and
a nonintonational system; this dichotomy cuts across that
into cenematics and plerematics.
English, for example, has segmental (nonintonational) mor

phemes that are mapped into segmental phonological features,
and intonational morphemes that are mapped into intonational
phonological features. A speaker transmits, simultaneously, a
nonintonational and an intonational message. The hypothesis
is a guess that this basic organization is true in all human lan
guages. It does not imply that the phonic "raw material" for
intonation is invariably the pitch of the glottal tone, as it is, in
large part, for English.
If true, this generalization is striking, since there seems (at the
moment) absolutely no reason why an otherwise languagelike
system should have this property. Most writing systems do not
carry it over.
Another generalization about intonation is tempting, on the
basis of very limited observation, but the evidence is scarcely
strong enough to present it as a numbered point: Many highly
diverse languages (English, other languages of Europe, Chinese,
Japanese, Samoan, Fijian) share a "most colorless" intonation
for flat statements, in the face of (1) different phonemic structures
for the intonation (which is phonetically similar from one language
to another) and (2) wide disparity in the remainder of the into
national system.
20
Charles F. Hockett
3.8. In every human language, plerematic patterning and cene~

malic patterning are both (independently) hierarchical.
Grammatically, an utterance consists (let us say) of clauses, a
clause of phrases, a phrase of words, a word of morphemes.
Phonologically, an utterance consists of macrosegments, a mac
rosegment of microsegments, a microsegment of syllables, a
syllable of phonemes, and a phoneme of phonological com~
ponents. (Except for morpheme, phonological component, and
perhaps utterance, the terms used for this explanation are not part
of the generalization.)
3.9. Human languages differ more widely in cenematics than
in plerematics.
3.10. Human languages differ more widely, at least in their
plerematic subsystems, at small size-levels than at large.
These two assertions are not of universals, but perhaps point
toward. some. For example, 3.10 suggests that all languages share
certain large-scale syntactical patterns, however varied may be
the smaller~scale patterns by which the constituents for the larger
patterns are built up. Point 3.9 can be challenged on the grounds
that we have no reliable way of measuring and comparing the
differences referred to. At present this is doubtless true; but the
assertion seems impressionistically valid to the writer, and formal
ways to confirm the impression (or to disprove it) may be found.
4. Grammatical Universals
The generalizations of the preceding section mention grammar
(or plerematic design), but do not belong in a set of generalizations
about grammar proper because they involve the relationship of
grammar to other aspects of language design. From what has
already been said, we know (or assume) that every language has
a grammatical system, and that grammatical patterning is hierar
chical. In addition, we can with reasonable confidence propose
the following points:
21
4.1. Every human language has a stock of elements that shift

their denotations depending on elementary features of the
speech situation.
That is, every language has deictic elements ("substitutes," in
Bloomfield's terminology): in English, the personal pronouns,
demonstrative pronouns and pro adverbs, and so on.
4.2. Among the deictic elements of every human language ;s
one that denotes the speaker and one that denotes the
addressee.
The first and second person singular pronouns are universal.
There seems to be no reason internal to our definition of language
why this should be so; yet, if we try to imagine a system that lacks
them, the results seem quite alien.
4.3. Every human language has some elements that denote
nothing but that make a difference in the denotation of the
composite forms in which they occur.
Such elements are markers, for example English and. Match
and book denotes something different from match or book or
match book, but and denotes nothing. The assumption that such
elements must denote something just as do man, sky, honor, or
unicorn has generated much bad mentalistic philosophizing,
populating the universe with abstract entities or the human mind
with concepts, both of which are as useless as the luminiferous
ether.
There are also impure markers, for example English in, on,
that have some denotation as well as a marking function. It may
be that we should go only so far as to assert the universal presence
of markers (pure or impure).
.1
4.4. Every human language has proper names.

A proper name is a form that denotes just what it denotes. If
it denotes more than one thing in different occurrences, the class
of things that it can denote has no criterial property in common
i'
1.:1
22
Charles F. Hockett
other than the (extrinsic) property of being denoted by the proper

name. All Americans named Richard are probably males, but
many males are not called Richard, and when one meets someone
for the first time, it is in no way possible to examine his properties
and infer that his name must be Richard.
A form may be a proper name and also something else: Robin/
robin, John/john, Brown/brown. The generalization does not deny
this.
4.5. Every language has grammatical elements that belong to
none of the three special categories just itemized.
Comparatively, it is worthy of note that all signals in bee

dancing are deictic elements, and that no gibbon calls are of any
of the three special types.
4.6. In every human language there are at least two basic orders
of magnitude in grammatical patterning.
Where there are just two, the traditional terms "morphology"
and "syntax" do very well. When the morphology-syntax boun
dary appears fuzzy, closer scrutiny often reveals a separate order
of magnitude of grammatical patterning sandwiched between. As
familiar a language as Spanish offers an example. The internal
organization of dando, me, and 10 is mofphology; the participation
of ddndomelo in larger forms is syntax; the patterns by which
dando, me, and 10 are conjoined to yield ddndomelo are not
conveniently classed as either.

However, 4.6 is shaky in another direction: a deeper under
standing of languages of the Chinese type may yet show that they
are best described without either the two-way morphology-syntax
dichotomy or a more complex three-way layering.
In many languages in which the morphology-syntax dichotomy
is clear-cut, phonological patterning correlates: that is, gram
matical words are also, for the most part, phonological units of
a distinctive sort. But there are many exceptions, so that this
points toward morphophonemic taxonomy rather than toward
universals.
23
4.7. Apart from the three special categories of elements already

mentioned (deictic elements, markers, and proper names),
no human language has a grammatically homogeneous
vocabulary.
There are always forms with different ranges of privileges of

occurrence, so that one can always validly speak of form classes,
\
4.8. A major form-class distinction reminiscent of "noun" versus

"verb" is universal, though not always at the same size-level.
This was discussed in connection with 1.5.
I
I
4.9. Every human language has a common clause type with

bipartite structure in which the constituents can reasonably
be termed "topic" and "comment."
The order of the constituents varies. Typically in Chinese,

Japanese, Korean, English, and many other languages, one first
mentions something that one is going to talk about, and then says
something about it. In other languages, the most typical arrange
ment is for the comment, or part of it, to precede the topic. Of
course, the generalization refers only to a "common clause type."
Every language seems to have clauses of other types as well.
4.10. Every language has a distinction between one-referent and
two-referent predicators.
In Mary is singing, the predicator is singing is of the one-referent

sort (and Mary is the referent); in John struck Bill, the predicator
is of the two-referent sort.
Both 4.9 and 4.10 are shaky in a special way. Although we tend
to find these patterns in language after language, it is entirely
possible that we find them because we expect them, and that we
expect them because of some deep-seated properties of the
languages most familiar to us. For some languages, some scheme
that is far less obvious to us might actually fit the facts better.
Although this is true of aU proposed generalizations, it never
theless seems especially true of these two.
If
Ii
I'
i;
24
Charles F. Hockett
25
I'
into some seventy or eighty phonemes, which in turn occur in

about twice that many syllables. Each syllable consists of one of
the seventy-odd consonant phonemes, followed by one of the two
From what has already been said, we know (or assume) that
vowel phonemes. It seems clear in such a case that the vowel
every human language has a phonological system, and that
"phonemes" are better regarded simply as two additional phono
phonological patterning is always hierarchical. Purely phono
logical generalizations are then to be considered within that
logical features, so that a unit such as Ikal is just a phoneme-or,
alternatively, that the term "phoneme" be discarded and one
tentatively established framework.
discuss the participation of features directly in syllables. Either
5.1. In every human language, redundancy, measured in pho
way, one does not need both the term "phoneme" and the term
nological terms, hovers near 50%.
"syllable." The case may be extreme, but it is real, and under
scores the importance of the "anti-universal" given as 5.2.
The notion is that if redundancy tends to increase much above
this figure, communication becomes inefficient, and people speak
5.3. Every language makes use of distinctions of vowel color.
faster or more sloppily, while decrease much below the figure leads
Vowel color is defined as combination of formants. Acousti
to misunderstanding, and people slow down or articulate more
cally,
it is known that for languages like English differences of
clearly.
vowel
color do much of the work of keeping consonants apart,
It may be that the redundancy figure would be about the same
as well as distinguishing vowel phonemes.
were it measured in grammatico-lexical terms; and it may be
that this approximate figure is the rule for a wide variety of com
5.4. A historical tendency toward phonological symmetry is
municative systems, at least among human beings. Printed

universal.
English yields the same figure (Shannon, 1951), in terms of

Jakobson has offered a number of synchronic generalizations
letters.
about phonological systems, to some of which there seem to be
a few marginal exceptions. One, for example, is the assertion
5.2. Phonemes are not fruitful universals.

that a language does not have a spirant of the type [6] unless it
We can, indeed, speak quite validly of phonemes in the discus

has both a [t] and an [s], nor an affricate like [c] unless it has both
sion of any language, but their status in the hierarchy of phono

a {t] and an [8]. However, Kickapoo has [t] and [6] but no [s].
logical units varies from one language to another, and also, to

Another is that a language does not have nasal continuants at
some extent, through varying preference or prejudice of analysts.

more. contrasting positions of articulation than it has stops of
The status of phonological components, on the other hand, is

some one manner of articulation. It is possible to analyze certain
fixed once and for all by definition-phonological components

varieties
of Brazilian Portuguese so as to violate this generaliza
are the minimum (not further divisible) units of a phonological

is that a language does not contrast unaspirated and
tion.
A
third
system. Given that all phonological patterning is hierarchical,
aspirated
stops
unless it has a separate phoneme Ih/ Mandarin
the exact organization of the hierarchy, varying from one lan

is
almost
an exception, in that the nearest thing to an
Chinese
guage to another, becomes a taxonomic consideration of impor
is normally a dorso-velar spirant.
tance, but not the basis of a generalization in the present context.

Yet these generalizations seem far too widely borne out to
There are certain languages of the Caucasus (Kuipers, 1960)

merely
be thrown into the scrap heap by virtue of a handful of
where one can, if one wishes, describe the phonological system

exceptions.
When facts invalidate a hypothesis, one tries modify
in terms of perhaps a dozen phonological features organized
5. Phonological Universals
!
~.1
,
26
.,
"
Charles F. Hockett
ing the hypothesis before one discards it altogether. In each of

the cases given, we seem to have an indication of a historical
tendency toward some sort of symmetry. The tendency can be
disrupted, so that not every system viewed in synchronic cross
section will conform to the rule; but diachronically the tendency
is real.
5.5. There are gaps, asymmetries, or "corifigurational pressures"
in every phonological system, no matter when examined.
Most systems, by virtue of a sort of semi magical logistics of
maneuvering on the part of analysts, can be forced to appear
neat and symmetrical. The maneuvering is always worth under
taking, not in order to force symmetry where there is lack of it, but
because it is heuristically valuable-it helps to show relationships
within a system that might otherwise be missed. But the asym
metries, however pushed about, remain in the system.
5.6. Sound change is a universal. It is entailed by the basic deSign
features of language, particularly by duality of patterning.
By "sound change" is meant a mechanism of linguistic change
that is not reducible to other mechanisms (see, e.g., Hockett,
1958, chs. 52-54). When a system has duality of patterning, the
basic role of its cenematic system is to identify messages and keep
them apart. Usually an utterance produced in given circumstances
is far more than minimally different from any other utterance that
might be produced in the same language in the same circum
stances. Thus there is room for much nondistinctive variation in
details of articulation and even more in the shape of the speech
signal by the time it reaches the ears of the hearer. Therefore there
will be sound change. The implications of sound change for the
phonological and grammatical systems of a language are another
matter. (See the reference given earlier.)
5.7. Every phonological system contrasts phonemes that are
typically stops with phonemes that are never stops.
Stops are sounds produced with complete oral closure and com
plete velic closure. By "phonemes that are typically stops" we
27
mean phonemes that are stops in slow careful speech or in key

environments, though they may be weakened or spirantized in
some environments or in faster speech. The contrasting nonstops
vary widely from one language to another. In a few languages of
New Guinea, the nearest nonstops are nasal continuants. More
commonly, they are spirants.
5.8. No phonological system has fewer than two contrasting
positions of articulation for stops.
The only attested cases with two are Hawaiian and a slightly
archaic Samoan, with labial versus lingual. (In contemporary
Samoan a new apical-versus-dorsal contrast has developed.)
5.9. If a language has a vowel system, it has contrasts of tongue
height in that system.
5.10. If we define a "vowel system" to include all the segmental
phonemes that occur as syllable peaks, then every language
has a vowel system.
For 5.10, clearly some adjustment is required in order to

subsume the languages of the Caucasus referred to earlier. If we
define a "vowel system" to include all the segmental phonemes that
occur only as syllable peaks, then at least one language, Wishram,
apparently has a vowel system of one element, which is only
trivially a "system." With these adjustments, 5.9 becomes a true
universal, applying to all human languages.
Another way to express 5.9 is to say that if a language has vowel
contrasts other than those of tongue height, it also has those of
tongue height, but not necessarily vice versa.
Further generalizations along the line of the last three men
tioned can probably be formulated, although all of them are
subject to modification at any time by empirical information on
some as-yet-unanalyzed language. As a set, however, they point
toward something rather puzzling. It would seem easy enough to
devise a phonemic system that would have no stops at all, or no
vowels at all, or the like. The phonological systems of th~ world,
28
Charles F. Hockett
despite their great variety, all seem to have more in common than
is strictly "necessary." That is, the degree of resemblance strikes
one as greater than is required merely by the defining features of
language and the known cultural and biological properties of our
species. Granting that the variety may actually be somewhat
greater than we currently realize, there is still a problem in this
degree of similarity. Are there constraints imposed by as-yet
unrealized properties of the organs of speech and of human
hearing? Is the resemblance due to a common origin, in relatively
recent times-say forty or fifty thousand years ago-of all
human languages on which we have any direct evidence or can
obtain any? (The latter hypothesis does not, of course, propose
that human language is only that old, merely that all other older
strains have died out.) These questions are open; the answers may
actually lie in some totally different direction.
ACKNOWLEDGMENTS
The author is indebted to Sidney Lamb for detailed criticisms and
suggestions. He also wishes to thank Fred Householder and Joseph H.
Greenberg for comments on certain aspects of the present paper.
REFERENCES
Bloomfield, L. (1933). Language. New York.
Cowan, G. M. (1948). "Mazateco Whistle Speech." Language 24.
280-286.
Hockett, C. F. (1958). A Course in Modern Linguistics. New York.
- - (1960). "Logical Considerations in the Study of Animal
Communication." W. E. Lanyon and W. N. Tavolga, eds.,
Animal Sounds and Communication, Publication No.7 of The
American Institute of Biological Sciences, 392-430. Washington,
D.C.
- - - (1961). "Linguistic Elements and Their Relations." Language
37.29-53.
Jakobson, R. (1961). "Why 'Mama' and 'Papa'?" Perspectives in
Psychological Theory, 124-134.
29
Kemeny, J. G. (1959). A Philosopher Looks at Science. Princeton, N. J.

Kuipers, A. H. (1960). Phoneme and Morpheme in Kabardian. The
Hague.
Murdock, G. P. (1945). "The Common Denominator of Cultures."
R. Linton, ed., The Science of Man in the World Crisis, 123-142.
New York.
Shannon, C. (1951). "Prediction and Entropy of Printed English."
Bell System Technical Journal 30. 50-65.
UNIVERSALS OF LINGUISTIC CHANGE

CHAPTER 2
ARE THERE UNIVERSALS OF LINGUISTIC CHANGE?

HENRY M. HOENIGSWALD
1. Introduction
On the subject of synchronic universals (see Casagrande, 10.2),

linguistic thinking may be said to have gone through three stages.
In the first stage-that period with which contemporary linguistics
has found so much fault because it was its direct heir-a glosso
centric attitude was likely to prevail. Features recognized in a
handful of familiar languages, and recognized, more often than
not, under the guise of spelling or rhetoric or logic rather than on
their own terms, were easily taken to be necessary properties of
language per se. Insofar as this is a fair description, the question of
universals did not arise, since everything was, in a sense, a universal.
The relativism of the stage which followed was extreme.
Nothing in the languages of the world was .to be taken for granted.
The investigator took care to keep his concepts unspecific. They
had to be as formal and as empty as possible so as not to prejudge
the case, and to permit the objective discovery of specific proper
ties in each instance. Perhaps it was safe to believe that all
languages "have" phonemes, morphemes, and constructions; but
it was especially safe if it meant no more than that any language
may be analyzed as made up of utterances that partly do, and
partly do not, mutually contrast. Our findings may then tell us that
all languages observed agree in some respects, and may thus give
us universals by induction.
30
31
It seems to me that the third stage in the handling of synchronic

universals takes off from here. Universals may have to be dis
covered inductively, from a neutral', formalistic foundation. But,
thus discovered, what are they like? There is an ineradicable
conviction that all of them are not mere random coincidences;
that their identification as universals is not necessarily so pre
carious as to be thrown out by tomorrow's addition to our store
of data; in other words, that universals may form some sort of
system in their own right.
This, of course, is not news to our conference. But in turning
from the synchronic point of view to the diachronic, we must first
of an realize that diachronic studies have developed at a different
pace. To be sure, the same tl1ree levels or stages are recogniz
able. But, as has often been said, even the prerelativistic first stage
of nineteenth-century historical linguistics, so far from merely
being the old rehash of Stoic logic, Latin school grammar, and
grammaire raisonnee, was a new and aggressive discipline with
productive, if somewhat improvised, working concepts. The
neogrammarians' ideas on the general (that is, universal) attri
butes of change processes are elaborate and substantive. They
are therefore on the surface much less vulnerable to the charge
of glossocentricity; and they are still very much with us. Any
text on historical linguistics contains detailed generalizations
concerning change processes. If proof were needed regarding the
claims made for them, it would be sufficient to point to a di
chotomy in the presentation of sound change which has almost
become traditional. Schwyzer's Historical Greek Grammar,l for
instance, discusses sound change under two headings: "general lan
guage phenomena" and, presumably, others less generaL Gram
mont's Traite de phonetique, a great classic (which, incidentally,
one hesitates to label prestructural), deals precisely with the uni
versals of sound change. Universalistic beliefs have also long been
held in the area of semantic change. The existence, in all languages
and at all times, of such happenings as narrowing, widening, ele
vation, pejoration, metaphor, and metonymy is a venerable tenet.
The elaborateness and substantiality of older historical lin
guistic results and methods of which these notions are a fair
32
reflection have made a thorough reexamination seem less neces

sary. They have thus retarded the emergence of the second stage
in diachronic studies, let alone the third. We are only beginning
to construct a framework of minimal formal concepts-not a
table of universals at all; but simply a skeleton of irreducibles
which will keep to a minimum any preconceived ideas about proc
esses of change that may be specific characteristics of some lan
guages but not of others. These thereby enable us, conversely, to
discover truly universal change processes, if there are any. This
is not to say that speculation, and frequently valuable speculation,
has not gone beyond that stage, too. As always in the history of
a discipline, one must avoid giving too narrow a chronological
meaning to its successive manifestations. Yet it seems fair to say
that the powers of diachronic linguistics to generalize are at
present circumscribed by the factors and in the fashion here
indicated.
Even in establishing the very simplest generalizations one soon
encounters difficulties which betray weak spots in the foundations.
For instance, it seems reasonable to assert that all languages
change. But this cannot simply mean that, given a certain idiolect,
i, it is impossible, after a lapse of time, to find another idiolect i'
such that the description and analysis of i will fit it as well, since
the same is true of contemporary idiolects. Perhaps it means that
after a lapse of time defined as considerable no idiolects can be
found that, under some such criterion as mutual intelligibility,
"resemble" i, while during the intervening period idiolects can be
found which provide a chain of "similar" forms of speech linking
i with i'. For greater concreteness some means is needed to recog
nize two forms of speech (one, as it happens, more recent than the
other) as having the relationship of ancestor and descendant-or,
in the case of less serious chronological differences, what are
called successive "stages" of the "same" language. Only where this
identity is established can we acknowledge "change" as such.
Many a controversy of the past shows that this is not mere
abstract quibbling over self-evident propositions. Much ink and
bile was spilled to prove that Italian is not a later stage of
Ciceronian Latin, or that Old Persian is indeed the true ancestor
33
of Middle Persian. It is interesting to reflect that these decisions

depend on the so-called comparative method of reconstruction.
Language descent, somewhat paradoxically, is a special case of
language relationship. Some pairs of related languages, when
subjected to the comparative method, yield a reconstruction
essentially identical with one member of the pair. It is thereby
that this member language is defined as the ancestor (or older
stage) of the other member language. A comparison between the
two is then known as a statement of change.
2. Change
Change is a general phenomenon in the sense that every lan
guage (perhaps with the exception of ex-novo creations like stan
dard Norwegian or modem Hebrew, but apparently not excluding
the pidgins) has an infinite ancestry of earlier and earlier stages.
The converse is of course not valid, since languages do become
extinct. These are truisms. But a more specific question has been
raised within the last decade or two which has to do with the rate
of change. It is obvious that the extremes are excluded. After
considerable time there is always a noticeable degree of change,
however measured; on the other hand, language will not change
overnight. It is reasonable to suspect the interplay of two forces
at work: one which holds back change in the interest of mutual
intelligibility at any given time; and another force, much more
obscure in its working, which makes for change, even intense
change. That this picture should not be oversimplified is suggested
by the famous Charmey experiment and its follow-up by Hermann,
who showed that what had been the younger generation did not
persist in all of its initial innovations but, in growing older,
moved into the subsociety of the middle-aged and in doing so also
adopted the subsocietal dialect much as it had existed before.
More such studies should be made now that some of the difficulties
in the way of making them have become less forbidding. However
that may be, Swadesh and his associates have asked whether the
long~term results of that balance of forces may not be a signifi
cantly constant rate of change and thus a true, specific universal. II
34
It is worth observing that this glottochronology is, in particular,

lexicostatistics. What is measured is the replacement of items in
the so-called basic vocabulary-a factor which some would regard
relatively minor as change processes go. This is not adverse
criticism; quite the contrary. It is necessary to understand the
surface in order to reach greater depths. It may also be that the
clear-cut and sometimes (not always!) enumerable sudden
switches in vocabulary are at the same time the events interfering
most dramatically with that mutual intelligibility which acts as
the governor on the engine of change-if indeed it does. In other
words, a clear-cut near-constant rate of change may apply more
reasonably to vocabulary replacement than to other varieties of
change. A suggestion has been made also that the rate of vocab
ulary loss, while not an absolute universal, may yet be constant
in a given language family or language area. Finally, we note that
lexicostatistics casts a new light, obliquely, on the old motif that
literacy holds back linguistic change. There is of course no reason
to think that this is really so; but in the special case where a
language borrows, in the "learned" faShion, from its own liter
arily preserved ancestor, vocabulary change (though not other
change processes) may appear retarded, since at least some such
borrowings will not be distinguishable. from material that is
simply retained. Thus it would not be literacy in general, but
Italian or Hindi literacy rather than English or Japanese literacy
which can be expected to show that minor effect.
Swadesh's work has the even greater merit of throwing into
relief the distinction between replacement processes and others.
For the moment We are back at that second level in the search
for universals, which we need so badly if we want to keep our
bearings. For this purpose, we may picture the analysis of lin
guistic change as being performed upon a "translation"-a trans
lation, ideally, of the texts of the earlier stage into texts of the
later stage. 3 (I need not emphasize that by "texts" I am not refer
ring to anything literary or recorded; I mean "self-sufficient
portions of utterances as they can be elicited," or "discourses.")
Now we cannot expect all the texts or discourses of the earlier
stage to have such a "translation." Nor can we expect the reverse:
...
35
the later stage includes texts which are lacking in the earlier.
This is so because the opportunity for the utterances of a given
text may disappear, or, on the other hand, it may not yet exist.
The missing text is potentially there in predictable shape but for
the lack of a stimulus. The change, as the saying goes, is in the
world at large, but not in the language. In fact, the circumstance
that a change in stimuli should produce the obsolescence or fresh
emergence of utterances is precisely a measure of the constancy
of the language. And what holds for entire texts may also be
asserted for their component elements. Words limited to obsolete
texts disappear, as the terms of medieval crafts and trades
disappeared from English. Words limited to newly emergent texts
are new: loanwords like giraffe or coffee were once new. Other
elements are adequately described as partly (or, rather, condi
tionally) dropped, or as conditionally added. Whelm has survived
after over-, but it has dropped out elsewhere, along with the texts
in which it figured. Rail in rai/splitter is relatively old, but railroad
or rails held firm at closing are recent additions.
3. Replacement Pattern
True linguistic chang\,':, however, involves replacement. It is

pointless to ask what word was replaced by coffee, or what word
in active usage has replaced juller. But, clearly, in our hypo
thetical quasi-translation or matching procedure, the earlier
English inwit has gone to conscience just as, merely in comparing
yesterday'S English with today's English, yesterday'S coffee and
conscience are trivially replaced by today's coffee and conscience.
There are very good reasons why the distinction between plain
amorphous addition and deletion, on the one hand, and replace
ment, on the other, must remain somewhat blurred at certain
points; but the distinction is nevertheless of central importance
for any unified view of linguistic change. To the extent that there
is replacement-that, wherever inwit drops out, conscience takes
its place-and to the extent that the replacement is neat, there is
change in the proper sense. The replacement process par exceIIence
is of course sound change. After 100 years of practice we are
~
36
thoroughly familiar with such formulations as that Germanic Idl

"becomes"-that is, is replaced bY-/tl in standard German, or
that proto-Algonquian lei and It I both "go to" the ItI of Cree.
Mere amorphous addition to and deletion from the language
cannot be among the typical effects of sound change.
It does make a difference whether we make our statements, as
we have just done, for the morphemes, phonemes, distinctive
features, constructions, etc., of our two languages; or else for
the morphs, phones, order arrangements, zeros, and what not,
all of which may be said to occupy or fill the distinctive units
comprised in the first list. In the former instance the emphasis
is on the pattern of replacement itself. Insofar as all Germanic
Id/'s are replaced only by German It/'s, and insofar as the only
source for a German ItI is a Germanic Idl, the replacement pattern
is one-to-one; and in one important respect" no structural change
has occurred. Just so, to the extent that conscience moves in only
where inwit has moved out, the vocabulary structure is, in a sense,
unaltered. 5 We might say that the phoneme and the morpheme
(or morphemic construct-the words are compounds) have
remained what they were in a table of mutually defined, Saussurian
elements, and that only their phonetic or their morphic content
has shifted: a phone, [t1, has changed its location; a morph, con
science, has done likewise, even more radically. But then, all
replacements are not one-to-one; there is merger, and there is
split. The plural morpheme of the modern Indo-European lan
guages has replaced both the dual morpheme and the plural
morpheme of the ancestor stages. Spanish tio as well as French
oncle replace two different, contrasting morphemes designating
the paternal and the maternal uncle in Latin. In Cree, as we
have said, both lei and ItI are replaced by It/. On the other hand,
a Latin expression like hominem "translates" into Spanish in some
contexts (that is, in some environments) as un hombre, in other
contexts as el hombre, and in still others as hombre. Or, an older
lui goes to Old English Iyl in some environments, but to lui in
others.
These are examples of various replacement patterns, both in
morphemics and in phonemics. For phonemics this is indeed the
37
customary style of statement. Nor is it entirely unfamiliar in

dealing with morphemes. For instance, it is not unusual to say that
the "concept expressed by the word inwit is later expressed by the
word conscience," or that "the name for a certain part ofthe body
changes from (Old English) wonge to cheek." But much more
commonly talk about such changes is couched in exactly opposite
terms. It is the morphs-that is, the stretches as defined by their
phonemic shape rather than their own contrastive status-that
are followed as they move through the framework of the table of
morphemic contrasts, a framework which in itself may remain
rigid or undergo its peculiar splits and mergers, as the case may be.
From this more usual point of vantage conscience is classified as
a "borrowing." Tio is a borrowing also, although its role in the
replacement pattern is quite different from that of conscience.
One/e, although its replacment function neatly duplicates that
of tio, is a case of semantic widening inasmuch as the morph is the
"same" (under sound change, disregarded on this level) as the
Latin morph which used to fill one of the two morphemes ('mater
nal uncle') that were later merged (into 'uncle'). So is cheek, as
a morph, an example of semantic change; it used to mean 'jaw'
and was replaced by jaw.
With this in mind, what may we expect to be the nature of any
diachronic universals? Conceivably, such universals might be
of three kinds. First, the replacement process itself may turn out
to have certain generally valid properties. 8 Second, there may
be universals of the sort that make the nature of the resulting
language structure predictable-either quite generally (in the sense
that all languages work toward certain identical goals) or specif
ically (in the sense that the resulting structure can in some degree
be predicted from the initial structure). Third, there may. be
something predictable about the movements of the stretches of
lower-level linguistic material-phones and morphs 7-aS they
take up their positions in the interests either of structure preserva
tion or of structural innovation.
First, then, consider the replacement process itself. We have
argued all along on the assumption that the concept of replace
ment is generally valid. To be sure, this is only a "second stage"
38
concept, so markedly formal that perhaps it deserves the name

"universal" no more than does the principle of the synchronic
amenability of all language to phonemic analysis, or the like.
But there are certain corollaries. Above all, there is the much
labored "regularity" of change. Sound change (certain raw effects
of borrowing and of "allophonic analogy" are not considered
here), for example, is known to be primarily merger, and only
secondarily split. Two phonemes may be replaced by one without
producing split, but two phonemes will not replace one without a
merger somewhere. Perhaps the machinery involved here is truly
still something of a puzzle. But it is far more than likely that a
great many mergers, at least, must go forward as the result of con
tact among subphonemically differing varieties within the speech
community. S The minimal extent of the difference may be ex
pected to favor something like "total borrowing," complete with
the "misinterpretation" of diaphones so commonly observed in
the cruder forms of borrowing from language to language. To
a very large extent the so-called regularity of sound change flows
rather directly from a universally present condition: namely from
the subdistinctive heterogeneity of speech communities. Recent
comparative work from all over suggests that sound change "is"
indeed regular everywhere on earth. This means, incidentally, that
the comparative method of reconstruction based on that regularity
is universally applicable. Disdainful opinion to the contrary
effect, even if held by serious scholars, is hardly more than a bit
of naIvete.
It would be wrong to think, and it would hamper us if we did
think, that the regular, "Iaw"-like character which impressed
earlier generations of scholars so much is a mysterious prerogative
of phonemic change. [ts exact morphemic counterpart is plain for
all to see, and indeed is so much on the surface that it has been
remarkable only to those interested in this very analogy. After all,
just as all Algonquian instances of leI go to Cree It/, so all in
stances of in wit go to conscience. Likewise with conditioned change
(split): Just as those particular instances of luI which were
followed in the next syllable by an Ii! went to Iyl, so also those
particular instances of early modern Englishflesh which occurred
39
in certain constructions (say, in jleshwound or mort(fy the flesh)

went to a later morpheme occupied by a homonymous morph
flesh, whereas the other instances of flesh were replaced by a
morph-and-morpheme meat. I have treated certain disparities in
this confrontation elsewhere-I believe there is a satisfactory
theory for them which strengthens rather than weakens the
paralleL 9 According to Leumann, the mechanics of semantic
change parallels very closely what we have been able to suggest
for sound changes: 10 Semantic changes are essentially dialect
borrowings, reinterpreted in the receiving dialect, through the
agency of a "misunderstanding"-the word sometimes to be
taken in its weakest possible meaning. Martinet has made fun of
the notion that a homogeneous language, if left alone, would
remain immutable. But since no language community is homo
geneous and none isolated, it is not so unreasonable to link the
universal incidence of change with the observed universality of
synchronic differentiation and of outside contacts.
Perhaps the doctrine of the gradualness of linguistic change
should be touched on here. It is a commonplace view that changes
begin on an infinitesimal scale, as nonrandom deviations from
some norm, and then grow in extent ("imperceptibly" is the fa
vorite adverb), until, somehow, the threshold is crossed. It is
just possible that this view is simply a carryover from prestruc
tural days. We would now prefer a picture in which changes are
made up of discrete steps, some very small, some (physically) zero
(namely, where a structural reinterpretation of the "same" phys
ical entity is the decisive event: allophones coming to stand in
contrast; allomorphs like' shade and shadow taking on contrasting
meanings, etc.). Since, if we are right on this, these discrete steps
depend on the amount of nondistinctive variation which exists in
the speech community in the first place, they are bound to remain
small. Perhaps they are even bound to affect only one distinctive
component at a time, as William Austin 11 acutely observed in the
special area of sound change. If we have a diachronic universal
here, it is one that flows from one of the synchronic universals
governing the typology of dialect areas.
40
Henry M. Hoelligswald
4. Target Structure
This brings us to the consideration of diachronic universals in

the sense that certain structures mayor may not be favorite targets
of change processes. Here, too, the typology of change is sub
ordinate to the typology of existing states; and our findings can
be no stronger than the findings on descriptive universals. We
first select two subjects for discussion: the alleged regularizing
action of "analogic change"; and the principle of widening
conditioning.
In classical historical linguistics sound change and so-called
analogic change (which must not be confused with the factor of
analogical creation present in all change processes and somehow
connected with the intimate machinery of those processes as
adumbrated here) are often regarded, and sometimes even ex
plicitly defined, as opposing principles. It is said that (conditioned)
sound change creates "irregular paradigms," that is, morpho
phonemes, whereas analogic change serves to eliminate morpho
phonemic ~lternation. This is a tribute to one admittedly typical
role assumed by those two forms of linguistic change; but it is
also an exceedingly obliq ue approach to analogic change. Analogic
change is essentially a replacement of one allomorph by another
within the morpheme; and it is that quite regardless of possible
morphophonemic consequences. As shoon, etc., has changed to
shoes, for instance, the allomorph -en has evidently receded one
step to the point where it is limited to oxen, and perhaps a few
other plurals; the allomorph z has been extended to one new
class of environments, namely into all texts where the plural
morpheme occurs after shoe. And it is not true that this dovetailing
reciprocal distributional movement within the morphemic unit,
will inevitably and exclusively favor one allomorph, or a selection
of allomorphs, over all the others, thereby minimizing irregular
alternation and "leveling" the paradigm. Analogic change has
been known to create new allomorphs (although perhaps not new
morphophonemes) and to extend irregular (grammatically con
ditioned) allomorphs at the expense of their regular, phonemically
conditioned competitors. English plurals in -sz -iz may generally
41
have the advantage over their competitors; but newly imported

names of fish, like muskellunge, are likely to receive the zero
alternant seen in trout, bass, andfish itself. Nor, on the other hand,
does sound change always increase alternation. Suppose that a
given sound change merges two phonemes which alternate. This
will reduce, and not increase, morphophonemic complexity. The
history of well-known language families is shot through with
examples of this sort. In other words, both the processes (sound
change and analogic fluctuation of aUomorphs) serve the same
structural purposes-to use a teleological turn of phrase which
is almost inevitable at the present state of our theoretical pene
tration. It is as though the structural goal were somehow given,
subordinating to itself all manners of available machinery.
And the goals, alas, impress us again and again as very specific
goals, characteristic perhaps of given areas and given broad
periods-they do not appear to be universals at all. I would
suspect that the older Indo-European languages, which represent
one of the best-studied areas and periods in the world, happened
to work toward a structural type characterized, among many
other features, by comparatively little allomorphic variety, and
that both ana logic change and a goodly fraction of phonemic mer
ger were merely pressed into that particular service, as it were.
There may be more to it, but we would be rash to be as sure of
it as Paul and Sturtevant had a right to be.
Greenberg and others feel that sound change has a typical
mechanism of successively widening scope. 12 Sound change, they
say, may begin as "sporadic," then become phonologically
conditioned, and finally unconditional. Before commenting on this
notion, let me strengthen it somewhat by what seems to me to be
an essential parallel. Once again the parallel comes from the
morphemic level (and once again, regrettably, from Indo
European I). One of the most sweeping trends in the history of
Indo-European languages, or possibly of the area to which the
Indo-European languages belong, is the merging of the two non
singular numbers, the dual and the plural, into one. The various
morphs which a componential analysis reveals to contain com
ponents for the dual morpheme or for the plural morpheme
42
become complementary in various ways and take on the combined

meaning of both (the category is customarily labeled "plural," as
though it were synonymous with the old, nondualic plural). But
not all such morphs are affected at once. In some conditions the
two components contrast far longer than in others. In analogy to
the parallel, if much simpler, process observed in phonemic
change, we should say that the conditioning of the merger widens
until it it becomes complete. [n some forms of Greek the dual in
the noun outlived the verbal dual. In classical Latin where the
same trend was obeyed earlier, both inflectional duals are gone,
but contrasting forms still exist for "which (of two)" versus "which
of more than two)" (uter versus qui). The modern English com
paratives, as well as the words either and neither, still contain a
contrastive dualic component (vis-a-~is the superlatives, and
any, none). 13 Some Indo-European languages have no remnant
left. There is no doubt that here we have an important principle.
To be sure, it operates in one direction only. It is not possible to
predict that all conditioned change will become unconditional, or
even (as we should really expect) go on to engender parallel chang
es, that is, acquire a wider and wider conditioning for the distinc
tive feature contained in the phoneme (or morpheme) in question.
But it is certainly appropriate to suspect that what is in the end
effect an unconditional or nearly unconditional merger has gone
through intermediate periods of such characteristics.
So long as we concentrate on sound changes, we have relatively
little difficulty with the so-called first step from alleged lexical
"sporadicity" to phonemic, but still narrow, conditioning. In
stances of so-called sporadic sound change can also be labeled
instances of dialect borrowing, and under favorable conditions
this is not necessarily inferential but demonstrable from a knOWl
edge of the dialect picture at the time of change. The step from
"sporadic" to regular change, where the changing unit is a "pho
neme in a given environment," that is, an allophone-that step
is, as we have tried to show, merely the step from ordinary
selective borrowing to the kind of total borrowing for which there
seem to exist typical favorable settings in history. The remainder
of the story is more difficult. Why should an allophone, once it
43
has left the fold, continue to attract its erstwhile fellows? 14 There
is no simple answer, but a few associations come to mind. It may
be argued that in addition to total dialect borrowing, some
apparent sound changes are in fact extreme manifestations of
analogic leveling. There are perfectly good theoretical criteria to
distinguish the two effects, but the evidence available does not
in all actual cases permit their application so that ambiguous
instances occur. Once a genuine conditioned sound change has
phonemically separated an allophone, or block of allophones,
from the rest, this may produce a morphophoneme. Further ana
logic generalizing is now possible, and, under conditions which
I will not here stop to define, such generalizing may place the
"new" component of the morphophoneme into positions which
make the resulting morphs appear as though more sound change
had taken place. A second association is more in the nature of a
priori speculation. It is possible that the initial amount of pho
nemic split which is triggered by a bit of dialect borrowing
(involving, say, a restricted and particularly vulnerable diaphone)
results in a structurally weak, unsymmetrical, poorly integrated
phonemic system, one in which the soft spots call for remedy.
Reversal of the initial process (which would not be recoverable
under most ordinary circumstances anyway) is a less likely pos
sibility since the particular area of distribution in which the smal1
amount of merger had occurred is now presumably the more stable
portion of the unbalanced subsystem-and Just for that reason,
we may expect the pressure to continue in the same direction with
somewhat increased vigor. Note that now we are again at the
mercy of synchronic typologies-possibly of their universals, but
more likely of their individual complexion. The very instances in
which the widening becomes arrested halfway are cases in point.
In several Algonquian languages vowels are lost in word-final
position, but not much in other positions. The result is a power
fully restricted canonical shape for words. III Just for this reason,
this is not the kind of sound change that we would expect to go
on widening until it engulfs nonfinal vowels, and most certainly
not to the point where these languages become vowelless
reportedly a rare, even precarious structural type.
[.~, ', ~l
II
i:",:,il
44
I,"
:"':1
, II
r
At this point the old question arises whether or not the syn
chronically observable structures (here regarded in their function
as targets of change) have themselves significantly changed in
the course of history. Of course, our information covers avery
small fraction of the history of the species, and within that
fraction is monstrously uneven. There is little in it to prove that
earlier ideas of "progress in language" (to quote a famous
phrase) are more than ethnocentric circularities, matters of fact
mistaken for matters of course. Those who point out that ad
vances in material culture are consistently "reflected" in all the
languages of the world-earlier or later, as the case may be-have
had to base their argument almost entirely on nonreplacement
changes, that is, on the amorphous additions and deletions
which are a delight to students of general history but do not really
alter the language. The more far-reaching claims are also more
doubtful. The trend from so-called synthetic to so-called analytic
structure may be observable in certain areas, but so is the opposite
trend, sometimes even in the same language family or area. It is
probably only because the Indo-European idea of progress (or
degeneration) from inflection to construction, from morphology
to syntax, from bound to free had become such a cliche, that the
equally typical notion of "grammaticalization"-the emptying of
lexically meaningful morphs (compound members, etc.) and their
transformation into "function" elements-was not presented as
a counteraction, although at least in a minor way it has served to
build up forms that look like new inflections (e.g., the Romance
adverbs in -mente, from mente 'with (such and such) a mind'; the
Osco-Umbrian locatives, with former enclitic adverbs intruding
into the case system; and so on). This is not saying that such a
presentation would have been particularly justifiable-only that
it would have been no worse than other popular attempts at in
terpreting the linguistic history of the species.
Greenberg, Osgood, and Saporta believe that, other things
being equal, "the more uncommon a phoneme is in human speech
in general, the more likely it is to be merged with another pho
neme." 16 What do they mean? Aside from the question of factual
support, one would like to know whether it is implied that there
45
are only processes which eliminate "uncommon phonemes," or

also others, not named, which give rise to new uncommon
phonemes. (The other generalizations offered in the same list
do not exclude the latter possibility.) If the former is meant, this
would attribute to the human race a consistent history of uni
formization, at least in point of sound structure. Would we not
then have to extrapolate back to a phylogenetic babbling stage,
and extrapolate forward to a radical reduction in the number of
existing areal types? P. Friedrich reminds me that such strikingly
aberrant traits as are revealed to the present-day typologist (say,
the southern African clicks?) could then ipso facto be recognized
as survivals-not in terms of ordinary micro history with its
mutually compensatory, plus-ra-change-plus-c' est-Ia-meme-chose
reshuffling, but as true fossils against a background of serious
mutation.
5. Morph Histories, etc.
We must now turn to this microhistory. Of course it is made
up of the same processes considered thus far: the replacements
with their various patterns. (See Casagrande, 10.3.) As far as we
can observe, these processes lead very rarely, if ever, to alterations
of fundamental importance for the species. They do somewhat
more frequently lead to structural changes which are typically
either episodes in an areal trend (as are many of the more system
atic transformations during the last two millennia in Indo
European or Semitic history), or, more flamboyantly, episodes
in the transfer of a language from one area to another, or from
a given position in one area (central, marginal) to another
position (marginal, central). As we have said, the typology. of
such structural changes is thus constrained by the typology of
possible or probable structures. The exchange which took place
between lakobson and Allen at the Eighth International Congress
of Linguists furnishes two interesting viewpoints on these aspects,
one sanguine and one carping.
In much ordinary language history the replacement processes
function as devices for the preservation rather than for the al
:1
oJ
46
Henry M. Hoelligswald
teration of the basic plan. They "fill gaps" left by other replace
ment processes or "restore a balance" which had been upset
temporarily-or even only potentially, since the separate for
mulation of illness and cure is not infrequently only a matter of
statement. Where these dramatic interpretations do not simply
depend on one out of many possible phonemicizations or morphe
micizations of the data, the "shifts," the "drag-chain," and "push
chain" displacements in phonemic systems, the "moves" attributed
to vocabulary items on the chessboards of semantic fields, can
often be shown to be concretely definable chronological realities
which elucidate the dynamics of linguistic history. In dealing
with them it is natural not to fix one's attention on the "emic"
structure points, but on the "etic" lower-level utterance stretches
that Occur first at one and then at the next structure point. We
have said at the start that it is in this realm that the early persis
tent and persuasive claims were made. We possess "general" clas
sifications of sound change (like Grammont's) where the criteria
are phonetic rather than concerned either with the replacement
pattern or with the resulting structure. We possess similar
classifications for semantic change by meaning content (or by
grammatical "function") where the morphs, as identified accord
ing to their phonemic shape (and not the morphemes defined
by their pattern of contrast!) are the heroes. These classifications
may sometimes be intended as mere conveniences. More often,
however, they will lay claim to some systematic or even predictive
power. This is quite clear in such works as Havers' manual of
"explicatory syntax" with its catalog of "conditions" and
"motive forces"; or Kurylowicz' theory of analogic change; or the
structuralist theories of change by Martinet and others in which
a great deal is said not only about preserved structures and
altered structures, but also about the movements of the particular
phones (and, where attention is given to grammar and lexicon,
morphs) whose shifting privileges of occurrence constitute in the
aggregate the systemic reality that matters in the end.
There is wide agreement, for example, to the effect that sound
changes are largely assimilatory changes. A given sequence of
sound segments is often replaced by an articulation which is in
47
some way less taxing. The same is true, as must be added, and as
Martinet has in fact added, of given combinations of distinctive
features occurring simultaneously. The assimilatory principle is
for that reason not as neatly tied to the superficial category of
"conditioned" (as against "spontaneous," that is, unconditional)
sound change as some take it to be. With more and more com
ponential analysis brought to bear, more and more presumably
unconditional changes turn into conditioned ones. The difference
is, of course, more important to those who advocate uniqueness
for phonemic solutions. Frequently the articulatory simplifica
tions produce "long components": sequences of consonants
between vowels acquire the voicing of the vowels; sequences of
back vowel syllables and front vowel syllables acquire an over-all
front vowel quality; there is loss and unvoicing at the! end of
utterance, thereby assimilating, totally or partially, to the adjacent
stretch of silence, and so on ad infinitum. I am not implying, by
bringing this well-known point into the discussion so late, that
assimilatory changes are not "utilized" for purposes of lasting
transformations as well as for microhistorical change; but I do
not think that anyone would seriously champion the idea that
constant assimilatory activity leads to more and more articulatory
simplicity in the languages of the world. Somehow, it seems, new
"difficulties" are always created. We also encounter the notion
that the assimilations constitute the speakers' (erosive) contri
bution to the flow of history; the hearer, striving for more
redundancy and indifferent to the speakers' tendency to ease of
articulation, keeps the extent of assimilation within bounds. It is
clear that without quantitative controls such a pair of governing
factors, located at the two extremes of a scale, has little to com
mend it: any sound change, by dint of its having occurred, will
be shown to have carried just the right amount of erosion and of
preservation. But it is possible that the numerical sophistication
at which information theory aims will help make these concepts
less trivial.
Kent and other scholars have observed that among the known
instances there is probably more anticipatory than progressive
(progressive = moving with, rather than against, the flow of
48
Henry M. Hoenigswaid
speech in time) or mutual (e.g., [sk) > [s)) assimilation. I? But

attempts at finding special offsetting features in the still numerous
cases of progressive assimilation have been disappointing, and
there is a grave po~sibility that the preponderance of anticipation
over "lag" is a specific areal feature characteristic of the more
thoroughly studied languages. In any case, not much joy can be
derived from a potential universal that presents itself in terms of
a rather modest statistical lopsidedness. To be sure, a further effort
was made, by Greenberg, to correlate the predilection for regres
sive assimilation with the reported (Sapir) synchronic predilection
for suffixing: prefixes tend to be swallowed up by the regressive
assimilations which originate in the core of the word. 18 It would
have to be shown, however, that suffixes are not equally threatened
by what is a notorious variety of regressive sound change, the
anticipation of post utterance silence through merger or loss of the
final portion of the word. It is quite true that lapses are also more
often anticipatory than otherwise in character; but the observa
tion of lapses is pretty much limited to language structures of one
and the same type. Typologies in which word-initials rather than
word-finals are morphophonemically variable or otherwise weak
have not, to my knowledge, been included; and, in any event, the
relationship between individual lapsing and sound change is
tenuous.
It has been proposed, as a structural refinement of the doctrine
of the assimilatory character of most sound changes, that the
distinctive status of an articulatory quality must be taken into
account when judging the probability of its functioning as a con
ditioning factor. Martinet has thought that "a voiced environment
in which voice has no distinctive value does not under ... ordinary
conditions have the effect of voicing a voiceless sound and
thereby neutralizing the opposition of voice"; 19 but the very
ordinary appearance of many a contrary case (e.g., nt > nd in
Late Greek) puts a heavy burden on the term "ordinary con
dition." One may well feel that more evidence is needed. But in the
absence of a satisfactory theory, more evidence-which can at
best have statistical characteristics-is not going to create over
much confidence. In this context we ought also to remember that
49
there is room for nonassimilatory processes in linguistic history,

too. I am not so much thinking of dissimilations which although,
they are much more part of "regular sound changes" than is often
asserted, may yet have special distinctive traits that make them
different in kind rather than opposite in degree from the more
usual processes. 20 But hyper/orms are very much part and parcel
of ordinary language history, and it is the essence of hyperforms
to produce effects which run counter to phonetic plausibility.
Thus, Italian dimestieo from Latin domesticum presupposes an
assimilatory change from im to om, with lip rounding as a "long
component"-the apparent dissimilation from 0 to i is probably
the result of dialect borrowing after the event. As usual, we feel
much surer about our knowledge of the replacement histories than
about such generalizations as we care to make with regard to the
behavior of phones in phonemic change.
With semantic change the same problems recur in an even more
drastic form. We have seen earlier how the question of "semantic
laws" (that is, of the regularity of semantic change) was vitiated
by the wrong parallel. If we formulate semantic changes from the
morphemic point of view, they are as regular as the sound changes
which indeed we do customarily state in phonemic and not in
phonetic language-:-<>nly we are disappointed, in the case of
semantic change, by the relative emptiness of the result. (One of
the reasons for this is that there is so much one-to-one replace
ment, without merger or split, in the lexical and grammatical
field.) When we consider morphs instead of morphemes, we start
having difficulties with our generalizations. Thus, it will be very
hard, if not impossible, to discover any kind of predictability for
the movements of individual morphs across the table of mor
phemes. For the very special cases of analogic change, Kurylo
wicz' concept of polarization, favoring the better characterized
of the two competing allomorphs, is perhaps the most promising
attempt. There is, I hope, no need to dwell on the weaknesses of
the content-oriented classifications for semantic change; it may
suffice to name metaphor as perhaps the most popular quasi
universal. There is no doubt that many languages have put
portions of their vocabulary through change processes which are
50
similar in essential ways from language to language and for which

the term metaphor is a fair label. But the probability that a given
morph will be used "metaphorically" is a probability of the same
kind (though not necessarily of the same degree) as that for a given
phone to undergo a certain sound change. In a sense the chief
governing factor is found to be, once again, the synchronic
typology. We must not forget that the major content-labels for
changes of meaning are at the same time useful in synchronic
semantics. Metaphor, in fact, was a rhetorical figure long before
it became a term for a class of semantic changes.
We are not surprised, therefore, at the failure to find semantic
laws for morphs and their meanings in history. The search for
such laws has not been fruitless-for example, Stern's work,
particularly his example: "English adverbs which have acquired
the sense 'rapidly' before 1300 always develop the sense 'im
mediately'. This happens when the adverb is used to qualify a
verb, etc." 21 Note especially to what extent the "regularity" of
this law rests on the semantic (Le., the morphemic, and not the
morphic) identification of the entities affected. But if we are to
learn more about the general properties of semantic change, some
of the all-too-familiar concepts will have to be abandoned.
I"
6. Conclusion
What is needed in the study of diachronic universals is, of
course, more widely based historical and comparative work than
was available for many decades; and a great deal of such work is
now being done. Furthermore, it must be understood that change
has been seen, for many generations of scholarship, in three
different ways: as a pattern of (superficial addition or deletions,
and of) replacement; as a process culminating in the production
of new structures; and as the Saussurian "chessboard" movement,
often to no lasting avail, of the counters as they are recognized by
their acoustic, articulatory, or otherwise measurable character
istics. Our interest naturally centers on the second and on the
third conception of change more than on the first. It is possible
that new approaches like transformational grammar which
!~
"
51
promise to unify synchronic typology in a hitherto unsuspected

sense may also bring new principles of importance to an under
standing of the universals of change.
ACKNOWLEDGMENTS
In addition to helpful comments by Charles F. Hockett, to be
mentioned in the notes, I would like to thank Joseph H. Greenberg,
Einar Haugen, Roman Jakobson, and Rulon S. Wells, who have put
me in their debt by suggesting changes.
Notes
1. Schwyzer, Historical Greek Grammar 1.234ff.
2. See S. C. Gudschinsky, Word 12.175ff. On Charmey, see Gottin
gen Nachrichten, Ph.-Hist. Klasse 1929, 195-214.
3. Thanks are due to Charles F. Hockett for his searching comments
on this and other points. While none has been ignored, it is impos
sible here to discuss, adopt, or try to refute all of his comments
in full. He takes exception to the notion of quasi-translation as
used here, emphasizing that language is a set of habits and not a
collection of "texts." But "is" is a difficult word. We are only,
after all, trying to analyze the historian's approach, which has
essentially been one of matching ("comparing") parts of two or
more bodies of utterance; it is not likely that that approach
should be irrelevant for an understanding of the habits in question,
even granting that habits and texts are things apart. Hockett
further objects that that kinds of change ought to be rigorously
distinguished from mechanisms of change in a manner which,
if I understand him correctly, goes somewhat beyond the position
taken in his Course (ch. 52 and elsewhere). This issue needs a full
debate.
4. Namely, in economy (Hockett).
5. For the sake of the argument, these words are treated as though
they had one morpheme each. Actually, in-, -wit, con- may be
said to take part in conditioned replacement processes.
6. "The first and foremost," Hockett suggests, " ... is that replacement
occurs. This is not trivial. There are communicative systems
in which it does not, or in which it takes place by mechanisms so
totally different ... that the difference is worthy of note."
52
,I
7. The latter insofar as their relationship to the morphemes are

concerned. The morphs themselves are to be thought of as pho
neme stretches.
8. Sound changes can apparently not be entirely predicted from
internal, systemic stresses and strains, nor can they be explained
as the effect of scatter around a target or norm; they have direction
and are in that sense specific, much like other happenings in
history. The factors which make for prestige differentials, for the
development of regional centers, for demographic change, etc.,
are specific and unique (i.e., not redeemed by the greater simplicity
of language structure as contrasted with the often-invoked
.
"world at large") in the sense required.
9. Language Change and Linguistic Reconstruction, 38 and 75f.
10. Indogermanische Forschungen 45. 105ff.
11. Language 33. 538ff.
12. Psycholinguistics, 148.
13. Fred W. Householder's watchfulness has saved this sentence from
serious misformulation.
14. See Joos, Readings in Linguistics, 376.
15. Restricted insofar as vowels are excluded. On the other hand,
after the change, "words could end in any of a much larger number
of consonants and consonant clusters"-larger, that is, than that
of the four vowels (Hockett).
16. Psycho linguistics, 148.
17. Language 12. 245ff.
18. Essays in Linguistics, ch. 8.
19. Economie des changements phonetiques, 111.
20. The doctrine of the special unruliness of dissimilations, metatheses,
etc., occupies a peculiar position in the history of linguistics.
But, in fact, even the dissimilations of liquids familiar from
Indo-European and Semitic are much less untypical than they are
said to be. They are only in part progressive; their "regularity" is
quite considerable (often the counterexamples occur in paradigms
only), and even if it were true that graduality is a requirement
of ordinary regular sound change, quite a few dissimilatory
processes (e.g., the one known as Grassmann's law) may easily
be thought of as taking place gradually, while some nondissimil
atory and nonmetathetic changes would seem to preclude such
a picture. See Hoenigswald, Phonetica 11. 202-15.
21. Meaning and Change of Meaning, 190.
CHAPTER 3
ASSUMPTIONS ABOUT NASALS: A SAMPLE

STUDY IN PHONOLOGICAL UNIVERSALS
CHARLES A. FERGUSON
Center for Applied Linguistics
1. Introduction
Although linguists hesitate to make statements of universal

(panchronic, cross-language) validity about the details of phono
logical structures, they often operate either in their own field
work or in their evaluation of others' descriptions as though they
held certain assumptions of this kind. Certain common features
of the sound systems of human languages are so fundamental, of
course, that linguists would exclude from the label "language"
a signaling system that lacked them. Such universals may be
regarded as definitional; that is, they are implicit in the linguist'S
concept of language, whether included in his formal definitions
or not. For example, the linguist would find it inconceivable that
a language should operate without phonemic contrasts, without
a small set of distinctive features (or phonetic and distributional
classes) in terms of which phonological elements of a segmental
sort could be identified, or without differences in frequency of
occurrence of such phonological elements.
2. Value of Nondefinitional Assumptions
The present paper is an attempt to formulate several non

definitional assumptions which the author holds in one section
of phonology. Most of them are probably shared with many other
53
54
:I
Charles A. Fergusol1
linguists; some may be of little validity. No attempt will be made

to provide a theoretical framework for the assumptions, but the
formulation of a set of statements like this may prove of value
in at least three ways.
First of all there is the advantage gained in any field of science
from making unspoken assumptions explicit. This process may
reveal mistaken or mutually inconsistent assumptions, or may
give new insights into the theory of the particular science.
Second, there is the value which universal, nondefinitional
statements have for linguistic typology. For example, it is a
widely held assumption that a language never has a greater
number of phonemic contrasts in the vowels of unstressed syllables
than it has in stressed syllables. 1 Any attempt to formulate this
assumption carefully and to investigate its validity empirically
soon shows that this point classifies languages with distinctive
stress into three main types: (I) Languages, typified by Spanish,
in which stress has no important effect on the quality of vowels.
In languages of this kind, apart from accidental gaps in distri
bution, there is only one system of vowels appearing in both
stressed and unstressed syllables. (2) Languages, like English
and Russian, in which stress has the effect of lengthening a vowel
and enhancing its characteristic coloration. In such languages
the stressed vowels are clearer, and there are in effect two vowel
systems, one for stressed syllables and one of fewer contrasts
for unstressed syllables, always with some slight tendency for
analogical formations to create unstressed vowels from the
stressed system (e.g . Russian lei) or to transfer a neutral vowel
of the unstressed system to the stressed system (e.g., English IV).
(3) Languages, like certain Tajik dialects 2 and certain varieties
of Syrian Arabic, in which stress has a leveling effect on vowels.
In such languages the stressed vowels are not greatly lengthened
and are less clear than the unstressed vowels. These languages
usually have a greater number of vowel contrasts in unstressed
position. 3
The third value of the formulation of phonological universals
lies in the materials it provides for extralinguistic treatment.
A nondefinitional universal in linguistics may serve either as
ASSUMPTIONS ABOUT NASALS
55
an exemplification of principles of some other field of knowledge

or as a suggestion toward reformulation of such principles. For
example, it is commonly assumed that extensive voiced-voiceless
neutralization in a language takes place most commonly in final
position and never intervocalically. 4 Or again, in diachronic
studies it is generally assumed that a phoneme of [s] type may
change to one of [h] type but not vice versa, or of the [k] type to
[t] but not vice versa, and so on. Linguistic statements like these
suggest interpretation in physiological or psychological terms.
An experienced linguist working in the field of phonology
probably operates with many assumptions which could be iden
tified and formulated. This may be shown by the ease with which
one can construct an artificial phonemic system which would
seem to be perfectly adequate for communication purposes but
which the practicing linguist would regard as implausible. 5
3. Phonological Assumptions -
Nasal Phenomena
A full list of phonological assumptions could run well over a

hundred. The list of fifteen statements that follows (identified by
Roman numerals) is limited to nasal phenomena and is offered as
a sample. 6 Three of the statements are diachronic, two are syn
chronic frequency statements, and the remainder are synchronic
existence statements. The statements are generally explained in
all-or-none terms, although most are probably only statistically
valid; that is, the probability of "exceptions" is very low, and a
language showing an exception may be regarded in some sense
as abnormal or pathological.
Nasal phonemes are of four general types, which are called
here primary nasal consonants, secondary nasal consonants,
nasal vowels, and nasal syllabics. These types will be defined, and
universal statements will be listed where most appropriate under
each type. Two kinds of nasal phenomena have been excluded
from the listing: (I) nasal or nasalized allophones or phonemes
the most characteristic allophones of which are nonnasal, and (2)
prosodic features of nasality. The first are felt to be outside the
system of nasal phonemes and not covered by the kinds of univer
56
Charles A. Ferguson
sal statements given here. The second are usually analyzable

alternatively in terms of segmental phonemes, and in that case
the universal statements made here are held to be valid.
3.1. Primary nasal consonants (PNC)
~.;i
I;jfl
:~ ~J
f~
;J
H
I'
Definition: A PNC is a phoneme of which the most charac

teristic allophone is a voiced nasal stop, that is, a sound produced
by a complete oral stoppage (e.g., apical, labial), velic opening,
and vibration of the vocal cords.
When, in a given language, there are no nasal phonemes of
other types with ranges of phonetic values which might conflict
with the PNC's, a PNC may have allophones without full oral
stoppage, with incomplete velic closure, or without voicing. Even
such a language, however, a PNC in phonological positions or
communication situations calling for maximum clarity will have
the normal voiced nasal quality. In some languages the PNC's
function is in part like that of the vowel phonemes of the language;
for example, it may constitute syllable peaks or bear accents, but
this is always in addition to consonantal function.
I. Every language has at least one PNC in its inventory. (Com

plete absence of nasals is reported for three Salishan languages 1
where the PNC's assumed for an earlier period are said to have
become voiced stops.)
II. If in a given language there is only one PNC, it is In/, that is,
its most characteristic allophone is apical. When there is no other
nasal phoneme in the language with a range of phonetic values
which might conflict with the In/, the /n/ may have labial, velar,
or other allophones, but in positions or situations of maximum
clarity it has the normal apical value. In the rare instances where
a language has only /m/, there seems always to be an apical [n] as
an allophone of something else. (In Hockett's analysis of Winne
bago [nJ is /rl plus nasality; 8 in Ladefoged's analysis of Yoruba
[n] is III next to nasal vowels. 9 Examples of languages with In/ as
the only PNC are chiefly in the Western Hemisphere, for example,
Tlingit, a number of Iroquoian languages, Arapaho.
57
III. If in a given language there are only two PNC's, the other
one is Iml, that is, its most characteristic allophone is labial.
Languages with 1m nl are extremely common, including exam
ples from Indo-European, Semitic, American Indian (various
families), Altaic, Caucasic.
IV. In a given language, the number of PNC's is never greater
than the number of series of obstruents. For example, if the
language has stops and affricates in four positions (e.g., Ip t c k/)
the number ofPNC's will be four or fewer (e.g., I m n il 't)/, 1m n't)/,
n ill, or 1m nf), never five or more. A number of different
arrangements are possible, for example:
Bengali
pqck
Nuer
Fiji
ml}nilu
tk
t c k
French
m n lJ
p t k
mnil
~8
mnU
V. When in a given language there is extensive neutralization
among the PNC's, this occurs in prejunctural and/or precon
sonantal positions. (Examples include Spanish, Classical Greek;
Trubetzkoy cites a number of others. 10)
3.2. Secondary nasal consonants (SNC)
Definition: An SNC is a nasal consonant phoneme the most
characteristic allophone of which is not a simple voiced nasal.
In many cases a phone type which may be analyzed as an SNC
may alternatively be analyzed as a cluster (e.g., Ihnl, Imb/). The
statements made here refer to languages where the monophone
matic analysis is required either because of contrast with clusters
or because of striking parallels of distribution. At least six sub
types occur:
voiceless nasals
aspirated nasals
glottalized nasals
pahitalized nasals
(e.g.,
(e.g.,
(e.g.,
(e.g.,
Kuanyama 11)
Marathj12)
Chontal (Oaxaca) 13)
Russian)
58
Charles A. Ferguson
"emphatic" nasals
prenasalized (voiced)
stops
"nasalized clicks"
(e.g., Syrian Arabic l4)

(e.g., Fiji 15)
(e.g., Zulu 16)
VI. No language has SNC's unless it also has one or more

PNC's. (Corollary to I.)
VII. In a given language the number of SNC's is never greater
than the number of PNC's.
VHI. In a given language the frequency of occurrence of SNC's
is always less than that of PNC's.
IX. SNC's are, apart from borrowing and analogical forma
tions, always the result of diachronic developments from clus
ters. (This assumption is based on the very few cases when the
history of SNC's is well known. It is quite probable that other
sources exist; in particular, it seems likely that prenasalized stops
have developed from voiced stops in certain languages.)
3.3. Nasal vowels (NV)
Definition: An NV is a phoneme the most characteristic
allophone of which has oral and velic opening and vibration of
the vocal cords. When in a given language there are no phonemes
with conflicting phonetic values, an NV may have allophones with
oral closure, velic closure, or lack of voicing, but clarity positions
and situations have normal nasal vowels. (Sample languages:
French, Bengali, Taos.)
X. No language has NV's unless it also has one or more PNC's.
(Corollary to 1.)
XL In a given language the number of NV's is never greater
than the number of non nasal vowel phonemes.
XII. In a given language the frequency of occurrence of NV's
is always less than that of nonnasal vowels. (Reliable frequency
counts of phonemes exist for very few languages with nasal
vowels. One small count for Bengali 17 shows an oral-nasal vowel
ratio of 50: I.)
59
XIII. When in a given language there is extensive neutraliza

tion of NV's with oral vowels, this occurs next to nasal conso
nants. (Two well-documented examples of this kind of neutrali
zation arc Bengali 18 and Yoruba. 111)
XIV. NV's, apart from borrowing and analogical formations,
always result from loss of a PNC. (This assumption is based on
a small number of languages where the history of the NV's is
known. These are chiefly Indo-European-[Indic, Slavic, Ro
mance]-and it is possible that this assumption will have to be
modified when more is found out about the history of NV's in
other families. One case where an NV may be of quite different
origin is in Iroquoian, where one of the NV's posited for the
protolanguage seems, on considerations of internal reconstruction,
to have derived from earlier lal + IiI or sequences like lawaI)
3.4. Nasal syllabics
Definition: A nasal syllabic is a nasal phoneme which
patterns like a syllable rather than like a consonant or vowel in
the language (e.g., Japanese Inl, 20 Ewe, Xhosa Im/21).
XV. A nasal syllabic phoneme, apart from borrowings and
analogical formations, always results from loss of a vowel.
Notes
1. Cf. Trubetzkoy, Principes de phon%gie, 255-256 (Paris, 1949).
2. Cf. Sokolova, Fonetika tadzikskogo jGzyka, 19 (Moscow, 1949).
3. It is interesting to note that in the Lebanese Arabic spoken in
Marjayoun, where the i-u contrast which has disappeared in
stressed position in much of Lebanese Arabic is still present, the

phonetic distance between IiI and luI is less than it is for the i-II
contrast in unstressed syllables.
4. A fully satisfactory formulation of this would be considerably more
complicated and would have to take into account the very rare
instances of initial neutralization (as asserted by Trubetzkoy
for one variety of Mordvinian) and initial-and-final neutralization
(asserted for Kirghiz). Cf. Trubetzkoy, op. cit., 254-255.
,
60
Charles A. Ferguson
"
5. The author has on a number of occasions reported to fellow
linguists a system summarized in the following list of phoneme
symbols: c k k W v s Z 6 m' n' fl' IJ' I A Y ? ii ii re reo AA.
"i 3 U. The reaction runs from mild surprise to disbelief.
6. Two brief treatments of nasal universals are known to the .
Trubetzkoy, op. cit., 189-196, and Hockett, Manual of Phonology,
119-120 (Baltimore, 1955).
7. Cf. Hockett, op. cit., 119.
8. Cf. ibid., 80-81. Even in the absence of further evidence Hockett's
analysis seems less convincing than a more traditional one of Inl
and Irl as separate phonemes; other treatments or Winnebago
recognize Iml, In/. and Ir/. with either an additional In 2/ or certain
morphophonemic interchange between Inl and frio
9. Cf. Ladefoged, A Phonetic Study of West African Languages, 23-24
(Cambridge, 1964).
10. cr. Trubetzkoy, op. cit., 193.
11. Cf. Westermann and Ward, Practical Phonetics, 65 (London, 1957).
12. Cf. Lambert, Introduction to the Devanagari Script (London, 1953).
13. Cf. lJAL 16:35 (1950).
14. Cf. Language 30:566-567 (1954).
15. Cf. Hockett, op. cit., 124.
16. Cf. Doke, The Phonetics of the Zulu Language (Witwatersrand,
1926).
17. Cf. Language 36:51 (1960).
18. Cf. Language 36 :44 (1960).
19. Cf. Ward, An Introduction to the Yoruba Language, 13 (Cambridge,
1952).
20. Cf. Language 26:112 (1950).
21. Cf. Doke, The Southern Bantu Languages, 92 (London, 1954).
CHAPTER 4
PHONEME DISTRIBlITION AND LANGUAGE
UNIVERSALS
SOL SA PORTA
University of Washington
1. Language Universals Closely Related to Language Typology

It is clear that the problem of language universals and that of
linguistic typology are closely related. Indeed, the two are mere
ly different sides of the same coin. The typological statement
that there are languages with a feature X and languages without
X is only as meaningful as the identification and definition of X
in the abstract, that is, independent of any language. Thus, to
say, for example, that all (or many) languages have 'a contrast
between nasal and oral consonants' presupposes a universal def
inition, for terms like nasal and oral. Universal definitions are,
then, a prerequisite for typology, universal in this sense meaning
'universally available,' that is, belonging to some metatheory of
linguistics. 1 Furthermore, a similar statement to the effect that
languages with feature X also have feature Y is a statement about
language universals which may be viewed as a special kind of
datum resulting from a typological analysis. Thus, typological
classification is a prerequisite for a statement of language uni
versals, where now universal means 'universally present.'
There have been two types of objections to work in language
typology. The first resolves itself into a lack of criteria for
evaluating alternative classifications. The number of possible
classifications for any number of languages soon gets large; for
the languages of the world the possibilities are unmanageable.
61
62
Sol Sapor/a
However, the conclusion that consequently any choice is arbitrary

seems unnecessarily pessimistic. The fact that one can always find
one feature to justify grouping two languages together does not
imply that all classifications are equally useful. We view a
typology as a device which yields predictions of this type: Given
two languages A and B which are assigned to one type by virtue
of sharing a certain feature X, and given the occurrence of another
feature Y in A, we predict its occurrence in B. Note that this is
nothing more or less than the positing ofa special kind of universal
of the type X _ Y; 2 and, in fact, the most useful typology is
exactly the one which provides a maximum of such universals.
The second objection is based on disagreement as to what terms
properly constitute a part of the metatheory, since, as we have
suggested, some universal definitions are a prerequisite for
typology. Now there are two types of terms in the metatheory:
those of substance and those of form. The first are terms like nasal,
stop, length, etc., or on the level of meaning, like past, dual,
animate, etc., which are defined in either phonetic or semantic
terms. The formal terms, however, do not require any such
'external' reference. Thus, a term like infix is presumably a formal
term in the metatheory, that is: (1) universally available-defin
able without reference to a particular language-and (2) formal
definable without reference to phonetics or semantics. The ques
tion arises as to whether a term like noun, on the other hand, is
a formal term in the metatheory. Most definitions of noun are
either semantics-oriented (the 'person, place, or thing' type of defi
nition), or specific-language-oriented (the 'noun-in-English' type of
definition). 3 It is precisely this fact that has hampered work in ty
pological classification and language universals on the grammati
cal level. The semantic definition seems inappropriate, or at least
undesirably imprecise, in a discipline which claims to be formal
and rigorous; the specific-language-based definition makes any
cross-language identification on formal terms impossible or, at
best, arbitrary. Similarly, the statement that two languages both
have /p/ has little meaning. The statement that they both have
phonemes which are distinctively voiceless, bilabial, and stops is
useful because the last three terms can be given physical (articu-
63
PHONEME DISTRIBUTION AND UNIVERSALS
latory or acoustic) definitions with considerable precision. In

other words, the fact that phonological typologies have proved
more feasible than grammatical ones is in part due to the ad
vanced status of theories of universal phonetics as compared to
semantics. In any case, an important prerequisite for a precise in
vestigation of language typology and language universals is spe
cifying the terms of the metatheory and the bases for their inclu
sion, that is, whether their definitions are purely formal or not.
So far, then, we have proposed three two-way distinctions:
universal/specific, form/substance, and phonology/grammar. We
can illustrate the possibilities as follows, filling in those boxes
for which we have cited examples:
PHONOLOGY
Substance
Substance
Form
infix
Universal
nasal
Specific
Form
GRAMMAR
animate
noun
In reference to the figure, there is no need to subdivide the columns

headed Substance, since phonetic and semantic terms are by
definition universals and not specific. However, any such term
may theoretically reappear as a formal term in a specific language
and, rarely perhaps, as a formal term in the metatheory. Thus the
term vowel (or vocoid) defined in terms of the presence of a certain
acoustic or articulatory feature may yield the same class of
phenomena, either for a specific language or for all languages,
as the definition of vowels as 'the minimum repertory of phonemes
such that at least one of these phonemes must occur in every
word.' 4 We are familiar with the case where semantically defined
terms in the metatheory correspond to formally defined classes in
a specific langauge, since it is these terms which yield the familiar
grammatical categories, such as the contrasts animate/inanimate or
singular/plural.
But there is a group of terms, such as endocentric construction,
or immediate constituent, which do not readily fit into such a
64
Sol Saporta
classification as the one proposed here. These terms have usually

been identified with the grammatical aspect of language, but this
restriction is probably unwarranted, since they seem rather to
apply to certain types of rules in a description. A language 'has'
endocentric constructions if it has rules of the form:
x=
A (B) or X
(A) B
But A and B may represent phonological terms, say, consonant

and vowel, as well as grammatical terms, like adjective and noun.
In short, certain terms are universally available because they refer
to aspects of the models of description. 5
Now, some but not all of the features which are universally
available are also universally present. It seems necessary to
distinguish, however, those which are present by definition (uni
versally necessary) from those which are not, though the basis
for such a distinction is vague. Though every language has a
contrast between consonants and vowels, presumably we would
not use the absence of such a contrast as sufficient evidence for
excluding a particular system of communication from being a
language. On the other hand, a system of vocal signs all of which
were onomatopoeic would be suspect, violating the requirements
of arbitrariness which characterizes language by definition.
We now ask the following sets of questions about the terms
phoneme, syllable, and phonological word: (la) Are they terms in
the meta theory; that is, can we define phoneme, or can we only
define phoneme-in-L, where L is some language? (lb) If they are
in the metatheory, we ask: Are they formal terms, that is, definable
without reference to phonetics? (2a) If they are in the metatheory,
that is, universally available, we ask: Are they universally present?
(2b) If they are universally present, we ask whether they are
present by definition, that is, universally necessary, or not. To say
that they are necessary will mean that they are universally
present no matter what model of description is used.
The investigation of these questions presupposes some frame
work within which the nature of phonological statements can be
made explicit.
65
2. Grammar as a Generative Theory

Adopted here is the view, formulated most clearly by Chom
sky, 6 that a grammar is a theory which ideally generates all and
only the grammatical sequences ofa particular language. A phono
logical description or generative phonology describes all and only
the phonologically acceptable sequences, including the as yet
unobserved ones, such as English fet but not fte. The adequacy
of the description is determined in part by the accuracy of its
predictions about the acceptability of unobserved sequences.
Chomsky has demonstrated that three models underlie genera
tive grammars or, phrased differently, that there are three different
kinds of rules: a Markov-process or finite-state rule, an immediate
constituent or phrase-structure rule, and a transformation rule.
Now, we are concerned here with two notions. The first is the
output of such a grammar, that is, the sequences to be generated.
The ideal end-product of a phonological description is the smallest
unit X about which it can be said that all combinations of phono
logically grammatical X's are themselves phonologically gram
matical. This unit may be called a phonological word. Any lan
guage which has a phonology has a phonological description, and
any description has an output, so that this unit is present in all
languages by definition, that is, universally necessary.
Now the acceptable sequences, to the extent that they constitute
a finite set, can always be listed. But we are presumably unwilling
to accept such a list as the most adequate description for any
natural language. That is, we insist that all languages have words
which are partially similar in sound, in the form of minimal pairs
or rhyme, alliteration, or something of the sort. Consequently any
adequate phonology will have to describe the acceptable sequences
as combinations of recurring elements or primes. During the
Conference, I suggested that, for the purpose of stating permissible
phonological sequences, the inventory of these recurring elements
would correspond closely enough to what we usually call
phonemes to warrant applying that term. 7 I did not make clear
enough, however, that for stating and comparing the phonological
inventory of languages independent of the distribution, features
66
Sol Saporta
are more likely to provide a useful basis; typological statements

about inventories are inevitably phrased in phonetic terms. In
any case, the present argument attempts to demonstrate only that
any phonological description must have two kinds of units, one
to be the output and the other the elements which combine to
form this output.
Now, one difference between a finite-state grammar, on the one
hand, and a phrase-structure or transformation grammar, on the
other, is the introduction in the latter of intermediate levels be
tween the output and the primes, levels like onset, coda, and
syllable, etc. Specifically, syllable is to phonology what phrase is
to grammar, namely a unit intermediate between the output and
the primes. Thus, phoneme, syllable, and phonological word are all
formal terms in the metatheory, that is, universally available. But
it is, in theory, possible to have a language in which all words are
exactly one syllable long, in which case the distinction between
phonological word and syllable disappears. Similarly, if there were
a language in which all syllables were exactly one phoneme long,
the distinction between syllable and phoneme would disappear
(and, incidentally, so would the distinction between vowel and
consonant). 8 Phoneme and phonological word are, therefore,
universally present by definition, that is, universally necessary.
Syllable, if it is universally present, is so by empirical observation.
In other words, to say that a language 'has' syllables is to say that
in stating the distribution of phonemes an immediate-constituent
model will prove simpler than a finite-state model.
Distributional statements are often phrased in phonetic terms.
Elsewhere, 9 we have suggested a universal based on such pho
netic statements, namely that the frequency of any C1 C2 would
be a function of the difference between C1 and C2 with both
extreme similarity and extreme difference tending to be avoided.
It is the main point of this discussion that phonological typologies
have rarely explored purely formal bases for classification. I do
not necessarily imply that these are more fruitful, only that they
are different and that it is appropriate that they be investigated.
Hereafter, universal, meaning 'universally present,' is used in the
broadest sense to include all phenomena of other-than-chance
67
frequency. Ideally, merely a cataloguing of such frequencies

would be inadequate without some rationale leading to the
identification of what the other-than-chance factors were, the latter
often being determined by nonlinguistic considerations. Thus, for
example, the frequency of diphthongs in Spanish 10 is a function
of stress, so that diphthongs are more likely to occur in stressed
syllables than in unstressed. To what extent this applies in other
languages is an open question, but some physiological analysis of
the relationship between the physical bases of stress and the
duration of a vocalic nucleus would presumably provide a basis
for suspecting that such a situation is the rule rather than the
exception. In short, the most interesting hypotheses are those
which relate regularities in linguistic data to regularities elsewhere.
During the Conference this question came up several times in
connection with the quotation from Bloomfield that 'useful gener
alizations ... are inductive.' 11 Some of the participants felt,
however, that the most suggestive hypotheses about universals
are those which follow from our definition of language as human
behavior and as being subject to laws governing such behavior.
Indeed, the question of how many languages constitute an ade
quate sample for a generalization is not a linguistic problem at all,
but rather a statistical one.
3. Hypotheses Illustrating Possibilities of Generalization in the
Area of Phoneme Sequences
Several tentative hypotheses are presented here. Underlying

them may be a general principle of economy whereby the presence
of the complex pattern implies the presence of the more simple
one. These hypotheses are presented purely by way of illustration
of the possibilities of generalization in the area of phoneme
sequences.
HypothesiS 1. In languages with both dissolvable and non
dissolvable medial clusters, the former will be significantly
more frequent than the latter. (A dissolvable cluster is defined
as a sequence whose first part occurs in final position and
whose second part occurs in initial position.)
68
:,:i
"I
'I'.'.1
11
,t
'I
'I
Sol Saporta
Hypothesis 2. In languages with phonemic stress, the num

ber of phonemic contrasts in stressed syllables will be greater
than or equal to the number of phonemic contrasts in un
stressed syllables.
Hypothesis 3. The presence of C1 C 2- makes -C2 C 1 as likely
as or more likely than -C1 C2
Hypothesis 4. For languages with phonemic stress, the
presence of n unstressed syllables flanking a stressed syllable
implies the presence of n- I unstressed sylIables.
Now, there may indeed be languages where Hypothesis 4 is
violated and the limitation on this hypothesis may serve to for
mulate the relation between universals and typology in a different
way. Given a general principle such as the one about a more
complex ~ a less complex, which in spite of wide application is
not, strictly speaking, a universal, it seems reasonable to suggest
. that precisely this limitation would provide the basis for a fruitful
typological classification. In other words, languages which violate
Hypothesis 4 form a type. We posit that these languages will tend
to share other features as well, and furthermore that it is just such
a clustering of features which characterizes a useful typology. In
other words, a good candidate for a diagnostic feature upon which
to base a classification is any case where there seems to be a
limitation on some general principle which can be established on
(physio- or psycho-) logical grounds, the assumption being that
deviations in such cases are more likely to affect other aspects of
linguistic structure.
It is of some importance to note that the degree of error in
typological and universal statements can be no less than the ex
tent to which we permit truly alternative, that is, nonconvertible,
solutions. 12 Furthermore, it is not enough, for example, that the
unit-phoneme versus cluster analyses of affricates be convertible
in some sense. They must, in fact, be converted before any precise
statements can be made which relate to numbers of phonemes,
types and frequency of clusters, etc. However, once the form of
a phonological description is agreed on, evaluation criteria become
69
feasible, and we can eliminate some portion of the nonunique

solutions.
Besides the purely phonological types of relationships explored
earlier, there are those between phonology and morphology
which deserved to be investigated. For example, consider the
following data on the length of morphs in phonemes in Spanish: 13
Number of Phonemes
o
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Number of Different Morphs
6
59
97
307
387
327
261
143
64
19
4
1
2
1
1
1679
The sample is a list of morphs, and it is by no means clear
to what extent the results are generalizable to texts. U Never
theless, there is a regularity in the fact that the distribution yields
a one-peak curve with the mean (4.4) near the peak. It is not
clear that such a distribution would result by chance alone and,
if not, what factor other than chance might be operating to
produce such a distribution. Nevertheless, one cannot help won
dering whether or not such a distribution is universal and, if not,
what other factors correlate with different distributions.
Another relation which suggests itself is that between the length
of morphs and number of phonemes in the inventory.
70
Sol Saporta
Hypothesis 5. The mean length of morphs will be inversely
related to the number of phonemes in the inventory. 15
4. Conclusion
One last thought: There is no reason to include only rules of
dependency in a statement of language universals. It is just as
interesting to point out factors which are independent. We will
not be surprised to find that questions of phonemic inventory have
no relation to the presence or absence of certain grammatical
categories, say, that the presence of a voiced/voiceless distinc
tion in stops is independent of the presence of a category of per
son in verbs. This is merely a trivial example of the general rule
of the arbitrariness of the linguistic sign, that is, the independence
of sound and sense. However, it might be of some interest to find
that there is no relation between, say, the number of vowels and
the number of consonants, either in the inventory, or in per
missible sequences. This would, of course, be a peculiar use of the
term universal, but such statements result from exactly the same
procedures which yield statements about the other-than-chance
cooccurrence of features.
Notes
1. Cf. Fred W. Householder, "On linguistic terms," in Sol Saporta
(ed.), Readings in Psycholinguistics (New York, 1961), pp. 15-25.

The author's research is supported in part by grant M 2385 from the
National Institutes of Health, Public Health Service, and by grants
G2502 and G5555 from the National Science Foundation.
2. The symbol -io- is to be read 'implies the presence of (but not vice
versa).' Rulon Wells pointed out during the Conference that
every hypothesis of the form X-io-Y had a diachronic counterpart
-namely, given a language with X and no Y, we predict either
the disappearance of X or the appearance of Y.
3. Householder now suggests the following as a sketch of such a
definition:
"(a) [f a language L has a favorite sentence-type of two parts
(X and Y) such that personal names occur more frequently as all
71
or the nucleus of one part (say X) than of the other (Y), then X
may be called 'noun phrase.'
"(b) If there is a (syntactically or morphologically defined)
form-class which regularly occurs as head of 'noun phrase', but
less often (or never) as head of Y, then the members of this
form-class are 'nouns.' "
Even the preceding includes one term, personal names, which
requires some purely formal bases for identification.
See now, however, a more meaningful discussion of such notions
by Jerrold J. Katz and Paul M. Postal, An Integrated Theory of
Linguistic Descriptions (Cambridge, 1964).
4. C. E. Bazell, Linguistic Form (Istanbul, 1953), pp. 47-48. A number
of the participants in the Conference pointed out that, in fact,
the two definitions do not yield the same classes for all languages.
Hockett, who commented in writing, cites Bella Coola as an
example of a language which has "utterances without vowels."
Householder suggests that "if a language has alI 'words' of phonetic
CVC form, then one V is predictable (and therefore non-phonemic),
and the statement will not be true." This is an example of the
effec.t of nonunique solutions on any cross-language comparison.
5. In the original version of this paper, I suggested these were formal
terms, presumably in the grammar, but I prefer to modify this
view now.
6. Cf. Noam Chomsky, Syntactic Structures ('s-Gravenhage, 1957),
and "Three Models for the Description of Language," IRE
Transactions on Information Theory, IT-2, No.3. 113-124 (1956).
The relevance of these models for phonological descriptions is
suggested in Sol Saporta and Heles Contreras, A Phonological
Grammar of Spanish (Seattle, 1962). Recent work in generative
phonology has made clear the need for modification of the view
suggested here. See, for example, Morris Halle, "Phonology in
Generative Grammar," Word 18 (1962), 54-72.
7. The correspondence will not be perfect. It is conceivable, and for
some languages perhaps inevitable, that the elements will occasion
ally correspond, on the one hand, to features and, on the other, to
clusters. [Indeed, recent work in generative phonology points
out the weakness of this position. The relevance of features in
distribution was unexplored in this paper. Similarly, the phono
logical segments referred to as phonemes are not identical to the
now classical position regarding segmentation.]
8. In his written comments, Hockett suggests that "there are languages
where olle does not need both phoneme and syllable," and quite
correctly points out that "in such languages it is purely a matter
of convenience whether one drops the one or the other term."
.'
,.
72
Sol Saporta
He concludes, however, that consequently neither unit is a universal.

Jakobson, during the Conference seemed to propose, on the con
trary, that both phoneme and syllable were universals and that any
distinction is unwarranted. However, there seems to be no a priori
reason for positing both, and this suggests that they are entities
of a different sort. Languages of the type Hockett mentions are
only one kind of evidence that they are different.
9. Sol Saporta, "Frequency of Consonant Clusters," Language 31.
25-30 (1955). See, too, Kathryn C. Keller and Sol Saporta,
"Frequency of consonant clusters in Chontal," lJAL 23.28-35
(1951); G. L. Bursill-Hall, "Consonant clusters in French,"
JCLA 2.66-17 (1956); John B. Carroll, "The assessment of pho
neme cluster frequencies," Language 34.261-218 (1958).
10. Sol Saporta and Rita Cohen, "The distribution and relative
frequency of Spanish diphthongs," Romance Philology II. 311
377 (1958).
11. Leonard Bloomfield, Language (New York, 1933), p. 20, cited
with apparent approval in the work papers by Hockett and
Ullmann.
12. Fred Householder proposed a number of plausible phonemic
solutions which affected the validity of certain hypotheses. For
example, Hypothesis 3 is almost certainly false for English if
one posits I-pl/ in apple, etc.
13. Posing the question in these terms assumes an item-and-arrange
ment model of morphemics. However, some such questions can
be framed for any model. For an item-and-process model one
could determine the length in morphophonemes of (the basic
form of) morphemes.
14. Actually, the sample is restricted to those morphemes with more
than one alternant. Sol Saporta, "Morpheme alternants in Spanish,"
in Henry R. Kahane and Angelina Pietrangeli (eds.), Structural
Studies on Spanish Themes (Salamanca, 1959), pp. 15-162.
15. Roman Jakobson pointed out a number of "compensatory"
alternatives which would relate to a small phonemic inventory:
(1) either the morphs might be longer than usual, or (2) a larger
than average percentage of possible combinations are in fact
utilized, or (3) there are a larger number of homonyms. WilJiam
Gedney suggested that the presence of phonemic tone or stress
might be particularly relevant in any precise formulation of this
hypothesis.
CHAPTER 5
SOME UNIVERSALS OF GRAMMAR WITH
PARTICULAR REFERENCE TO THE ORDER OF
MEANINGFUL ELEMENTS
JOSEPH H. GREENBERG
Stanford University
1. Introduction
The tentative nature of the conclusions set forth here should

be evident to the reader. Without much more complete sampling
of the world's languages, the absence of exceptions to most of the
universals asserted here cannot be fully assured. As indicated by
the title, attention has been concentrated largely, but by no
means exclusively, on questions concerning morpheme and word
order. The reason for this choice was that previous experience
suggested a considerable measure of orderliness in this particular
aspect of grammar. In the body of this paper a number of univer
sals are proposed. A large proportion of these are implicational;
that is, they take the form, "given x in a particular language, we
always find y." When nothing further is said, it is understood that
the converse, namely, "given y, we always find x," does not hold.
Where the two sets of characteristics are binary, the typical distri
bution in a tetrachoric table is a zero as one of the four entries. 1
From the point of view of scientific methodology, there is nothing
to apologize for in such results, and this is so for two reasons.
First, the lowest-level laws as described in manuals of scientific
method take precisely this form. It Second, what seem to be non~
implicational universals about language are in fact tacitly impli~
cational since they are implied by the definitional characteristics
of language. II Further, to assert the definitional characteristics
themselves is obviously tautologous.
73
74
Joseph H. Greenberg
It is perhaps worth while to point out that a number of uni4

versals of the second type-that is, those implied by the defini4
tional characteristics of language-although not usually formally
stated in this paper, are in fact involved in the notion of the
general comparability of languages in the grammatical sphere
which underlies the specific statements found here. For example,
a whole series of universals in the usual sense are assumed in such
a statement as the following: If a language has verb4subject-object
as its basic word order in main declarative clauses, the dependent
genitive always follows the governing noun. It is here assumed,
among other things, that all languages have subject-predicate
constructions, differentiated word classes, and genitive construc
tions, to mention but a few. I fully realize that in identifying such
phenomena in languages of differing structure, one is basically
employing semantic criteria. There are very probably formal
similarities which permit us to equate such phenomena in different
languages. However, to have concentrated on this task, important
in itself, would have, because of its arduousness, prevented me
from going forward to those specific hypotheses, based on such
investigation, which have empirical import and are of primary
interest to the nonlinguist. Moreover, the' adequacy of a cross
linguistic definition of 'noun' would, in any case, be tested by
reference to its results from the viewpoint of the semantic phe
nomena it was designed to explicate. If, for example. a formal
definition of 'noun' resulted in equating a class containing such
glosses as 'boy', 'nose', and 'house' in one language with a class
containing such items as 'eat', 'drink', and 'give' in a second
language, such a definition would forthwith be rejected and that
on semantic grounds. In fact, there was never any real doubt in
the languages treated about such matters. There is every reason to
believe that such judgments have a high degree of validity. If.
for example, someone were to dispute the specific assignment of
order type of a genitive construction given in this paper. it is quite
clear on what evidence such an assignment would be accepted or
rejected.
For many of the statements in this paper, a sample of the
following 30 languages has been utilized: Basque, Serbian, Welsh.
SOME UNIVERSALS OF GRAMMAR
7S
Norwegian, Modern Greek. Italian, Finnish (European); Yoruba.

Nubian, Swahili, Fulani. Masai, Songhai, Berber (African);
Turkish, Hebrew, Burushaski, Hindi, Kannada, Japanese, Thai,
Burmese, Malay (Asian); Maori, Loritja (Oceanian); Maya
Zapotec, Quechua, Chibcha, Guarani (American Indian).
This sample was selected largely for convenience. In general,
it contains languages with which I had some previous acquaint
ance or for which a reasonably adequate grammar was available
to me. Its biases are obvious, although an attempt was made to
obtain as wide a genetic and areal coverage as possible. This
sample was utilized for two chief purposes. First, it seemed likely
that any statement which held for all of these 30 languages had a
fair likelihood of complete or, at least, nearly complete universal
validity. Second, less reliably, it serves to give some notion of the
relative frequency of association of certain grammatical traits. In
this respect, of course, it is not to be taken literally. On some
questions I have gone well outside the sample.
The main section of the paper, which follows, is concerned with
the establishment of universals on the basis of the empirical
linguistic evidence. These are presented with a minimum of
theoretical comment. The final section is exploratory, seeking
to discover what general principles may exist from which at least
some of the. generalizations of the earlier sections might be de
duced. For convenience of exposition, the universals scattered
though the text are repeated for cross reference in Appendix III.
The theoretical section is far more speculative and uncertain than
the sections devoted to the universals themselves. In a certain
sense we would prefer to have as few universals as possible, not
as many. That is, we would like to be able to deduce them from
as small a number of general principles as possible. However,
the establishment of a relatively large number of empirical gen
eralizations must, on the whole, come first. For one thing, it
would be embarrassing to deduce a particular universal from
what seemed like a valid general principle, only to discover that
the generalization was not empirically valid.
I'
"
76
Joseph H. Greenberg
2. The Basic Order Typology"

Linguists are, in general, familiar with the notion that certain
languages tend consistently to put modifying or limiting elements
before those modified or limited, while others just as consistently
do the opposite. Turkish, an example of the former type,
puts adjectives before the nouns they modify, places the object
of the verb before the verb, the dependent genitive before the
governing noun, adverbs before adjectives which they modify,
etc. Such languages, moreover, tend to have postpositions for
concepts expressed by prepositions in English. A language of the
opposite type is Thai, in which adjectives follow the noun, the
object follows the verb, the genitive follows the governing noun,
and there are prepositions. The majority of languages, as for
example English, are not as well marked in this respect. In
English, as in Thai, there are prepositions, and the noun object
follows the verb. On the other hand, English resembles Turkish in
that the adjective precedes the noun. Moreover, in the genitive con
struction both orders exist: 'John's house' and 'the house of John'.
More detailed consideration of these and other phenomena of
order soon reveals that some factors are closely related to each
other while others are relatively independent. For reasons which
will appear in the course of the exposition, it is convenient to
set up a typology involving certain basic factors of word order.
This typology will be referred to as the basic order typology.
Three sets of criteria will be employed. The first of these is the
existence of prepositions as against post positions. These will be
symbolized as Pr and Po, respectively. The second will be the
relative order of subject, verb, and object in declarative sentences
with nominal subject and object. The vast majority of languages
have several variant orders but a single dominant one. Logically,
there are six possible orders: SVO, SOY, VSO, VOS, OSV, and
OVS. Of these six, however, only three normally occur as
dominant orders. The three which do not occur at all, or at least
are excessively rare, are VOS, OSV, and OVS. These all have in
common that the object precedes the subject. This gives us our
first universal:
77
Universal 1. In declarative sentences with nominal subject

and object, the dominant order is almost always one in
which the subject precedes the object. II
This leaves us with three common types: VSO, SVO, and SOY.
These will be symbolized as I, II, and III, respectively, reflecting
the relative position of the verb.
The third basis of classification will be the position of qualifying
adjectives (Le., those designating qualities) in relation to the
noun. As will be seen later, the position of demonstratives, ar
ticles, numerals, and, quantifiers (e.g., 'some', 'all') frequently
differs from that of qualifying adjectives. Here again there is
sometimes variation, but the vast majority of languages have a
dominant order. Dominant order with adjective preceding noun
will be symbolized by A and dominant order with noun preceding
adjective by N. We thus arrive at a typology involving 2 X 3 X 2,
that is, twelve logical possibilities. The 30 languages of the sample
are distributed among these twelve classes as shown in Table I. 8
TABLE
Po-A
Po-N
Pr-A
Pr-N
0
0
0
6
1
II
III
4
6
The table has been arranged so that the 'extreme' types Po-A
and Pr-N are in the first and fourth row, respectively. It is evident
that with respect to these extremes, I and III are polar types,
the former being strongly correlated with Pr-N and the latter
with Po-A. Type II is more strongly correlated with Pr-N than
with Po-A. It is also clear that adjective position is less closely
related to types I, II, and III than is the Pr/Po contrast. The table
is, I believe, a fair representation of the relative frequency of
these alternatives on a world-wide basis. Type II is the most
frequent; type III almost as common; type I is a definite minority.
This means that the nominal subject regularly precedes the verb
in a large majority of the world's languages.
78
Joseph H. Greenberg
Turning for a moment to gemtlve order, we note that this

characteristic might fittingly have been utilized for typological
purposes. The reason for not employing it is its extremely high
correlation with Pr/Po, a fact generally known to linguists. It
would thus virtually have duplicated the latter criterion. It was
not chosen because Pr/Po on the whole is slightly more highly
correlated with other phenomena. Of the present sample of 30
languages, 14 have post positions, and in everyone of these the
genitive order is genitive followed by governing noun. Of the
14 prepositional languages, 13 have the genitive following the gov~
erning noun. The only exception is Norwegian, in which the
genitive precedes. Thus, 29 of the 30 cases conform to the rule.
If anything, 1/30 is an overestimation of the proportion of
exceptions on a world-wide basis. We therefore have the following
universal:
Universal 2. In languages with prepositions, the gemtIve
almost always folIows the governing noun, while in languages
with post positions it almost always precedes.
Turning once more to the data of Table I, we find striking evi

dence of lawful relationships among the variables in that of the 12
possibilities 5, or almost half, are not exemplified in the sample.
All of these types are either rare or nonexistent. 7 For type I, we
see that all 6 languages of the sample are Pr/N. This holds with
extremely few exceptions on a world-wide basis. There are,
however, a few valid examples of [/Pr/A, the mirror image, so
to speak, of the fairly frequent III/Po/N. On the other hand,
there are, as far as I know, no examples of either l/Po/A or
l/Po/N. Hence we may formulate the following universal:
Universal 3. Languages with dominant VSO order are
always prepositional.
Languages of type 11 r are, as has been seen, the polar opposites

of type I. Just as there are no postpositionallanguages in type I,
we expect that there will be no prepositional languages in type III.
This is overwhelmingly true, but I am aware of several excep-
79
tions. 8 Since, as has been seen, genitive position correlates highly

with Pr/po, we will expect that languages of type III normally
have ON order. To this there are some few exceptions. However,
whenever genitive order deviates, so does adjective order, whereas
the corresponding statement does not hold for Pr/Po.9 We
therefore have the following universals:
Universal 4. With overwhelmingly greater than chance
frequency, languages with normal SOy order are postposi
tional.
Universal 5. If a language has dominant SOY order and the
genitive follows the governing noun, then the adjective like
wise follows the noun.
An important difference may be noted between languages of

types I and III. In regard to verb-modifying adverbs and phrases
as well as sentence adverbs, languages of type I show no reluc
tance in placing them before the verb so that the verb does not
necessarily begin the sentence. Further, all YSO languages ap
parently have alternative basic orders among which SVO always
figures. On the other hand, in a substantial proportion, possibly
a majority, of type III languages, the verb follows all of its mod
ifiers, and if any other basic order is allowed, it is OSV. Thus
the verb, except possibly for a few sentence modifiers (e.g.,
interrogative particles), is always at the end in verbal sentences.
It is not logically required, of course, that languages all of whose
basic orders involve the verb in the third position should also re
quire all verb modifiers to precede the verb, but this seems to hold
empirically. Thus, languages in which the verb is always at the
end may be called the "rigid" subtype of III. In the present sample,
Burushaski, Kannada, Japanese, Turkish, Hindi, and Burmese
belong to this group, while Nubian, Quechua, Basque, Loritja,
and Chibcha do not. 10 These considerations permit us to state
the following as universals:
Universal 6. All languages with dominant VSO order have
SVO as an alternative or as the only alternative basic order.
l'
80
Joseph H. Greenberg
Universal 7. If in a language with dominant SOY order, there

is no alternative basic order, or only OSY as the alternative,
then all adverbial modifiers of the verb likewise precede the
verb. (This is the rigid subtype of III.)
3. Syntax
Having defined the basic order typology and stated some of the
universals that can be most immediately derived from the con
sideration of its defining properties, we turn to a number of syn
tactic universals, many but not all of which are associated with
this typology. One set of criteria employed in this typology was
the order of nominal subject, nominal object, and verb in de
clarative sentences. One reason for stating the criteria in this
manner was that interrogative sentences tend to exhibit certain
characteristic differences as compared to declarative statements.
There are two main categories of questions, those of the yes-no
variety and those involving specific question words. A common
method ofdifferentiating yes-no questions from the corresponding
statement is by a difference of intonational pattern, as in English.
Our knowledge of these patterns still leaves much to be desired.
However, the following statement seems to be sufficiently docu
mented:
Universal 8. When a yes-no question is differentiated from
the corresponding assertion by an intonational pattern, the

distinctive intonational features of each of these patterns are
reckoned from the end of the sentence rather than from the
beginning.
For example, in English a yes-no question is marked by a rise
in pitch in the last stressed syllable of the sentence and the corres
ponding statement by falling pitch. The reckoning of distinctive
patterns from the end of the sentence may well hold for all
intonational patterns.
Yes-no questions may likewise be signaled by a question par
ticle or affix. Some languages use both this method and intonation
as alternatives. The position of such question markers is fixed
81
;~
by either reference to some specific word, most frequently the

verb, or the emphasized word of the question, or it may be fixed
by position in the sentence as a whole. In languages of the rigid
subtype III, it is of course impossible to distinguish between
position after the verb and position at the end of the sentence. In
the present sample, there are 12 languages with such initial or
final particles. With reference to the basic order typology, these
12 examples are distributed as shown in Table 2.11
TABLE
Initial particle
Final particle
';1
l
!,
i:~"
II
III
The two examples of a final particle in group II are preposi

tional languages (Thai and Yoruba). The table includes only
cases where there is a single such particle or affix in the language,
or there are several following the same rule. In two of the lan
guages in the samples, there is more than one such element, each
with differing rules. Zapotec (I/Pr) has either an initial particle
alone or this same particle in conjunction with a final particle.
Songhai (II/Po) has three such particles, two of them an initial and
one a final particle. These complications as well as the fact that at
least one language outside of the sample belonging to (II/Po),
namely, Lithuanian, has an initial particle suggest the following
rather cautious statement:
Universal 9. With well more than chance frequency, when
question particles or affixes are specified in position by refer
ence to the sentence as a whole, if initial, such elements are
found in prepositional languages, and, if final, in post
positional.
Where specification depends on some particular word, the par

ticle almost always follows. Such particles are found in 13 lan
guages of the present sample. 12 Examples of the rigid subtype III
are counted in both this and the previous category. Of these 13,
12 are suffixed. They include both prepositional and post
'i\
i,
l'
"1t
"
82
Joseph H. Greenherg
positional languages, but none in group 1. The following, there

fore, probably holds:
Universal 10. Question particles or affixes, when specified in
position by reference to a particular word in the sentence,
almost always follow that word. Such particles do not occur
in languages with dominant order VSO.
The other basic kind of question, that involving an interroga

tive word, likewise shows a definite relationship to the basic
order typology. In such sentences, many languages have a different
word order than that of the corresponding declarative sentence.
Characteristically, the question word comes first, except for the
possible retention of normal order within smaller units (e.g.,
phrases). This holds in English, for example, where the question
word is first in 'What did he eat?' as against the statement, 'He
ate meat'. The second point is illustrated by 'With whom did he
go?' as against 'He went with Henry', where the question phrase
comes first but the order within the phrase itself is not disturbed.
Many languages which put interrogatives first likewise invert the
order of verb and subject (e.g., German 'Wen sah er?'). Such lan
guages sometimes invert for yes-no questions, (e.g., 'Kommt
er?'). It appears that only languages with interrogatives always
initially invert, and only languages which invert in interrogative
word questions invert for yes-no questions. 13
In the present sample, 16 languages put the interrogative word
or phrase first. They are distributed as shown in Table 3.
TABLE
Question word first

Question and statement
order identical
Question word first

Question and statement
order identical
II
III
10
II
Pr
Po
14
12
83
A definite relationship thus appears, and we have the follow

ing universals:
Universals 11. Inversion of statement order so that verb
precedes subject occurs only in languages where the question
word or phrase is normally initial. This same inversion
occurs in yes-no questions only if it also occurs in interrog
.ative word questions.
Universal 12. If a language has dominant order VSO in
declarative sentences, it always puts interrogative words or
phrases first in interrogative. word questions; if it has
dominant order SOV in declarative sentences, there is never
such an invariant rule.
Verbal subordination to verb will be considered next. Seman

tically, the concepts to be considered here include time, cause,
purpose, and condition. Formally, we have one or more of the
following: introductory words (Le., "conjunctions"); and verbal
inflections, whether finite, involving categories of person and
number (e.g., subjunctives) or nonfinite forms such as verbal
nouns and gerundives. It seems probable that conjunctions are
more frequent in prepositional languages, nonfinite verb forms
in postpositional languages, and that finite verb forms are found
in both, but this point was not investigated. In accordance with
the over-all emphasis of the paper, attention was directed to the
question of the relative order of subordinate and main verbal
forms. Since the subordinate verb qualifies the main verb, we
would expect it to precede the main verb in all languages of the
rigid subtype of III. Since this subtype was defined merely in
terms of the invariable precedence of noun object, the question
remains for empirical verification. In fact, this turns out to be
true for aU the languages of this subtype in the sample and, no
doubt, holds generally.14 In languages of other types certain
characteristics of individual constructions appear. The normal
order everywhere is for the protasis of conditional constructions
to procede the apodosis, that is, for the condition to precede the
conclusion. This is true for aU 30 languages of the sample. In
84
These data suggest the following universal:
languages of the rigid subtype of III the protasis never follows,

but in other languages it will do so occasionally.
On the other hand, in expressions of purpose and volition the
normal order is for these to follow the main verb except in lan
guages of the rigid subtype of III. Here again there are no ex
ceptions in the sample. We have therefore the following uni
versals:
Universal 16. In languages with dominant order VSO, an

inflected auxiliary always precedes the main verb. In lan
guages with dominant order SOV, an inflected auxiliary
always follows the main verb.
Uninflected auxiliaries will be considered later in connection

with verb inflections.
In nominal phrases, the position of attributive adjectives in
relation to the noun modified is a key factor. The position of the
qualifying adjective shows a definite though only statistical rela
tion to the two other bases of the typology. A summary of these
data for the languages of the sample is given in Table 5.
Universal 13. If the nominal object always precedes the yerb,

then verb forms subordinate to the main verb also precede it.
Universal 14. In conditional statements, the conditional
clause precedes the conclusion as the normal order in all
languages.
Universal 15. In expressions of volition and purpose, a
subordinate verbal form always follows the main verb as the
normal order except in those languages in which the nominal
object always precedes the verb.
Another relation of verb to verb is that of inflected auxiliary to

main verb. For present purposes, such a construction will be
defined as one in which a closed class of verbs (the auxiliaries)
inflected for both person and number is in construction with an
open class of verbs not inflected for both person and number. For
example, in English <is going' is such a construction. This defini
tion, of course, excludes the possibility of such a construction in
languages in which the verb has no category of person and
number (e.g., Japanese). In the sample of 30 languages, 19 have
such inflected auxiliaries. They are distributed among the order
types as shown in Table 4. 15
TABLE
II
I
3
0
II
7
1
Po
Auxiliary precedes verb

Auxiliary follows verb
Pr
9
0
1
9
III
0
8
TABLE
I;
NA
AN
6
0
II
8
5
NA
AN
Pr
12
4
Po
7
7
:i
I'
I.
~
11
Auxiliary precedes verb

Auxiliary follows verb
85
Joseph H. Greenberg
1)
!i:
III
5
6
In general, then, the tendency is for adjectives to follow the

noun in prepositional languages, and most strongly so in lan
guages of type I, which are always prepositional as has been
noted. There are a few rare exceptions, not in the sample, of
languages of type I with adjective before the noun, as was noted
earlier. Hence, we have the following near universal:
Universal 17. With overwhelmingly more than chance fre
quency, languages with dominant order VSO have the
adjective after the noun.
From the data of Table 5, it will also be noticed that there are
19 languages with adjective after the noun, as against 11 with the
adjective before the noun. This is representative of a general

tendency which very nearly overrides the opposite rule to be
expected in languages of type HI.
r
86
Joseph H. Greenberg
The position of demonstratives and numerals is related to that

of descriptive adjectives in individual languages. However, these
items show a marked tendency to precede even when the descrip
tive adjective follows. On the other hand, when the descriptive
adjective precedes, then the demonstratives and numerals vir
tually always precede the noun likewise. The data from the sam
ple languages are given in Table 6.
TABLE
I
1
12
11
8
11
AN
o
10
In one language, Guarani, numbers may either precede or

follow the noun, and this case was not included in the table. In
Guarani, the adjective follows the noun, as would be expected. In
the case of numbers, it should be noted that for languages with
numeral classifiers, it was the position of the numeral in relation
to the classifier which was taken into account. 16 There seems to be
no relation between the position of the numeral and the demon
strative outside of that mediated by adjective position. Languages
in which the adjective follows the noun may have numeral
preceding while demonstrative does not, demonstrative pre
ceding while numeral does not, both preceding or neither
preceding. Outside of the sample, however, there are a small
number of instances (e.g., Efik) in which the demonstrative
follows while the adjective precedes. It may be noted that other
quantifiers (e.g., 'some', 'all') and interrogative and possessive
adjectives show this same tendency to precede the noun, as evi
denced, for example, in the Romance languages, but those cases
were not studied. We have then the following universal:
Universal 18. When the descriptive adjective precedes the
noun, the demonstrative and the numeral, with over
whelmingly more than chance frequency, do likewise.
87
An additional related observation may be noted:

Universal 19. When the general rule is that the descriptive
adjective follows, there may be a minority of adjectives which
usually precede,but when the general rule is that descriptive
adjectives precede, there are no exceptions.
This last universal is illustrated by Welsh and Italian in the
present sample.
The order within the noun phrase is subject to powerful con
straints. When any or all of the three types of qualifiers precede
the noun, the order among them is always the same: demonstra
tive, numeral, and adjective, as in English, 'these five houses'.
When any or all follow, the favorite order is the exact opposite:
noun, adjective, numeral, demonstrative. A less popular alter
native is the same order as that just given for the instances in
which these elements precede the noun. An example of the latter
is Kikuyu, a Bantu language of East Africa, with the order,
'houses these five large', instead of the more popular 'houses large
five these'. We have, then, a universal:
Universal 20. When any or all of the items (demonstrative,
numeral, and descriptive adjective) precede the noun, they
are always found in that order. If they follow, the order is
either the same or its exact opposite.
The order of adverbial qualifiers of adjectives in relation to the
adjective will noW be considered. This order also shows a definite
relation to that between the descriptive adjective and the noun, as
shown by Table 7. Tn the third row are cases in which certain
adverbs precede and others follow Y
6
NA
Oem. - Noun
Noun-Oem.
Num. - Noun
Noun - Num.
-+
TABLE
Adverb _ Adjective
Adjective _ Adverb
Adj. _ Adv. and Adv. - Adj.
7
AN
NA
11
0
0
5
8
2
From Table 7 it can be seen that there is a tendency for the

adverb to precede the adjective, which can be overridden only in
some cases when the adjective follows the noun. The situation
88
Joseph H. Greenberg
thus far is similar to that obtaining with regard to demonstratives

and numerals. However, if we look further, we note that all of
those languages in which some or all adverbs follow the adjective
not only have the nO'un followed by the adjective, but also are
all of types I and II. Thus we have a universal:
Universal 21. If some or all adverbs follow the adjective they
modify, then the language is one in which the qualifying
adjective follows the noun and the verb precedes its nominal
object as the dominant order.
One other topic concerning the adjective to be considered is

that of comparisons, specifically that of superiority as expressed,
for example in English, by sentences of the type 'X is larger than
Y'. A minority of the world's languages have, like English, an
inflected comparative form of the adjective. More frequently a
separate word modifies the adjective, as in English, 'X is more
beautiful than Y', but in many languages this is optional or
does not exist at all. On the other hand, there is always some
element which expresses the comparison as such, whether word
or affix, corresponding to English 'than', and obviously both the
adjective and the item with which comparison is made must be
expressed. We thus have three elements whose order can be
considered, as in English larg(er) than Y. These will be called
adjective, marker of comparison, and standard of comparison.
The two common orders are: adjective, marker, standard (as
in English); or the opposite order: standard, marker, adjective.
These two alternatives are related to the basic order typology,
as shown by Table 8. 18 A number of languages are not entered
this table because they utilize a verb with general meaning 'to
surpass'. This is particularly common in Africa (e.g., Yoruba):
'X is large, surpasses Y', Loritja, an Australian language which
has 'X is large, Y is small', is likewise not entered.
TABLE
Adjective - Marker - Standard

Standard - Marker Adjective
Both
......::..:-::---:..: :. :~~-.
\'J..:q;,~ P~"
I'
i:
' r:
\ \b
\,
Adjective - Marker Standard

Standard Marker - Adjective
Both
o
o
I
II
III
. ~~
Po
I
10
1
o
o
A clear relation to the basic order typology is likewise found

in constructions of nominal apposition, particularly those in
volving a common along with a proper noun. A number of se
mantic and formal subtypes are involved (e.g., titles of address,
'Mr. X,' as against appellations 'Avenue X'). The latter type is, in
certain cases, assimilation to the genitive, and may therefore be
expected to show a similar order (e.g., 'the city of Philadelphia').
English is somewhat ambivalent, doubtless because of adjective
noun order, as can be seen from '42nd Street' versus 'Avenue A',
or 'Long Lake' versus 'Lake Michigan'. Most languages, how
ever, have a single order (e.g., French, 'Place Vendome', 'Lac
Geneve', 'Boulevard Michelet'), My data here are incomplete
because grammars often make no statement on the subject, and I
was dependent on text examples. 19
In Table 9, contrary to usual practice, the genitive construction
is used instead of Pr/Po since it gives more clear-cut results.
TABLE
Common Noun - Proper Noun

Proper Noun - Common Noun
Common Noun - Proper Noun
Proper Noun - Common Noun
.f
,-'
Pr
13
Universal 22. If in comparisons of superiority the only order,
or one of the alternative orders, is standard-marker-adjective,
then the language is postpositional. With overwhelmingly
more than chance frequency if the only order is adjective
marker-standard, the language is prepositional.
89
9
I
II
111
2
0
7
2
0
6
ON
NO
8
0
Universal 23. If in apposition the proper noun usually
precedes the common noun, then the language is one in
\{,
,1
Ii,
t;
90
Joseph H. Greenberg
which the governing noun precedes its dependent genitive.

With much better than chance frequency, if the common
noun usually precedes the proper noun, the dependent
genitive precedes its governing noun.
As the concluding item in the discussion of nominal construc
tion, we take the relative clause which modifies a noun (e.g.,
English, 'I saw the man who came', 'I saw the student who failed
the examination'). Here again there is considerable diversity of
format means from language to language. All that will be consid
ered here is the order as between nominal antecedent and the verb
of the relative clause (e.g., 'man' and 'came' in the first sentence).
Once more the distribution of the rules of order, as set forth in
Table 10, shows a clear relation to the categories of the basic order
typology. 20
TABLE
Relational expression precedes noun

Noun precedes relational expression
Both constructions
Relational expression precedes noun

Noun precedes relational expression
Both constructions
10
1
II
III
0
6
0
0
12
2
I
Pr
Po
0
16
0
7
4
From Table 10 it is clear that if the relational expression pre

cedes the noun either as the only construction or as alternate con
struction, the language is post positionaL However, outside of
the sample there is at least one exception, Chinese, a preposi
tional language in which the relational expression precedes the
noun. It is plausible to explain this deviation as connected with
the fact that in Chinese the adjective precedes the noun. As with
adjective-noun order there is a pronounced general tendency for
the relative expression to follow the noun it qualifies. This
tendency is sometimes overcome but only if (I) the language is
prepositonal or (2) if the qualifying adjective precedes the noun.
91
Universal 24. If the relative expression precedes the noun

either as the only construction or as an alternate construction,
either the language is postpositional, or the adjective precedes
the noun or both.
Thus far nothing has been said about pronouns. In general,

pronouns exhibit differences regarding order when compared
with nouns. This was the reason for specifying nominal subject
and nominal object in the definitions of the basic typology. One
peculiarity of pronominal order is illustrated by French where we
have, 'Je vois l'homme' but 'Je Ie vois'; that is, the pronominal
object precedes, whereas the nominal object follows. Similar
examples are found in a number of languages of the sample. In
Italian, Greek, Guarani, and Swahili, the rule holds that the
pronominal object always precedes the verb, whereas the nominal
object follows. In Italian and Greek, however, the pronoun
follows just as does the nominal object with imperatives. In
Berber the pronoun objects, direct or indirect, precede the verb
when the verb is accompanied by the negative or future particle.
In Loritja, the pronominal object may be an enclitic added to
the first word of the sentence. In Nubian, the usual nominal order
is SOY, but the alternative SVO is fairly frequent. For pronominal
object, this alternative never occurs. In other words, the pro
nominal object always precedes the verb, whereas the nominal
object may either precede or follow. In Welsh, in an alternative
order with emphasis on the pronoun subject, the pronoun subject
comes first in the sentence. In such sentences the pronominal
object precedes the verb, but the nominal object follows. Finally,
in Masai, whereas normal order for nominal object is VSO, a
pronominal object precedes a nominal subject and immediately
follows the verb.
No contrary instances occur in the sample of a pronominal
object regularly following the verb while a nominal object pre
cedes. We may therefore state the following universal:
Universal 25. If the pronominal object follows the verb, so

does the nominal object.
92
Joseph H. Greenberg
4. Morphology
Before proceeding to the question of inflectional categories,

which will be the chief topic of this section, certain general con
siderations relating to morphology will be discussed. Mor
phemes within the word are conventionally divided into root,
derivational and inflectional. As elsewhere in this paper, no
attempt at definition of categories will be attempted. Derivational
and inflectional elements are usually grouped together as affixes.
On the basis of their order relation to the root, they may be
classified into a number of categories. By far the most frequent
are prefixes and suffixes. Infixing, by which a derivational or
inflectional element is both preceded and followed by parts of the
root morpheme, may be grouped with other methods involving
discontinuity. Examples of such other methods are intercalation,
as in Semitic, and what might be called ambifixing, where an
affix has two parts, one of which precedes the entire root, while
the other follows. All such discontinuous methods are relatively
infrequent, and some languages do not employ any of them. The
following universal on this topic is probably valid:
Unirersal 26. If a language has discontinuous affixes, it
always has either prefixing or suffixing or both.
As between prefixing and suffixing, there is a general predomi

nance of suffixing. Exclusively suffixing languages are fairly
common, while exclusively prefixing languages are quite rare. In
the present sample, only Thai seems to be exclusively prefixing.
Here again a relationship with the basic order typology appears. 21
TABLE
Both
Exclusively prefixing
Exclusively suffixing
Both
93
Universal 27. If a language is exclusively suffixing, it is post

positional; if it is exclusively prefixing, it is prepositional.
Where both derivational and inflectional elements are found

together, the derivational element is more intimately connected
with the root. The following generalization appears plausible:
Universal 28. If both the derivation and inflection follow the
root, or they both precede the root, the derivation is always

between the root and the inflection.
There are probably no languages without either compounding,
affixing, or both. In other words, there are probably no purely
isolating languages. There are a considerable number of lan
guages without inflections, perhaps none without compounding
and derivation. The following probably holds:
Universal 29. If a language has inflection, it always has deri
vation.
Turning now to verb inflectional categories, we can state that
since there are languages without inflection, there will obviously
be languages in which the verb has no inflectional categories. In
the far more frequent cases in which the verb has inflectional
categories, a partial implicational hierarchy exists.
Universal 30. If the verb has categories of person-number or
if it has categories of gender, it always has tense-mode cate
gories.
The greater externality of gender categories in the verb can be

seen from the following generalization:
11
I
Exclusively prefixing
Exclusively suffixing
II
III
010
2
10
o
10
Pr
I
Po
0
12
15
Universal 31. If either the subject or object noun agrees with

the verb in gender, then the adjective always agrees with the
noun in gender.
Gender agreement between noun (usually noun subject) and

verb is far less frequent than agreement in person and number;
yet examples of the former without the latter do occur (e.g., in
some Daghestan languages of the Caucasus). However, where
94
Joseph H. Greenberg
such gender categories appear, they always seem to be associated

with number also. Therefore we have the following:
Universal 32. Whenever the verb agrees with a nominal
subject or nominal object in gender, it also agrees in number.
A further obsenation about noun-verb agreement in number

may be made. There are cases in which this agreement is reg
ularly suspended. In all such cases, if order is involved, the fol
lowing seems to hold: 22.
Universal 33. When number agreement between the noun
and verb is suspended and the rule is based on order, the case
is always one in which the verb precedes and the verb is in
the singular.
Such phenomena as the suspension of agreement are analogous

to that of neutralization in phonemics. The category which does
not appear in the position of neutralizatiDn, in this case the plural,
may be called the marked category (as in classical Prague School
phonemic theory). Similar phenomena will be encountered in
the subsequent discussion.
The three most common nominal inflectional categories are
number, gender, and case. Among systems of number, there is a
definite hierarchy which can be stated in the following terms:
Universal 34. No language has a trial number unless it has
a dual. No language has a dual unless it has a plural.
Nonsingular number categories. are marked categories in

relation to the singular, as indicated in the following universal:
Universal 35. There is no language in which the plural does
not have some nonzero allomorphs, whereas there are lan
guages in which the singular is expressed only by zero. The
dual and the trial are almost never expressed only by zero.
The marked character of the nonsingular numbers as against

the singular can also be seen when number occurs along with
gender. The interrelations of these two sets of categories are
stated in the following universals:
95
Universal 36. If a language has the category of gender, it

always has the category of number.
Universal 37. A language never has more gender categories
in nonsingular numbers than in the singular.
This latter statement may be illustrated from Hausa, which

has a masculine and feminine gender distinction in the singular
but not in the plural. The opposite phenomenon, to my knowl
edge, never occurs.
Case systems may occur with or without gender systems and
with or without the category of number. The unmarked categories
of case systems are the subject case in nonergative systems and
the case which expresses the subject of intransitive and the object
of transitive verbs in ergative systems. Hence we have the
following universal:
Universal 38. Where there is a case system, the only case
which ever has only zero allomorphs is the one which
includes among its meanings that of the subject of the
intransitive verb.
As between number and case, where there is a distinct mor

pheme boundary, the following relation almost always holds:
Universal 39. Where morphemes of both number and case
are present and both follow or both precede the noun base,
the expression of number almost always comes between the
noun base and the expression of case.
The following general statement may be made about agreement

between adjectives and nouns:
Universal 40. When the adjective follows the noun, the
adjective expresses all the inflectional categories of the noun.
In such cases the noun may lack overt expression of one or
all of these categories.
For example, in Basque, where the adjective follows the noun,

the last member of the noun phrase contains overt expressions
of the categories of case and number and it alone has them.
96
Joseph H. Greenberg
Case systems are particularly frequent in postpositional lan

guages, particularly those of type III. In the present sample, all
the languages of this type have case systems. There are a few
marginal cases or possible exceptions.
Universal 41. If in a language the verb follows both the
nominal subject and nominal object as the dominant order,
the language almost always has a case system.
Finally, pronominal categories may be briefly considered. In

general, pronominal categories tend to be more differentiated
than those of the noun, but almost any specific statement in this
regard will have some exceptions. As a general statement we have
the following universals:
Universal 42. All languages have pronominal categories in
volving at least three persons and two numbers.
Universal 43. If a language has gender categories in the
noun, it has gender categories in the pronoun.
Gender categories show certain relations to categories of

person in pronouns, as might be expected.
Universal 44. If a language has gender distinctions in the first
person, it always has gender distinctions in the second or
third person, or in both.
There is likewise a relation to the category of number.

Universal 45. If there are any gender distinctions in the plural
of the pronoun, there are some gender distinctions in the
singular also.
5. Conclusion: Some General Principles
No attempt is made here to account for all of the universals

described in the preceding sections and repeated in Appendix III.
Some general principles, however, are proposed which seem to
underlie a number of different universals and from which they
may be deduced. Attention is first directed to those universals
which are most closely connected with the basic order typology
97
and the closely associated genitive construction. Two basic no

tions, that of the dominance of a particular order over its al
ternative and that of harmonic and disharmonic relations among
distinct rules of order, are introduced. This latter concept is very
obviously connected with the psychological concept of gener
alization.
We may illustrate the reasoning involved by reference to Uni
versal 25, according to which, if the pronominal object follows
the verb, the nominal object does so likewise. In other words, in
the tetrachoric table resulting from the alternative for each of the
combinations there is a single blank. Since the nominal object may
follow the verb whether the pronoun object precedes or follows,
while the nominal object may precede the verb only if the pronoun
precedes, we will say that VO is dominant over OV since OV only
occurs under specified conditions, namely when the pronominal
object likewise precedes, while VO is not subject to such limita
tions. Further, the order noun object-verb is harmonic with
pronoun object-verb but is disharmonic with verb-pronoun
object since it does not occur with it. Likewise verb-noun object
order is harmonic with verb-pronoun object and disharmonic
with pronoun object-verb. We may restate our rule, then, in
terms of these concepts as follows:
A dominant order may always occur, but its opposite, the
recessive, occurs only when a harmonic construction is
likewise present.
Note that the notion of dominance is not based o"n its more
frequent occurrence but on the logical factor of a zero in the
tetrachoric table. It is not difficult to construct an example in
which one of the recessive alternatives is more frequent than the
dominant. Dominance and harmonic relations can be derived
quite mechanically from such a table with a single zero. The entry
with zero is always the recessive one for each construction, and
the two constructions involved are disharmonic with each other.
Harmonic and disharmonic relations, as noted earlier, are
examples of generalization. In similar constructions, the cor
responding members tend to be in the same order. The basis for
98
Joseph H. Greenberg
99
From the overwhelming association of prepositIOns with

governing noun-genitive order and of postpositions with genitive
governing noun order but with a small number of exceptions of
both types, the conclusion is drawn that prepositions are harmonic
with NG and postpositions with GN.
The close connection between genitive order and Pr/Po is a
simple instance of generalization. The relation of possession is
assimilated to other relational notions, for example, spatial
relations. In English, 'of' which marks possession is a preposition
with the same order properties as 'under', 'above', etc. Further,
such spatial and temporal relations are often expressed by nouns
or nounlike words, for example, English 'in back of'. In many lan
guages 'behind' = 'the back + genitive'; hence: 'X's back' = 'in
back of X' parallels' X's house'; and 'back of X' = 'in back of X'
parallels 'house of X'.
The connection between these genitives and the analogous pre
positional or postpositional phrases on the one hand, and subject
verb and object-verb constructions on the other, is via the so
called subjective and objective genitive. Note that in English
'Brutus' killing of Caesar started a civil war' has the same truth
value as 'The fact that Brutus killed Caesar started a civil war'.
The order of elements is likewise similar. In other words, in such
transformations, the noun subject or object corresponds to the
genitive, and the verb to the governing noun. In fact, there are
languages in which the subject or the object of the verb is in the
genitive. For example, in Berber argaz 'man' is the general form
of the noun, and urgaz is either the dependent genitive or the
subject of the verb, provided it follows immediately. Thus ifJey
urgaz, 'went out the man', exactly parallels axam urgaz, 'the
house of the man'. Berber, it will be noted, is a language of type I,
and the genitive follows the noun. It likewise has prepositions
rather than postpositions.
A further relationship among the variables of the basic order
typology may be posited, that between genitive order and adjec
tive order. Both the genitive and qualifying adjectives limit the
meaning of the noun. There are further facts to support this. There
are languages like Persian, in which both adjective and genitive
the correspondence in the present instance is obvious, in that

pronoun and noun are both objects of the verb, and the other pair
verb-verb is identical. In regard to harmonic and disharmonic
relations, a fair amount of freedom will be exercised based on
transformational and other relations among constructions, not
merely the occurrence of a zero in a tetrachoric table.
Proceeding on this basis, we now consider Universal 3. It will
be noted that this universal amounts to an assertion of the non
existence of postpositional languages of type I. Since in all of the
types, I, II and III, S precedes 0, this is irrelevant for the present
context. This leads to the following conclusions:
Prepositions are dominant over postpositions, and SV order
is dominant over VS order. Further, prepositions are har
monic with VS and disharmonic with SV, while postpositions
are harmonic with SV and disharmonic with VS.
What distinguishes type II from type III is that in type II the
object follows the verb, a characteristic shared with type I. On the
other hand, type III has the object before the verb. From Uni
versal 4, which states that with overwhelmingly more than chance
frequency SOY is associated with postpositions, the conclusion
is drawn that OV is harmonic with postpositions while VO is
harmonic with prepositions. The constructional analogies which
support this are discussed later with reference to the closely
associated genitive constructions. For the moment it may be noted
that the relations between types I, II, and III and Pr/Po may now
be recapitulated in these terms: Type I has VS which is harmonic
with prepositions, and SO which is likewise harmonic with prepo
sitions. Further, prepositions are dominant. All languages of
type I, in fact, are prepositional. Type II has SV which is harmonic
with postpositions and VO which is harmonic with prepositions,
and prepositions are dominant. In fact, a definite majority of
languages of type II have prepositions. Type III has SV and OV,
both of which are harmonic with postpositions. However, pre
positions are dominant. In fact, the preponderant majority of
languages which have type III have postpositions, with but a
handful of exceptions.
100
Joseph H. Greenberg
dependence are marked by exactly the same formal means. Where

pronominal possession is involved, some languages use a derived
adjective, while others use a genitive of the pronoun. There are even
instances where adjectives are used in the first and second person,
while a genitive is used in the third person (e.g., Norwegian).
We may summarize these results by stating that all of the fol
lowing are directly or indirectly harmonic with each other: pre
positions, NG, VS, va, NA. We have here a general tendency
to put modified before modifier, and the most highly "polarized"
languages in this direction are those of type I with NG and NA,
a considerable group of languages. The opposite type is based
on harmonic relations among postpositions, GN, SV, OV, and
AN. This is also a very widespread type, as exemplified by
Turkish and others in the present sample. On the other hand,
the general dominance of NA order tends to make languages of
the Basque type (Le., III/Po/NA with GN order) very nearly as
common as the Turkish type. It should also be pointed out that
languages being highly complex structures, there are other
factors at work in individual cases not included among the five
factors cited at this point. One of them, demonstrative-noun
order, has already been mentioned.
It is more difficult to account for the dominances than for the
harmonic relations, to explain, for example, why the adjective
tends to follow the noun. It may be suggested, however, that
noun-adjective predominance arises from the same factor as that
which makes subject-verb the dominant order. In Hockett's ter
minology, there is a general tendency for comment to follow
topic. There is some evidence that noun-adjective does parallel
subject-verb in this way. In many languages all adjectival notions
are treated as intransitive verbs. The qualifying adjective is then
a relative or participle of the verb. The tendency of relative
clauses, it has been seen, is even stronger than that of adjectives
to follow the noun. In some languages such as Arapesh in New
Guinea, 'The good man came' would be literally translated 'The
man is-good that-one he came'. Adjective-noun order, then, is
somewhat ambivalent since analogies with other constructions
involving modifiers make it indirectly harmonic with VS while
101
the factor of topic-comment order makes it analogous with SV.

All this is far from a complete theory. Nevertheless, it does
suggest that one should examine instances in which, contrary to
the prevailing rules, the genitive construction is disharmonic with
Pr/Po. One would reason that in such cases the genitive construc
tion is, as it were, being attracted by the adjective-noun construc
tion which, as has been seen, has sources of determination that
are to some extent outside of the general framework of harmonic
relations connected with the order of modifier and modified.
For example, if, in spite of the general rule, we find genitive
governing noun order with prepositions, the reason might be the
opposing pull of order adjective-noun which is harmonic with
genitive-governing noun. Otherwise stated, the genitive con
struction should only be disharmonic with Pr/Po when Pr/Po is
disharmonic with the adjective-noun order. One may include here
cases in which a language has two genitive orders, indicating a
probable change of type since one must, in all likelihood, be older
than the other. One may further conjecture that if there are
exceptions, they will be in type II, which, having both SV and va
which are disharmonic, can provide an anchor in either case for
deviant genitive order.
It will be noted that Universal 5, insofar as it refers to post
positional languages of type III (the vast majority), gives a par
ticular instance of this hypothesis; for this statement asserts that
a language of type III if it has NG will also have NA. If such a
language is postpositional, then NG will be disharmonic with
postpositions but harmonic with NA. Ifwe include languages with
both genitive orders, then there are at least six cases, all favorable
(Le., with NA rather than AN). These are Somali and Maba with
both genitive orders, and Kanuri, Galla, Teda, and Sumerian
which have SOY, postpositions, NG, and NA.
This hypothesis will, however, produce some further predic
tions. For prepositional languages of type III, the hypothesis
will be that with varying genitive order or with GN, which is
disharmonic with prepositions, the adjective-noun order will be
AN. I know of only two cases, Tigrinya with both genitive orders,
and Amharic with GN. Both have AN in accordance with our
,I
:t11
,;
',i.i
f," l
',\
I'
;?
102
Joseph H. Greenberg
hypothesis. For languages of type II which are prepositional and

which have GN, and should therefore have AN, we have Danish,
Norwegian and Swedish (possibly a single case), and English
with two genitive orders. Both fulfill the hypothesis in that they
have AN. Among postpositionallanguages oftype II, we have the
Moru-Madi group in the Sudan and the fairly distantly related
Mangbetu, both of which, with alternative genitive orders, have
the predicted NA. We now encounter the only exceptions of
which f am aware, Araucanian in Chile, with both genitive orders;
and a group of Daghestan languages in the Caucasus, including
some like Rutulian with NG, and others like Tabassaran with both
genitive orders. Apparently all those languages of the Daghestan
group which are of type III have only GN harmonizing with both
postpositions and AN. If so, this is an important indication of
the general validity of our hypothesis. Finally, since all languages
of type I are prepositional, we have only a single case to consider,
prepositional languages with GN. I know of only one example,
the Milpa Alta dialect of Nahuatl described by Whorf. It has AN
as expected.
Another type of relation than those that have just been con
sidered is illustrated by Universals 20 and 29. These may be called
proximity hierarchies. What we have is a rule that certain elements
must be closer to some central element than some other satellite.
The central element may be the root morpheme or base of a word
or the head-word of an endocentric construction. Such a proximity
hierarchy is likely to be related to an implicational hierarchy in
in the instance of inflectional categories. Just as the category of
number is almost always closer t6 the base than expressions of
case, so there are many languages with the category of number but
without the category of case, and very few with case but without
number. Since, by the proximity hierarchy, number is closer, it is
more likely to become amalgamated with the base and so become
an inflection. These hierarchies are presumably related to degrees
of logical and psychological remoteness from the center, but no
analysis is attempted here.
These phenomena are likewise related to those of neutralization.
The more proximate category, or the implied category, tends to be
103
more elaborate, and it is the less proximate or the implying cate

gories which tend to be neutralized in its presence. Universals 36
and 37 are related in this manner. Number is the implied category.
Gender categories are often neutralized in the marked number (i.e.
nonsingular). It is much rarer for number to be neutralized in some
particular gender (e.g., the neuter in Dravidian languages). With
regard to number and case, number is, as has been seen, more
proximate and generally present when case is present, while the
opposite relation holds far more rarely. It is likewise common for
certain case distinctions to be neutralized in number, while the
opposite phenomenon perhaps never occurs.
Another principle is evident from Universal 34. We do not
have such systems as the following: a particular grammatical
category for the trial, while another embraces the dual and all
numbers larger than three. In other words, disjunctiveness or
lack of continuity in this respect is never tolerated.
Universals 14 and 15 possibly illustrate the same principle.
The order of elements in language parallels that in physical ex
perience or the order of knowledge. In the instance of conditionals,
although the truth relations involved are timeless, logicians have
always symbolized in the order implying, implied exactly as in
spoken language. If modus ponens is used in proof, then we have
a pragmatic example which follows the order of reasoning. No
one thinks to write a proof backwards.
Universals 7, 8, and 40, although superficially very different,
seem to be examples of the same general tendency to mark the
end of units rather than the beginning. For example,. in rigid
subtype HI. the verb marks the end of the sentence. When the
inflections occur only with the final member of the noun phrase,
this marks the end of the phrase. This is probably related to the
fact that we always know when someone has just begun speaking,
but it is our sad experience that without some marker we don't
know when the speaker will finish.
The existence of a rigid subtype III, whereas there are no ex
amples of a rigid subtype of I, is probably related to still another
factor. In general the initial position is the emphatic one, and
while there are other methods of emphasis (e.g. stress). the initial
104
Joseph H. Greenberg
position always seems to be left free so that an element to which

attention is directed may come first. Here Universal 12 is an
example. It seems probable that in 'all languages expressions of
time and place may appear in the initial positions in the sentence.
The discontinuity of the predicate, which commonly appears in
such instances (e.g., German, 'Gestern ist mein Vater nach
Berlin gefahren'), illustrates a further principle. On the whole,
the higher the construction in an immediate constituent hierarchy,
the freer the order of the constituent elements. It has been seen
that practically all languages have some freedom of order regard
ing subject and predicate as a whole; whereas only a small mi
nority have variant order in genitive constructions, and then al
most always along with other differences, not merely a difference
of order. Within morphological constructions, order is the most
fixed of all. On the whole, then, discontinuous constituents are far
less frequent than continuous ones.
As indicated in the initial section of this paper, the principles
described in this section are to be viewed as no more than sug
gestive. It is hoped that some of them at least will prove useful
for further investigation.
ACKNOWLEDGMENTS
In addition to my indebtedness to the work of Roman Jakobson,
to be mentioned in the notes, I would also like to thank Fred House
holder and Charles F. Hockett for making helpful critical comments
on the earlier version of this paper.
Notes
1. I am intlebted to the work of Roman Jakobson for directing my
attention. to the importance of implicational universals.

2. See, for example, the remarks of R. B. Braithwaite, Scientific
Explanation (Cambridge, 1953), concerning scientific laws. "The
one thing upon which everyone agrees is that it always includes
a generalization, i. e., a proposition asserting a universal connec
tion between properties" (p. 9).
3. That is, empirically, not logically, implied. All languages are
observed to have the characteristics in question. It should be
105
added that universals in the sense of non definitional character
istics, if found only in language, do have the additional logical
property of implying as well as being implied by the definitional
properties.
4. Some of the ideas regarding the basic order typology are found in
nineteenth-century linguistic literature. For example, the relation
between genitive position and prepositions versus postpositions
and the hypothesis that some languages favor the order modifier
modified and others the opposite order is already a familiar
notion in R. Lepsius' introduction to his Nubische Grammatik
(Berlin, 1880).
The most systematic treatment is that of W. Schmidt in Die

Sprachfamilien lmd Sprachenkreise der rde (Heidelberg, 1926)
and in several other works. Schmidt's basic conclusions may be
summarized here. Prepositions go with nominative-genitive order
and postpositions with the reverse order. The nominative-genitive
order tends to appear with verb before nominal object and
genitive-nominative with object-verb. Schmidt says nothing of
subject-verb order so that types I and II as treated in this paper
are not distinguished. Further, nominative-genitive is associated
with noun-adjective and genitive-nominative with adjective-noun.
This last correlation, particularly the latter half, is much weaker
than the others. Schmidt gives figures based on a world sample
which show good general agreement with the results from the
thirty-language sample utilized here. It should be added that
Schmidt's chief interest in this topic is as a vehicle for the inter
pretation of culture history. His results there verge on the fantastic.
5. Siuslaw and Coos, which are Penutian languages of Oregon, and
Coeur d'Alene, a Salishan language, are exceptions.
6. The manner in which each language has been assigned can be
determined from the data of Appendix I.
7. For details, see Appendix II.

8. Iraqw, a southern Cushitic language, Khamti, a Thai language,
standard Persian, and Amharic.
9. The single case where it does not hold seems to be Amharic, which
has SOY, GN, and AN, but is prepositional.
10. However, Householder informs me that in Azerbaijani, and in
most types of spoken Turkish, it is allowable to have one modifier,
especially a dative or locative noun phrase after the verb.
11. Languages of type I-Berber, Hebrew, Maori, Masai, and Welsh;
II-Thai, Yoruba; III-Burmese, Burushaski, Japanese, Kannada,
Nubian. For Yoruba, see further note 12.
12. In the following languages the affix or particle follows: II-Finnish,
Guarani, Malay, Maya, Serbian; Ill-Basque, Burmese, Japanese,
!~,
~i
A
1J~.'
F ,;
'i~:
':l
f".
(:i
106
li:
Ii
Joseph H. Greenberg
Kannada, Nubian, Turkish, Quechua. It precedes in Yoruba, but

may be accompanied by a final particle.
13. The question word is first in Berber, Finnish, Fulani, Greek,
Guarani, Hebrew, Italian, Malay, Maori, Masai, Maya, Nor
wegian, Serbian, Welsh, Yoruba, and Zapotec.
14. Again, this only holds for literary Turkish, according to House
holder. See note 10.
15. Auxiliary precedes verb in Finnish, Greek, Italian, Masai, Maya,
Norwegian, Serbian, Swahili, Welsh, Zapotec. Auxiliary follows
verb in Basque, Burushaski, Chibcha, Guarani, Hindi, Kannada,
Nubian, Quechua, Turkish.
16. For details, see Appendix I.
17. Languages with adjective-noun and adverb-adjective order are
Burushaski, Finnish, Greek, Hindi, Japanese, Kannada, Maya,
Norwegian, Quechua, Serbian, Turkish. Languages with noun
adjective and adverb-adjective order are Basque, Burmese,
Chibcha, Italian, Loritja. Languages with noun-adjective and
adjective-adverb order are Fulani, Guarani, Hebrew, Malay,
Swahili, Thai, Yoruba, and Zapotec. Languages with noun
adjective and the rule that certain adverbs precede and certain
follow the adjective are Maori and Welsh. Berber, Masai, Nubian,
and Songhai-no data.
18. Languages with adjective-marker-standard are Berber, Fulani,
Greek, Hebrew, Italian, Malay, Maori, Norwegian, Serbian,
Songhai, Swahili, Thai, Welsh, Zapotec. Languages with standard
marker-adjective are Basque, Burmese, Burushaski, Chibcha,
Guarani, Hindi, Japanese, Kannada, Nubian, Turkish. Both
constructions are found in Finnish.
19. Languages with common noun-proper noun are Greek, Guarani,
Italian, Malay, Serbian, Swahili, Thai, Welsh, Zapotec. Those
with proper noun--common noun are ijasque, Burmese, Burushaski,
Finnish, Japanese, Norwegian, Nubian, and Turkish.
20. The relational expression precedes the noun in Basque, Burmese,
Burushaski, Chibcha, Japanese, Kannada, Turkish. The noun
precedes the relational expression in Berber, Fulani, Greek,
Guarani, Hebrew, Hindi, Italian, Malay, Maori, Masai, Maya,
Norwegian, Quechua, Serbian, Songhai, Swahili, Thai, Welsh,
Yoruba, Zapotec. Both orders are found in Finnish and Nubian.
In Finnish the construction with the relational expression pre
ceding the noun is in imitation of literary Swedish (personal com
munication of Robert Austerlitz).
21. The exclusively suffixing languages are Basque, Burmese, Chibcha,
Finnish, Hindi, Japanese, Kannada, Loritja, Nubian, Quechua,
Songhai, Turkish.
107
22. The reason for specifying order is that there are instances of
neutralization of number agreement in which the order of the
item is not involved. For example, in classical Greek the neuter
plural goes with a singular verb without regard to order.
Additional note: The following facts were learned too late to be

included in the paper. According to information supplied by Einar
Haugen, Norwegian has both genitive orders. Note that Norwegian
had been the only exception in the sample to the generalization on
p. 64. In a discussion at the International Congress of Linguistics at
Cambridge in August 1962, it was pointed out that Papago, a Uto
Aztecan language, is I/Po. This is therefore an exception to Universal 3.
From Mason's data it should probably be assigned to type 7 of
Appendix II.
Appendix I
Basic Data on the 30-Language Sample
Pr
VSO
Basque
Berber
Burmese
Burushaski
Chibcha
Finnish
Fulani
Greek
Guarani
Hebrew
Hindi
Italian
Kannada
Japanese
Loritja
Malay
Maori
Masai
Maya
Norwegian
Nubian
Quechua
III
I
III
III
III
II
II
II
II
I
III
II
III
III
III
II
I
I
II
II
NA
ND
x
x
Xl
x
x
x
III
III
NNum
-2
X
x
x
x3
_2
x
x
x
x
x
x
x
x
x
x
X
_2
X
_2
~,
108
Serbian
Songhai
Swahili
Thai
Turkish
Welsh
Yoruba
Zapotec
Joseph H. Greenberg
VSO
Pr
II
II
H
II
HI
x
x
II
x
x
NA
ND
NNum
x
x
-2
x3
x
x
x
x
In the first column, I indicates that normal word order is verb

subject-object, ([ indicates subject-verb-object, and III subject-object
verb. In the second column, x indicates that the language has preposi
tions, and - that it has postpositions. In the third column, x indicates
that the noun precedes its modifying adjective, and - that it follows.
In the fourth column, x indicates that the noun precedes its modifying
demonstrative, and that it follows. In the fifth column, x indicates
that the noun precedes its modifying numeral, and - that it follows.
In any column, 0 means that both orders are found.
Notes to Appendix I
1. Participle of adjective-verb, however, precedes and is probably as
common as adjective following.
2. Numeral classifiers following numerals in each case. The construc
tion numeral + classifier precedes in Burmese and Maya, follows
in Japanese and Thai, and either precedes or follows in Malay.
3. In Welsh and Italian a small number of adjectives usually precede.
Appendix II
Distribution of Basic Order Types;
1. I/Pr/NG/NA. Celtic languages; Hebrew, Aramaic, Arabic, Ancient

Egyptian, Berber; Nandi, Masai, Lotuko, Turkana, Didinga;
Polynesian languages and probably other Austronesian languages;
Chinook, Tsimshian; Zapotec, Chinantec, Mixtec, and probably
other Oto-Mangue languages.
2. I/Pr/NG/AN. Tagabili and probably other Philippine Austronesian
languages; Kwakiuti, Quileute, Xinca.
109
3. I/Pr/GN/AN. Milpa Alta Nahuatl.

4. I/Pr/GN/NA. No examples.
5. I/Po/NG/NA. No examples.
6. I/Po/NG/AN. No examples.
7. I/Po/GN/AN. No examples.
8. I/Po/GN/NA. No examples.
9. I1/Pr/NG/NA. Romance languages, Albanian, Modern Greek;
West Atlantic languages, Yoruba, Edo group, most languages
of Benue-Congo group including all Bantu languages; Shill uk,
Acholi, Bari, most languages of Chad group of Hamito-Semitic
but not Hausa; Neo-Syriac, Khasi, Nicobarese, Khmer, Vietna
mese, all Thai languages except Khamti; many Austronesian
languages including Malay; Subtiaba.
10. Il/Pr/NG/AN. German, Dutch, Icelandic, Slavonic, Efik, Kredj,

Maya, Papiamento.
11. I1/Pr/GN/AN. Norwegian, Swedish, Danish.
12. II/Pr/GN/NA. Arapesh (New Guinea).
13. II/Po/NG/NA. No examples.
14. II/Po/NG/AN. Rutulian and other Daghestan languages in the
Caucasus.
15. II/Po/GN/AN. Finnish, Estonian, ljo, Chinese, Algonquian
(probably), Zoque.
16. I1/Po/GN/NA. Most Mandingo and Voltaic languages, Kru, Twi,
Ga, Guang, Ewe, Nupe, Songhai, Tonkawa, Guarani.
17. 1II/Pr/NG/NA. Persian, Iraqw (Cushitic), Khamti (Thai), Akka
dian.
18. llI/Pr/NG/AN. No examples.
19. III/Pr/GN/AN. Amharic.
20. I1I/Pr/GN/NA. No examples.
21. I1I/Po/NG/NA. Sumerian, Elamite, Galla, Kanuri, Teda, Kamila
roi and other southeastern Australian languages.
22. lll/Po/NG/AN. No examples.
23. III/Po/GN/AN. Hindi, Bengali, and other Aryan languages of
India; Modern Armenian, Finno-Ugric except Finnish group;
Altaic, Yukaghir, Paleo-Siberian, Korean, Ainu, Japanese, Gafat,
Harari, Sidamo, Chamir, Bedauye, Nama Hottentot; Khinalug,
Abkhaz and other Caucasian languages; Burushaski, Dravidian;

Newari and other Sino-Tibetan languages; Marind-Anim, Navaho,
Maidu, Quechua.
24. I1I/Po/GN/NA. Basque, Hurrian, Urartian, Nubian, Kunama,
Fur, Sandawe, Burmese, Lushei, Classical Tibetan, Makasai,
Bunak (Timor), Kate (New Guinea), most Australian languages,
Haida, Tlingit, Zuni, Chitimacha, Tunica, Lenca, Matagaipa,
Cuna, Chibcha, Warrau.
~i:d
',I
:\
;~
';'i',j;l
.:,..1
,ill
'I'
"r!
. !I
"~!H
).::1
i~ ~~
'I ~
:!~;
,,~.
:1~
110
Joseph H. Greenberg
Languages with Object before Subject:
Coeur d'Alene: VOS/Pr/NG/NA.

Siuslaw, Coos: VOS and OVS/Po/GN/AN/
Lal1lfuages with Variant Constructions:
Geez, Bontoc Igorot I, 2; Tagalog I, 2, 3, 4; Sango 9, 10; English

II; Lithuanian II, 15 (prepositions more numerous); Mangbetu,
Araucanian 12, 13; Takelma 12, 16 (prepositions more frequent);
Moru-Madi 13, 16; Tabassaran 14, 15; Luiseiio 15, 16; Tigre 17, 18,
19,20; Tigrinya 18, 19; Somali, Maba 21, 24; Afar, Ekari 23, 24.
Appendix III
l1niversals Ftestated
I. In declarative sentences with nominal subject and object, the domi
nant order is almost always one in which the subject precedes the
object.
2. In languages with prepositions, the genitive almost always follows
the governing noun, while in languages with post positions it
almost always precedes.
3. Languages with dominant VSO order are always prepositional.
4. With overwhelmingly greater than chance frequency, languages
with normal SOY order are postpositional.
5. If a language has dominant SOY order and the genitive follows
the governing noun, then the adjective likewise follows the noun.
6. All languages with dominant VSO order have SVO as an alternative
or as the only alternative basic order.
7. If in a language with dominant SOVorder there is no alternative
basic order, or only OSV as the alternative, then all adverbial
modifiers of the verb likewise precede the verb. (This is the
"rigid" subtype of Ill.)
8. When a yes-no question is differentiated from the corresponding
assertion by an intonational pattern, the distinctive intonational
features of each of these patterns is recleoned from the end of the
sentence rather than the beginning.
9. With well more than chance frequency, when question particles or
affixes are specified in position by reference to the sentence as a
whole, if initial, such elements are found in prepositional languages
and, if final, in postpositional.
III
to. Question particles or affixes, specified in position by reference to

a particular word in the sentence, almost always follow that
word. Such particles do not occur in languages with dominant
order VSO.
11. Inversion of statement order so that verb precedes subject occurs
only in languages where the question word or phrase is normally
initial. This same inversion occurs in yes-no questions only if it
also occurs in interrogative word questions.
12. If a language has dominant order VSO in declarative sentences, it
always puts interrogative words or phrases first in interrogative
word questions; if it has dominant order SOY in declarative
sentences, there is never such an invariant rule.
13. If the nominal object always precedes the verb, then verb forms
subordinate to the main verb also precede it.
14. In conditional statements, the conditional clause precedes the
conclusion as the normal order in all languages.
15. In expressions of volition and purpose, a subordinate verbal
form always follows the main verb as the normal order except in
those languages in which the nominal object always precedes the
verb.
16. In languages with dominant order VSO, an inflected auxiliary
always precedes the main verb. In languages with dominant order
SOY, an inflected auxiliary always follows the main verb.
17. With overwhelmingly more than chance frequency, languages
with dominant order VSO have the adjective after the noun.
18. When the descriptive adjective precedes the noun, the demonstra
tive and the numeral, with overwhelmingly more than chance
frequency, do likewise.
19. When the general rule is that the descriptive adjective follows,
there may be a minority of adjectives which usually precede, but
when the general rule is that descriptive adjectives precede, there
are no exceptions.
20. When any or all of the items-demonstrative, numeral, and
descriptive adjective-precede the noun, they are always found
in that order. If they follow, the order is either the same or its
exact opposite.
21. If some or all adverbs follow the adjective they modify, then the
language is one in which the qualifying adjective follows the noun
and the verb precedes its nominal object as the dominant order.
22. If in comparisons of superiority the only order or one of the alter
native orders is standard-marker-adjective, then the language is
postpositional. With overwhelmingly more than chance frequency,
if the only order is adjective-marker-standard, the language is
prepositional.
ll2
Joseph H. Greenberg
23. If in apposition the proper noun usually precedes the common

noun, then the language is one in which the governing noun
precedes its dependent genitive. With much better than chance
frequency, if the common noun usually precedes the proper
noun, the dependent genitive precedes its governing noun.
24. If the relative expression precedes the noun either as the only con
struction or as an alternative construction, either the language is
postpositional or the adjective precedes the noun or both.
25. If the pronominal object follows the verb, so does the nominal
object.
26. If a language has discontinuous affixes, it always has either pre
fixing or suffixing or both.
27. If a language is exclusively suffixing, it is postpositional; if it is
exclusively prefixing, it is prepositional.
28. If both the derivation and inflection follow the root, or they both
precede the root, the derivation is always between the root and the
inflection.
29. If a language has inflection, it always has derivation.
30. If the verb has categories of person-number or if it has categories
of gender, it always has tense-mode categories.
31. If either the subject or object noun agrees with the verb in gender,
then the adjective always agrees with the noun in gender.
32. Whenever the verb agrees with a nominal subject or nominal
object in gender, it also agrees in number.
33. When number agreement between the noun and verb is suspended
and the rule is based on order, the case is always one in which the
verb is in the singular.
34. No language has a trial number unless it has a dual. No language
has a dual unless it has a plural.
35. There is no language in which the plural does not have some
nonzero allomorphs, whereas there are languages in which the
singular is expressed only by zero. The dual and the trial are almost
never expressed only by zero.
36. If a language has the category of gender, it always has the category
of number.
37. A language never has more gender categories in nonsingular
numbers than in the singular.
38. Where there is a case system, the only case which ever has
zero allomorphs is th one which includes among its meanings
that of the subject of the intransitive verb.
39. Where morphemes of both number and case are present and both
follow or both precede the noun base, the expression of number
almost always comes between the noun base and the expression
of case.
113
40. When the adjective follows the noun, the adjective expresses all the
inflectional categories of the noun. In such cases the noun may
lack overt expression of one or all of these categories.
41. If in a language the verb follows both the nominal subject and
nominal object as the dominant order, the language almost
always has a case system.
42. All languages have pronominal categories involving at least three
persons and two numbers.
43. If a language has gender categories in the noun, it has gender
categories in the pronoun.
44. If a language has gender distinctions in the first person, it always
has gender distinctions in the second or third person or in both.
45. If there are any gender distinctions in the plural of the pronoun,
there are some gender distinctions in the singular also.
UNIVERSALS IN DIACHRONIC MORPHOLOGY

CHAPTER 6
A SEARCH FOR UNIVERSALS IN INDO-EUROPEAN

DIACHRONIC MORPHOLOGY
WARREN COWGILL
Yale University.
!I
In searching for universals of diachronic morphology, three

approaches come easily to mind. One is to make typological
studies of the morphologies of several sets of genetically related
languages, and then compare the typologies to see what similarities
in development are shown by related languages, and whether there
are any general rules that apply to all the languages studied regard
less of genetic relationship.
Two kinds of similarities can be looked for: similarities in
change and similarities in retention. Looking for similarities in
change among related languages should be useful in testing
the idea of drift-that is, whether or not the imbalances and
stresses of the common protolanguage inevitably lead to simi
larities in the innovations made by the daughter languages, and
what parts of the grammar are most likely to undergo such paral
lel change. Looking for similarities in change among unrelated
languages would help to determine whether there are any general
rules for the evolution of typologies-for example, would any
language similar in type to Proto-Indo-European or Proto-Uto
Aztecan show much the same evolution as Indo-European and
Uto-Aztecan have, or are the structural differences determining
the future development of a linguistic system too subtle to be
caught by any typology yet devised?
Looking for features retained with little change over a long
period by all or most of a set of related languages should reveal
114
115
whether any features of their common protolanguage were more

resistant to change than others, and so in some sense more basic
to its structure. If there should prove to be such features, one could
then look to see whether or not these belong to the same areas of
structure in all language families. The results, even if purely
negative, should be of interest to psychologists.
A slightly different approach is to compare the changes that
have taken place rather than a selection of the states that the
language assumes as it undergoes the effect of these changes in
its progress through time. That is, instead of stating that lan
guage A, spoken in 500 B.C., has prefixes and grammatical gen
der, while its daughter language B, spoken in 1500 A.D., has
neither, we ask what has become of the prefixes and gender
markers of A during the intervening two millennia, and then check
to see if any other language we know of has treated its prefixes or
gender markers in the same way. If enough data of this type could
be accumulated, it might become possible to predict the life
expectancy of a particular grammatical feature and the relative
probability of several alternative ways in which it might be
transformed, replaced, or lost without replacement.
More useful, perhaps, for the practicing historical linguist, but
also more laborious and exacting would be to look not at the gross
over-all changes in structure but rather at the individual, often
minute, innovations whose cumulation over centuries results in
the gross changes which can be measured by typology. A suffi
ciently large collection of such individual changes, appropriately
classified, should give linguists a measure of the relative plausi
bility of different solutions for problems in historical grammar. A
change which is recorded as having occurred 20 times will be more
plausible than one recorded only 5 times, and both will be more
plausible than one for which no parallel can be found at all.
At present each linguist judges the plausibility of a newly pro
posed solution pretty much by what he happens to remember of
the morphologic innovations which during his career he has been
led, for one reason or another, to accept as plausible. A reasonably
objective standard of plausibility should make it easier for
historical linguists to agree on solutions for problems of historical
116
Warren Cowgill
morphology that at present are still disputed, or, in the case of

many non-Indo-European languages, not yet even attacked.
However, the collection and classifying of parallel morphologic
changes would not be easy. The large number of individual cases
to be treated would make the work laborious and time-consuming;
the decision as to what constitutes an individual case would often
be delicate; the construction and application of a system of
classification might prove difficult; and the lack of an agreed-on
criterion of plausibility to begin with would bias the results in
favor of the investigators' preconceived ideas about linguistic
change.
117
typology is to determine the minimum size of sample needed to

be reasonably sure of having an accurate picture of a language.
An expanded and refined list of indices, perhaps along the lines
suggested by Voegelin, Ramanujan, and Voegelin (IJAL 26.198
205) would no doubt reveal more than the 10 used by Greenberg.
However, it seemed outside the scope of this paper to get involved
in trying to devise a new typology; besides, I wanted to be able
to use the figures already calculated for Indo-European languages
by Greenberg.
The Indo-European languages for which Greenberg gives
indices are Classical Sanskrit (a passage from the Hitopadda, date
of compilation unknown but certainly before the fourteenth
century A.D.); Old English (a passage from the Alfredian transla
tion of Boethius, about 900 A.D.); Modern English (a passage
from the New Yorker, 1952 A.D.); Modern Persian (a passage
from a chrestomathy published in 1889 A.D.). 1 I abbreviate these
as CISk., OE, NE, and NP. I have taken unchanged the figures
given in the table on page 193 of Greenberg's article (IJAL 26),
without trying to repeat his calculations, although discrepancies
the totals of some indices which, according to my understanding
of the method, ought to add up to identical figures suggest that
there are some inaccuracies. In the following I have marked
figures taken over from Greenberg with an asterisk.
As an experiment in looking for universals by comparing

typologies, I have compared the typologies of 14 Indo-European
languages, using the 10 indices described by Joseph H. Greenberg
in International Journal of American Linguistics 26.178-194 (1960;
reprinted from Method and Perspective in Anthropology: Papers in
Honor of Wilson D. Wallis, 1954).
As I am well aware, my study is of very limited value except
as a preliminary exploration. The lack of data from non-Indo
European languages means that any universal or near-universal
tendencies of change or retention found may be valid only for
Indo-European. Like Henry Ford and the color scheme of the
Model-T, we can say that any uniformities we observe will be
true of the descendants of any language, provided that language
happens to be Proto-Indo-European. But not even this much
generalization is possible, since the failure to include samples from
many of the important branches of Indo-European leaves it
uncertain how much of the regularity observable in our samples is
true of all Indo-European. At least my figures have some negative
value: any failure of the languages sampled to develop alike will
a fortiori be true of the world's languages as a whole.
Another weakness of the material used here is the small size
of the samples, 100 words of text from each language. Although I
am ignorant of statistical methods, I strongly suspect that many
of the indices calculated from such short samples have little
statistical validity. One of the urgent desiderata of current
The ten languages which I myself sampled were:

1. Vedic Sanskrit (abbreviated RV), Rig-Veda 1.32 from the
beginning (indrasya) to asya in stanza 6. Since Greenberg's
sample of Sanskrit comes from a period of the language differing
significantly in morphology and syntax from Vedic, I thought it
would be worth while to put beside it a specimen of genuinely old
Sanskrit and see what differences, if any, would be revealed. Al
though Rig-Veda 1.32 is not one of the oldest of Vedic hymns, it
seems old enough to be a fair sample of the earliest available lndo
Aryan. Its date is probably somewhere in the first half of the first
millennium B.C.
!t
118
Warren Cowgill
2. Asoka's Rock Edicts (Asoka), in the version of Girnar, from

the beginning of Edict [ (iytlf!1) to the end of the fourth line of
Edict II (kata), following the edition of Jules Bloch, Les in
scriptions d' Asoka (Paris, 1950). This was chosen as a sample
of early Middle [ndic, securely datable to the middle of the
third century B.C.
3. Bengali (Bg.), a passage from the 1862 edition of Kali
prasanna Sinha's Hutom penchar naksha as printed in S. K.
Chatterji's Bengali Phonetic Reader (London, 1928), from chele
biEla (page 73) to si!i in line 6 of page 75. I chose Bengali to
represent modern (ndo-Aryan because from my slight acquaint
ance with modern lndic I suspected that Bengali had diverged
more than most of its sister languages from ancient (ndic and that
it would be relatively easy to learn enough about it to make the
judgments needed for a typological analysis.
4. Old Persian (OP), the first column of Darius I's Bisotun
inscription, from 6iitiy in line 24 to pasii- in line 36, following the
edition of R. G. Kent, Old Persian 117 (second edition; New
Haven, 1953). This text is securely dated to the last two decades
of the sixth century B.C. Greenberg had already calculated indices
for Modern Persian, and therefore as a representative of ancient
Iranian I took Old Persian, the dialect closest to being a direct
ancestor of Modern Persian.
S. Cuneiform Hittite (Hitt.), The Apology of Hattusilis, tablet [,
column I, from the beginning of line 9 (A-BU- YA) to -as- in
line 21, as given in E. H. Sturtevant's Hittite Chrestomathy 42-45
(Philadelphia, 1935). This text belongs to the thirteenth century
B.C. Although Hittite texts older by several centuries are available,
my experience with the language leads me to doubt that they
would show up very different typologically from the sample here
studied. It should be noted that the readings of some of the
Hittite words written logographically are unknown, so that
sometimes I have had to guess about the number of morphemes
in a word and whether they were combined agglutinatively or not.
6. Homeric Greek (Hom.), Iliad 1.22 (enth') to para in line 34,
as printed by Walter Leaf in his second edition of the Iliad (Lon-
119
don, 1902). This is possibly from one of the older parts of the
Homeric Epic, and is probably not later than the early eighth cen
tury B.C. Since the fragments of Mycenaean provide no texts
usable for computing typological indices, Homer constitutes the
earliest available Greek.
7. New Testament Greek (NT), the seventh chapter of Luke
from the beginning (epeide) to gar in verse 6, using the text of A.
Souter, Novum testamentum graece (Oxford, 1910). This text,
dating from the first century A.D., seemed a convenient inter
mediate point between Homer and Modern Greek, and also
offered opportunity for comparison of the same passage in Gothic
and Old Church Slavic.
8. Modern Greek (NGk.), from the Khamena logia (1888) of
Jean Psichari, as printed by Albert Thumb, Handbuch der neu
griechischen Volksprache 254 (second edition; Strassburg, 1910).
The specimen starts at the beginning of the second paragraph
(apheste) and goes as far as pia in the third line from the end of the
paragraph.
9. Gothic (Go.), Luke 7.1 (bi) to uf in verse 6, corresponding
(except for a few words at the end) to the sample of New Testa
ment Greek. The Gothic translation of the Bible was made in the
fourth century A.D., and so provides a specimen of Germanic
about SOO years older than the Old English text studied by Green
berg. I have followed the readings of Wilhelm Streitberg, Die
gotische Bibel (second edition; Heidelberg, 1919).
10. Old Church Slavic (OCS), Luke 7.1 (egda) to sebe in verse
7, using the text of the Zographensis manuscript as printed by A.
Leskien, Handbuch der altbulgarischen Sprache (third edition;
Weimar, 1902). The Slavic translation of the Bible, made in the
latter part of the ninth century A.D., is the oldest available Balto
Slavic.
Typological studies of several more languages would have been
needed to provide an adequate sampling of Indo-European as a
whole. Particularly desirable additions would have been Latin
(with samples of early and late Latin, for example Plautus and
120
Warren Cowgill
Jerome) and one or more modern Romance languages; some

Celtic, say Old, and Modern Irish and one of the British lan
guages; a modern Slavic language; Lithuanian; Albanian; Clas
sical Modern Armenian; and Tocharian.
Something needs to be said about the methods 1 have used in
calculating the indices. In general I have not tried to make com
plete and rigorous grammatical analyses, but rather have pro
ceeded very often by rule of thumb in deciding on word, and
morpheme boundaries and classifying morphemes into radical,
derivational, and inflectional. Of the ten languages studied, Ben
gali and Modern Greek are the least familiar to me, and it is here
that I am most likely to have made mistakes out of ignorance.
Since I have not made counts of my own for the four languages
studied by Greenberg, I do not know how much my personal way
of calculating indices differs from his; however, the fairly close
agreement between most of his figures for Classical Sanskrit and
mine for Vedic suggests that our results for other languages
might not be very divergent.
In the, matter of word boundaries I have probably been over
influenced by whether or not a space was left between letters by
the editor of the text I was using. Thus I have considered tha a
word in Modern Greek, although I suspect that more information
on its privileges of occurrence might show it to be in fact a ver
bal prefix. \I In general, I have tried to apply Greenberg's rule
(IJAL 26.192) that 'a nucleus boundary is a word boundary if it
is possible to insert an indefinitely long sequence of nuclei.'
However, I have not followed Greenberg in considering a se
quence like Latin dominusque a single word because the accent of
dominus- differs from the accent of dominus in other positions.
Greenberg here seems to be introducing a phonologic criterion
into what is really a grammatical problem. Hence I have con
sidered enclitics as independent words, perhaps more liberally than
I should have.
In segmenting words into morphemes I have also differed from
Greenberg (IJAL 26.189) in not segmenting unless at least one of
the putative morphs existed elsewhere with a meaning similar to a
part of the total meaning of the sequence in question. In practice,
121
I think this has made a difference only in the case of etymolog

ically polymorphemic sequences which as the result of semantic
change have come to have a meaning not containing the meaning
of any of their parts.
In segmenting inflectional morphemes I have sometimes found
it hard to decide where to posit rather long morphemes with two
or three semantic components and where to make use of partial
resemblances between forms to segment out shorter morphemes
with fewer semantic components. In general, I have not set up
separate number morphemes (except in Bengali), but have con
sidered number as combined into single morphemes with the
case markers of nouns and the person markers of verbs. In
Vedic, Asokan, Homeric, and New Testament Greek, I set up
feminine gender markers as morphemes; in Old Persian, Modern
Greek, and Gothic, gender seemed best treated as one component
along with case and number in the noun-inflecting morphemes;
in Bengali and Hittite, this problem did not arise, and the Old
Church, Slavic sample contained no feminines. In Vedic and
Hittite, mediopassive verb endings were analyzed as two mor
phemes each, the active ending plus a middle voice marker; in
other languages with a distinction of voice, the mediopassive
endings were considered single morphemes. In Hittite, Vedic,
and Homeric, where a difference in the personal ending is some
times the sole mark distinguishing present from imperfect tense,
it had to be decided whether the present endings were different
morphemes from the imperfect endings or were composed of per
sonal endings plus a non past morpheme; only in the case of Hittite
did the structure of the language seem clearly to warrant segmen
tation.
I have posited zero morphs, but probably fewer than many
linguists would. Thus I have not analyzed verbs consisting only
of root and personal ending as containing a zero tense formant,
or nouns consisting only of root and case-number ending as con
taining a zero noun-deriving suffix. Present stems of denomi
native verbs I take to have only a single derivational morpheme,
not a derivational morpheme plus zero present-stem formant. On
the other hand, I have considered vocatives and second singular
122
Warren Cowgill
active imperatives as containing zero case-number or person

mood endings even where there was no overt vocative or second
singular imperative ending in the language.
For agglutination my indices are more likely to be too high
than too low, since it is more likely that I forgot (or was ignorant
of) non predictable allomorphs than that I failed to see how
a set of allomorphs could be so described as to be automatically
predictable.
In classifying morphemes into roots, derivational affixes, and
inflectional affixes, two main problems arose. The first involves
the status of morphemes which are uninflected when they are
separate words-so-called prepositions and adverbs. When they
are parts of larger words, it was not clear to what extent these
should be considered derivational prefixes and to what extent the
first roots of compound words. On the one hand, their numbers
are limited, and their meanings are usually not concrete, which
according to Greenberg (IJAL 26. 191) makes them prefixes. On
the other hand, the~e morphemes act like noun stems in Some of
the more archaic Indo-European languages in that they combine
into single words with following noun stems but do not do so
with verb stems. I have, therefore, decided to treat Vedic and
Homeric words like sam-udrd and eu-pHemos as compounds; in
other languages, where these morphemes combine with verb stems
as well as with noun stems, I have considered them prefixes.
As a result, in all of the ten languages studied by me there are
compound nouns (and verbs derived from them), but compound
verbs only in Bengali.
The second problem in classifying morphemes was to draw
the line between inflection and derivation. Greenberg's rule here
is: 'Derivational morphemes [are] morphemes which, when in
construction with a root morpheme, establish a sequence which
may always be substituted for some particular class of single mor
pheme '" without producing a change in the construction' (IJAL
26.191). Although this is probably a good criterion for arriving
at a useful classification of the affixes of the world's languages in
general, I feel that Indo-European affixes can be better classified
somewhat differently. In Proto-Indo-European there is a contrast
123
between derivational suffixes and inflectional endings. The latter

include only case-number morphemes in the nouns, and person
number-voice-nonpast-imperative morphemes and augment in the
verbs. All the rest of Proto-Indo-European morphology is deri
vation, including feminine gender markers and the formants of
tense-aspect and modal stems and participles. Greenberg's rule
would probably lead to much the same classification.
But one of the characteristic developments shared by all Indo
European languages since about 500 B.C. is that gender, tense
aspect, mood, infinitive, and participle-forming affixes have come
to be sufficiently ubiquitous and paradigmatic to justify a change
of status which within the frame of Greenberg's typology seems
best describable by saying that they have become. inflectional. I
am not at all sure whether Greenberg'S essentially syntactic
criterion for distinguishing inflection and derivation would lead to
this result.
My treatment of these morphemes was as follows: everywhere
that I recognized separate feminine gender morphemes I con
sidered them inflectional, perhaps wrongly. In Vedic, Old Per~
sian, and Hittite, tense-aspect, mood, and participial formants
were assigned to derivation. In Homeric, I treated these affixes and
infinitive formants as inflectional, although hesitantly. In Asokan,
Bengali, New Testament Greek, Modern Greek, Gothic, and Old
Church Slavic, they belong clearly to inflection.
In Bengali, where there is no grammatical number concord, the
plural formants of nouns were considered derivational.
The counting of prefixes and suffixes presented no problems
once the words had been segmented into morphemes and it had
been decided what to count as prefixes and what as the first ele
ment of compound words. Since the Vedic sample contained three
infixes, I added to Greenberg'S ten indices an index of infixes per
word (Inf/W), numbered 6a.
Similarly in counting nexuses and assigning them to isolating,
pure inflection, and concord I experienced little trouble. How
ever, my judgments of the number of nexuses were all ad hoc and
not rechecked, so that I may have been inconsistent in analyzing
different sentences of similar structure. Regarding the assignment
""I
Warren Cowgill
124
of nexuses to the three categories, I considered conjunctions,

adverbs, and prepositional phrases as connected outwardly by
isolating constructions; the relation of verb or preposition to its
object as pure inflection, except in Bengali, where the relation to an
object without case ending is isolating; and that of verb to subject
as concord. As a result of rounding off to the nearest hundredth,
the totals of these three syntactic indices are sometimes 1.0 I
instead of 1.00.
TABLE I
10
8
9
60
7
6
3
4
5
I
2
M/W All R/W D/W I/W P/W Inf/W S/W O/N Pi/N CoIN
.03 1.24 .26
.97 .19
2.56 .08 1.10 .49
RV
1.18 .16
.84 .16
*CISk.
2.59 .09 1.13 .62
.00 1.23 .40
.86 .07
Asoka
2.52 .26 1.22 .44
.80 .57
.00
.53 .01
1.90 .46 1.09 .28
Bg.
.00 1.20 .23
.98 .19
2.41 .20 1.02 .41
OP
.49 .52
1.52 .34 1.03 .10 .39 .01
*NP
.94 .35
.00
1.95 .42 1.00 .24 .71 .01
Hitt.
.00 1.00 .48
.85 .06
2.07 .10 1.01 .21
Hom.
.00 1.24 .34
2.45 .12 1.03 .28 1.14 .18
NT
.77 .53
.00
.68 .03
1.82 .40 1.02 .12
NGk.
.00 1.19 .37
2.31 .19 1.03 .30 .98 .09
Go.
1.03 .15
.90 .06
2.12 .11 1.00 .20
OE
.64 .75
.53 .04
1.68 .30 1.00 .15
*NE
.00 1.17 .41
.95 .12
2.29 .20 1.00 .34
OCS
Figures for these languages were calculated by Greenberg.
.48
.46
.18
.29
.39
.29
.32
. 27
.32
.21
.34
.47
.14
.33
.27
.38
.42
.14
.38
.19
.33
.26
.34
.26
.29
.38
.II
.26
The meanings of the index labels in this table are:

M/W:
morphemes per word (measuring synthesis)
All:
agglutinative intraword morph junctures divided by the total
number of intra word morph junctures
roots per word (measuring compounding)
R/W:
derivational morphemes per word
D/W:
inflectional morphemes per word
I/W:
prefixes per word
P/W:
infixes per word
Inf/W:
suffixes per word
S/W:
isolating constructions divided by the total number of grain
O/N:
matical nexuses
pure inflectional constructions divided by nexuses
Pi/N:
concordial constructions divided by nexuses
CoiN:
125
Table I presents the reSUlting indices, including those of the

four Indo-European languages calculated by Greenberg.
The first two indices, M/W and All, can be said to reflect the
over-all morphologic complexity of a language; the higher the
M/W index and the lower the A/J index, the more complicated
the morphology will be. According to a widespread view, Indo
European languages have generally been getting less synthetic and
more agglutinative as time goes by. Rearranging the first two
columns of Table I, with M/W indices in descending and All
indices in ascending order, will help to show how far this view'
is borne out by our material.
TABLE 2
All
M/W
*CISk.
I.
2.
RV
3.
Asoka
4.
NT
OP
5.
6.
Go.
7.
OCS
8.
OE
9.
Hom.
Hitt.
10.
Bg.
II.
12.
NGk.
*NE
13.
14.
*NP
* Languages calculated by
2.59
2.56
2.52
2.45
2.41
2.31
2.29
2.12
2.07
1.95
1.90
1.82
1.68
1.52
Greenberg.
RV
"'C\Sk .
Hom.
"'OE
NT
Go.
}
.08
.09
.10
.11
.12
.19
OP,OCS .20
Asoka
"'NE
*NP
NGk.
Hitt.
Bg.
.26
.30
.34
.40
.42
.46
In both columns of Table 2, there is considerable correlation

between time and position in the table. In both. the four modern
languages are at or near the bottom, and in both Sanskrit is at
the top. The differences between Vedic and Classical Sanskrit
turn out to be small and probably not significant, suggesting thal
in this area of its grammar Classical Sanskrit was effectively
stabilized at a level not measurably advanced over Vedic.
But between the extremes both lists show several departures
from chronological order. Hittite, the earliest attested of the
:1:1
~~;
126
I!
Warren Cowgill
14, is very far out of place in both columns, being tenth in syn
thesis and thirteenth in agglutination. This aberrancy has long
been noted and discussed by Indo-Europeanists. According
to one view, the Anatolian branch of Indo-European (to which
Hittite belongs) separated from the rest of the Indo-European
family before the development of some of the complications
common to or presupposed by all the other Indo-European lan
guages. Others think the difference is due to extremely rapid
evolution of Hittite, perhaps influenced by speakers of some lan
guage with a simpler or radically different morphology. My own
opinion is that although the low index of synthesis may be partly
an archaism (which will be discussed later), the high index of
agglutination is almost certainly largely the result of innovation,
and much of the analytic tendency of the language may be innova
tion also. Indeed, I would even hazard the speculation that Hittite
may be a Creole, that is, the descendant of a pidgin originally
used for communication between speakers of Indo-European and
non-Indo-European languages in Anatolia. 3
The MjW list shows two other departures from chronological
order. Old Persian follows instead of preceding Asokan and New
Testament Greek. If significant, this position probably reflects the
relatively rapid linguistic evolution which resulted in Old Persian
being no longer clearly understood less than two centuries after
the time of Darius: the inscriptions of the late Achaemenid kings
show a confusion in grammar and spelling which indicates that
the language they spoke was already essentially Middle Persian.
I do not know the reasons for this rapid evolution. However, it
seems significant that later in its.history, for about the last
thousand years, the evolution of Persian seems to have been quite
slow." This invites speculation. Is the structure of Modern
Persian in some sense in a state of equilibrium, relatively free of
the imbalances that presumably are the chief cause of structural
change? If so, are other Indo-European languages tending to the
same sort of structure? Or is the relative stability of Modern
Persian the counterpart of its preceding rapid evolution; in other
words, is there some more-or-Iess uniform rate of change in
language, such that a period of rapid change is regularly followed
127
by a period in which the language changes more slowly, covering

in the end the same ground that another language might travel at
a more uniform speed?
The remaining aberrancy is the position of Homeric. Approx
imately contemporary with Vedic, this language is far down on
the list, below Old Church Slavic and Old English of the ninth
century A.D. Indeed, Homeric Greek ranks even below its own
daughter language, New Testament Greek. That this is no mere
accident of the samples counted is suggested by supplementary
counts that I made of Hesiod's Works and Days (beginning with
Zeus in line 143 and ending with genos in line 156, according to
Rzach's third edition, Leipzig, 1913) and Plato's Apology (from
hoiltos at the end of 20e to the first pate in 21b, following
J. Burnet's edition, Oxford, 1900). The first of these is epic poetry,
undated, but linguistically clearly later than Homer; Plato is early
fourth century B.C. For Hesiod, I found an index of 2.19 and for
Plato 2.30, suggesting that there was indeed a steady increase in
synthesis within Greek of the first millennium B.C.
These figures show clearly that the popular conception of an
uninterrupted decrease in synthesis within the history of Indo
European is too simple. Furthermore, comparative reconstruction
suggests that the Indo-European protolanguage was probably less
synthetic even than Homer, to judge by its relatively large
number of nouns and verbs without derivational affixes (that is,
root nouns and athematic root verbs, including injunctives). There
is even some evidence that at one time a verb could not contain
more than one derivational affix-for example, a verb charac
terized by a tense-aspect suffix could not add to this a subjunctive
or optative mood suffix.
It therefore seems necessary to substitute for the oversimple
straight-line decrease in synthesis a curve with at least one in
flection. At the earliest period which our reconstructions can
reach, the number of morphemes per word was rising in the Indo
European dialects. Within each dialect it reached a maximum, and
then turned downward. in which direction it has been moving ever
since. (Further data, of course, could easily reveal that this scheme
also is too simple.) The low MjW index of Hittite would then be
128
Warren Cowgill
due partly to an early turn downward, partly to a steep rate of fall.

Of the subgroups from which more than one representative is
studied here, Indo-Iranian and Germanic evidently reached their
maxima before the earliest recorded texts. J suspect that this
would prove to be true of most other Indo-European subgroups
as well, so that only in Greek are we privileged to see an Indo
European language becoming more synthetic.
If one assumes that the history of Indo-European synthesis here
proposed is correct, it would be interesting to know the time
range within which the different dialects were at their M/W
maxima-that is, was Indo-Iranian abnormally quick in evolving,
or was Greek abnormally slow? Unfortunately, the late date at
which most Indo-European languages are attested makes this
question unanswerable. The only subgroup besides Anatolian,
Indo-Iranian, and Greek adequately attested before the writing
of the New Testament is Italic. Hence, languages like Germanic,
Balto-Slavic, and Celtic may have reached their maxima no earlier
than Greek did, or they may have reached them by 1000 B.C. or
even earlier.
[t is less easy to fit the indices of agglutination into a simple
scheme. The high Old Persian index could be correlated with the
language's low M/W index, and taken as one more symptom of
the accelerated development that Persian was beginning to under
go. But the low index of Old English and the high index of Asokan
suggest rather that the index of agglutination was generally
subject to much more oscillation than was the index of synthesis.
I think this can be plausibly explained: complex morpho phone
mics is mainly the result of sound changes which cause morphs
to develop differently in different environments. A language under
going no sound change would in time eliminate most of its mor
phophonemic alternations by analogic spread of one allomorph
for each morpheme, and the history of a language's morpho
phonemics is largely a seesaw between sound change and analogy.
Since sound changes presumably operate with little or no regard
for morphology, it follows that a language tending to become
more and more agglutinative can easily undergo a sound change
which will temporarily arrest or even reverse the general trend.
129
Comparative reconstruction suggests that Proto-Indo-European

was, if anything, less agglutinative than any of its descendants
here studied. By internal reconstruction a stage can be envisioned
in which part of the morphophonemic complexity (namely
quantitative ablaut) did not exist, but in general it seems safe to
say that if there ever was a maximum of morphophonemic com
plexity, it was a good deal earlier than the maximum of synthesis.
In the matter of compounding we find a radically different ar
rangement. Here Indic uniformly has a high R/W index, ranging
from) .09 in Bengali to 1.22 in Asokan. Apparently compounding
became early established as a favorite way of joining morphemes
in Indic, and remained popular there despite all the changes that
the language underwent in other respects. The high index for Aso
kan is partly due to my having taken as a compound the oft-re
curring devanaY(lpriya- 'dear to the gods'. If this is taken as a
phrase, the R/W index sinks to 1.16, still the highest in the corpus.
It may be noted also that Asokan yielded the sole example of a
three-root compound.
I strongly doubt that there is any significance to the ordering
of the remaining ten languages, with indices ranging from 1.00
to 1.03. Whatever real differences there may be among them in
frequency of compounds are evidently too small to be revealed
by counting only 100 words of text.
In another way the method used here fails to reveal an impor
tant feature of Indo-European historical grammar. By not making
a distinction (suggested by Voegelin, Ramanujan, and Voegelin,
IJAL 26.200) between compound verbs and compound nouns, it
fails to reveal that in all these languages compound nouns freely
occur, but that only in Bengali and Modern Persian are there
compound verbs (aside from rarities like English housekeep). 5
Considering how common verb compounding is in the world's
languages, its slowness to develop in Indo-European suggests that
compounding techniques may be a part of the grammar that is
especially tenacious of old patterns and slow to develop new ones.
It seems appropriate to mention here an oft-noted archaism
of Indo-European noun compounding, the appearance of the
first element in its stem form, with no case or number ending. Thus
130
!Il
li,i
Warren Cowgill
in Russian klinopis' 'cuneiform writing', the word klin 'wedge'

appears in a form which has otherwise disappeared from the lan
guage and which ignores the fact that the meaning ofthe compound
would require a plural, probably in the instrumental case (,writing
done with wedges'). It is commonly stated, probably correctly,
that this feature of IndoEuropean compounding technique dates
from a time before inflection of nouns for case and number had
developed. What is usually not stated is that this is no evidence
that case and number endings are a particularly late feature of
Indo-European; once established, the pattern of compounding
could have remained through tens of thousands of years.
Simply ranking the derivational, inflectional, prefixing, and
suffixing indices would have little value, since these figures are
essentially only a breakdown of information already contained in
the M/W index. If a language has a high morpheme-perword in
dex, it will automatically have high indices in at least two, and
probably all four, of the indices 4 through 7. 6 {For this reason, the
summation technique used by Kroeber, lJAL 26.175, seems
unlikely to prove very useful.}
These indices can be put to better use .ey calculating the ratios
in each language of D/W to I/W and ofP/W to S/W. In this way
we can see whether there is any trend to enhance derivation as
compared with inflection, or to change the relative frequency of
prefixes and suffixes.
The ratios of derivation to inflection do not reveal to me any
obvious tendency of change that might prove to be valid for all
of Indo-European. In Indic and English the ratio seems to have
increased with time; in Persian and Greek it has decreased. More
striking is the tendency for Indic and Greek to cluster at opposite
ends of the list, with Germanic tending to be low but not so low
as Greek, while Iranian, Slavic, and Hittite occupy the center.
This suggests that each subgroup, once a particular ratio of deri
vation to inflection had been established, tended to maintain that
ratio without a great deal of change. But all conclusions from these
figures are subject to a great deal of caution, in view of the some
what uneven way in which I decided what was inflection and what
was derivation. For instance, had I treated tense, mood, and parti-
131
ciple formants in Homer as derivational, as r did in Vedic, Old

Persian, and Hittite, the Derivation/Inflection ratio for Homer
would have been .67, well above everything but Classical Sanskrit
(in which, to be sure, I strongly suspect that Greenberg in fact
treated such morphemes as inflectional).
TABLE 3
I.
2.
3.
4.
5.
6.
7.
8.
D/W
I/W
*CISk.
Bg.
10.
II.
12.
RV. Asoka
.51
OP
OCS
Hitt.
Go.
.42
.36
.34
.31
*NE
*NP
9.
.74
.53
Hom., NT
.28
.26
.25
06
.22
.18
NGk.
Languages calculated by Greenberg.
\3.
14.
P/W
S/W
OP
.16
RV,NT
.15
CISk.
OCS
Go.
.14
.10
.08
f1om, A.,k.
*06 *N6
.06
NGk.
*NP
.04
.02
} Bg., Hitt.
.01
The ratios of prefixes to suffixes, on the other hand, seem to

show some correlation with the indices of morphologic com
plexity shown in Table 2, especially the morpheme-per-word index.
The four languages with prefix-to-suffix ratios of .14 to .16 have
correspondingly high morpheme-per-word indices; Gothic and
Old Church Slavic are next in both lists, followed by Homeric
Greek and Old English; and Hittite with the four modern lan
guages is at the end. Only Asokan seriously disturbs the picture,
being much lower in the prefix-to-suffix list than in the morpheme
per-word list. Also, the ran kings within each group are not the
same in both lists, and perhaps further data would show the
apparent correlation to be illusory. But, tentatively, we can say
that in Indo-European a high morpheme-to-word ratio appears to
go with a high prefix-to-suffix ratio. Greenberg has suggested
132
Warren Cowgill
(Essays in Linguistics, 89-92) that prefixes may be psychologically

more difficult to work with than are suffixes. Although Chomsky
(Word 15.202-203) has cast serious doubt on most of the reasons
advanced by Greenberg for the relative infrequency of prefixes
in the world's languages, it does seem possible that there may be
some fundamental difficulty about prefixes compared with
suffixes, so that a high ratio of prefixes and prevalence of poly
morphemic words are just two aspects of a basic trait of 'com
plexity', and a rise in one could be expected to entail a rise in the
other.
The correlation between prefix-to-suffix ratios and indices of
agglutination seems best toward the bottom of the list. In both,
Bengali and Hittite come at the end, preceded first by Modern
Persian and Modern Greek, and then by Asokan and Modern
English, and in both lists Gothic is sixth; but the arrangements
of the remaining seven languages inthe upper part of the list seem
unrelated to One another.
It seems worth while to remark that the extremely low prefix
to-suffix ratio of Hittite agrees well with the hypothesis that this
ratio is somehow tied up with the general morphologic complexity
of a language; otherwise we might expect Hittite to have a prefix
to-suffix ratio comparable to that of Homeric or Vedic.
Infixes were found only in the Vedic sample. Several of the other
languages tested have them, but the samples r examined did not
happen to contain any. (Whether Greenberg found any at all
his samples he does not say-hence the dashes in Table I.) As
far as the evidence goes, the ratio of infixes to suffixes, like that of
prefixes to suffixes, is correlated with the general morphologic
complexity of the language. Counts of longer samples ought to
show whether this correlation indeed holds good or not.
The three remaining indices are primarily syntactical, and I
do not see how they could be used to get information about mor
phology that cannot be gotten better and more easily in other
ways. Thus No.8, the index of Isolation, is directly proportional
to the number of uninflected words in the text, and hence could
conceivably be used to estimate the ratio of inflected to uninflected
words in the language, or the average number of inflectional
133
morphemes per inflected word. But both these figures can be

obtained more simply and more accurately by counts performed
on the text itself.
Similarly the Pure Inflection and Concord indices could be
compared with each other to give a ratio of concordial to noncon
cordial inflectional morphemes, but again this could be better
obtained by simply counting within the sample. And the informa
tion obtained would, I think, be considerably less interesting than
a study of the Proto-Indo-European concords themselves and the
ways in which they have been retained, lost, or replaced by new
ones. But this would lead into the second kind of investigation
mentioned at the beginning of this paper, in which the changes
themselves rather than the states resulting from the changes are
compared.
However, not to limit the scope of this paper too severely to
morphology, I have calculated the quotients obtained by dividing
indices of Concord by indices of Pure Inflection, in order to see
if any obvious trend through time would emerge. The results, in
descending order, are: Asoka 2.3; NGk. 1.2; NT 1.1; Hitt. 1.0;
OP .97; Hom..96; Go..85; *CISk .. 83; *OE. 81; *NE, OCS .79;
*NP .66; RV .56; Bg..48. (Again asterisks indicate languages
calculated by Greenberg.) I am unable to discover from these
figures anything that looks like a consistent direction of change.
It appears, then, that typology, even in its present very imperfect
state, reveals well some features of Indo-European drift, namely a
general rise in agglutination, an early rise in synthesis followed by
a decline, and a general decline in the prefix- (and infix)-to-suffix
ratio. The aberrant position of Hittite in all three of these features
is at once enough to require some special explanation (whether a
proto language distinct from the rest or creolization). On the other
hand, larger samples are needed to find out what' has been hap
pening to compounding, and the comparison of derivation with
inflection and of pure inflection with concord suggests that in
these areas there has been no general drift in either direction.
To look for uniform retentions within the typology of Indo
European seems of little value until indices for more non-Indo
European languages are available. Using the four non-Indo
134
Warren Cowgill
European languages for which indices are given by Greenberg, we

can only say such things as that synthesis has not (yet) gotten as
low as in Vietnamese (1.52 in Modern Persian, 1.06 in Vietnamese)
nor agglutination as high as in Swahili (.46 in Bengali, .67 in
Swahili); compounds and prefixes exist in all the Indo-European
languages studied (even though the samples of Hittite, Old Eng
lish, Modern English, and Old Church Slavic happen not to con
tain any compounds), while Yakut, Vietnamese, and Eskimo lack
prefixes altogether and Eskimo has no compounds. 7
Jij
l;lj
As an example of what might be done'comparing the changes

that have taken place in Indo-European subgroups without re
stricting oneself to the presently available typological indices, I
shall add here some remarks on the history of the three positional
classes of affixes in Indo-European.
Suffixes were numerous and important in Proto-Indo-European,
forming the principal machinery of derivation and inflection and
responsible for most of the morphophonemic alternation of root
morphemes. Suffixes are still numerous in all the Indo-European
languages, and to a considerable extent are etymologically derived
from Proto-Indo-European suffixes of similar meaning. This is
particularly true of the case-number endings of the noun and the
personal endings of the verb, the morphemes which in Proto-Indo
European were already fairly well organized into paradigmatic
structures. Most of the innovation among these has been the result
of sound change (often causing a suffix to disappear altogether),
analogic transfer from one class of stems to another, and reduc
tions in the number of distinct grammatical categories recognized
by the language. Indo-European derivational suffixes, including
the tense and mood formants ofthe verb, have been somewhat less
conservative, partly, I suspect, because derivation is naturally
more casual and less paradigmatic than inflection, and partly
because the tense-aspect-mood systems of the Indo-European
languages have generally evolved structurally much more than
the case-gender-number systems of the noun.
The history of prefixes and infixes has been quite different.
Proto-Indo-European had only one infix, -ne- (ablauting with
UNIVERSALS IN DIACHRONIC MORPHOLOG Y
135
-n-) inserted between the second and third consonant of three
consonant roots and forming present stems of verbs, for example

*k!-ne-w- to the root *klew- 'hear', or *yu-ne-g- to the root
*yewg- 'join, harness'. Already in the proto language such presents
were not common-I should guess there were at most about forty.
In most dialects of Indo-European their numbers have been
sharply reduced, and no new infixes created (unless, to be sure, one
analyze forms like English feet and sang as containing infixes
rather than zero suffixes). As far as I know, only Baltic now
possesses a productive nasal-infix formation, of the type Lithua
nian 'Sviiita 'grows bright', pret. svito. On the other hand, the
infixed presents had begun to give rise to nasal suffixes already in
the proto language : by recutting nasal presents to roots ending
in *w and *A, new suffixes *-new- (-nu-) and *-neA- (-nA-) were
created, of which at least the former enjoyed a fair popularity in
several branches of Indo-European.
What little can be said about the origin of the Proto-Indo
European infix does not bear out Greenberg'S suggestion (Essays
in Linguistics, 92) that infixes usually arise from prefixes. At least,
I know of no evidence whatever that the Indo-European nasal
infix was ever a prefix. On the other hand, it could conceivably
have begun as a suffix. According to one view, not entirely sat
isfactory, an early prestage of Indo-European had typically two
consonant roots, for example, *kl 'hear', *yw 'join'. To these
might be suffixed a nasal, *-ne-, or some other consonant, for
instance, *w or *g; if, now, such a consonant is suffixed both to the
plain root and to the nasalized form, we get stems like *kl-ew-:
*kl-ne-w- or *yew-g-: *yu-ne-g-, in which -ne-, originally a suffix,
has become, so to speak, embedded within the word as an infix.
As for prefixes, Proto-Indo-European had two kinds. One was
the so-called augment, a vowel (mostly e) that could be prefixed to
verbal forms which in themselves were ambiguous as to mood and
time (the so-called injunctive) but with the augment were unam
biguously indicative and (almost always) preterit. Certain features
of the augment's accentuation in Greek and Sanskrit and its failure
to occur outside a contiguous group of languages (Greek, Arme
nian, Indo-Iranian, perhaps Phrygian) suggest that up until late
'''II
136
Warren Cowgill
in the prehistory of Indo-European it was an independent word

(meaning roughly 'really') which came to be more and more
restricted in its freedom of occurrence until finally.it disappeared
altogether in most of Indo-European, and in the dialects ancestral
to Greek, Armenian, and Indo-Iranian was limited to the position
immediately preceding a finite indicative verb. The limitation to
this one position automatically entailed a change of status from
word to prefix.
The subsequent history of the augment in the three subgroups
where it is found is generally parallel. At first, in Vedic, it is a
meaningful morpheme, ma.rking as preterit indicative forms that
without it were unmarked as to time and mood. In Homer the use
of such unmarked forms in values other than past indicative has
already largely disappeared, so that augment is mostly an op
tional addition to a form already unambiguously characterized,
although it still performs some service in distinguishing a form
like epheromen 'we carried' from pheromen, which can mean
either 'we carried' or 'we carry'. Then there came a period, repre
sented by Classical Sanskrit, Old Persian, and Classical Greek,
in which the augment is universally required with certain preterit
tenses (even though most of these would have been unambiguously
characterized without it), or else is not used at all-so in Avestan.
Apparently it was psychologically easier to use an augment con
sistently everywhere or nowhere than it was to decide in every
case whether or not one was likely to be understood without it.
Yaghnobi, the one modern Indo-Iranian language to retain any
of the old preterits, is apparently still at this stage, using the
augment everywhere.
The next stage is that represented in Indo-Iranian by Pali (a
Middle Indic language roughly contemporary with Asokan),
Modern Greek, and Classical Armenian (fifth century A.D.). The
tenses that take augment are now unambiguously marked even
without it, and (consequently?) have begun to drop it, especially
from the longer words. In most of Indo-Iranian, from Prakrit
and Middle Persian onward, the old preterit tenses were replaced
by new forms based on the past participle, so that the augment
has now completely disappeared. In Classical Armenian the aug
,I,
.ii,
137
ment is already severely limited, occurring only in aorists that

without it would be monosyllabic. A cursory glance at a Modern
Armenian grammar indicates that there is now only one aorist
with augment, ekaw 'came'; since the present, gay, has a different
consonant, this is presumably a case of suppletion between stems
ga- and eka-, in which e- is no longer a separate morpheme. Thus
only Yaghnobi and Modern Greek now retain the Indo-European
augment.
The other type of prefixation in Proto-Indo-European is the
so-called reduplication, in which the initial consonant of a root,
followed by a vowel (and sometimes another consonant), is pre
fixed to the root. Reduplication occurred in the proto language

mostly as a formant deriving tense-aspect stems from roots. As
a formant of present and aorist stems it was in competition with
a number of suffixes, and never achieved any great popularity
except in Sanskrit, where it was used to form intensive presents,
causative aorists, and (in conjunction with a suffix) desideratives.
For the most part, reduplicated aorists and presents survived only
as isolated irregularities, becoming continually fewer and less
transparent in structure. Only in the perfect, where it had no
serious competition, did reduplication settle itself firmly, reaching
probably its maximum development in Hellenistic Greek. But
even in the perfect the reduplication was subject to restructuring
and loss in a way that seems quite unparalleled among the suffixes.
Thus in Modern Greek the perfect participle (the one form of the
tense surviving) regularly lacks reduplication. In Germanic, where
the preterit of strong verbs mostly continues the Indo-European
perfect, most verbs had lost reduplication already in Proto
Germa~ic; those that still had it in Proto-Germanic underwent
extremely violent remodelings in North and West Germanic,
resulting in its disappearance (and the creation of half a dozen new
ablaut patterns).
As a result of these changes, the Proto-Indo-European prefixes
have almost completely disappeared from modern Indo-European.
In the samples of Asokan, Bengali, Hittite, Modern Greek,
Gothic, and Old Church Slavic discussed earlier in this paper, not
one prefix inherited from Proto-Indo-European occurs, and I
':
i
,!,
;','
,'":!
i
"
,:,
138
Warren Cowgill
doubt strongly that Greenberg found any in his samples of Old

English, Modern English, and Persian.
But the stock of prefixes has made gains from another source
by a development which seems to have gone on parallel in all the
branches of Indo-European except Tocharian that survived the
middle of the first millennium B.C. In Proto-Indo-European, as
still in Hittite, Homeric, and Vedic, there were combinations of
verb and adverb comparable to English expressions like 'go up'
or 'come in'. The order of words was not fixed; the adverb
normally preceded the verb, but might do so immediately or sep
arated by other words. But by a process similar to that previously
assumed for the prehistory of the augment, the order of words
came to be more and more fixed, until (with a few exceptions,
such as Old Irish) the adverb came to stand always immediately
before the verb, forming a single accentual unit with it. The
adverb was thus no longer an independent word, but a prefix to
the verb. This in turn entailed a restructuring of the existing com
binations of such adverbs with noun stems: what had been com
pounds now became nouns with prefixes.
Such neoprefixes occur already in Old Persian and New Testa
ment Greek, and constitute most or all of the prefixes in the
samples of later languages that I discussed earlier in this paper.
(Only, as usual, Hittite is aberrant, having besides its free adverbs
two verbal prefixes, of which one happens to occur in the sample
studied.) To some extent even these new prefixes have by now
fused with the following roots on account of semantic or phono
logic change, and so ceased to be independent morphemes.
It seems, therefore, that Indo-European prefixes have shown
little tenacity compared to suffixes, and have been created anew
by only one process, the fusion of two words into one. There is
nothing to parallel the creation of new suffixes by combining old
ones or by reanalyzing the end of a stem as a suffix (type ox-en).
The history of Indo-European prefixing has yet another inter
esting feature. Of the three kinds of prefixes, reduplication,
augment, and adverbs, the origins of the first are lost in the dis
tances of prehistory, the augment probably became a prefix about
the time that strong dialectal differences were arising in the proto-
139
language, and the development of the last dates from the recorded
history of Indo-European. The morphophonemic complexity of
each of these classes is in direct proportion to its antiquity. Re
duplication, the oldest, is thoroughly nonagglutinating. The aug
ment was probably agglutinative in Proto-Indo-European, and
is nearly so in Sanskrit and Greek. In Old Persian it may in fact
be agglutinative (I have assumed that it is in calculating the Old
Persian A/J index), although the writing system does not permit
certainty. But adverbial prefixes are almost everywhere agglu
tinative.
What this means, of course, is simply that the more recently
created morpheme sequences have not been in the language long
enough to be seriously disrupted by sound change. Applied to
non-Indo-European languages, this observation might help to
determine the relative age of different sets of affixes: the more
morphophonemic alternation a set shows or entails in adjoining
morphemes, the longer it is likely to have been in the language.
Notes
1. I have since learned that the text sampled by Greenberg is a story
attributed to Saadi, who died in 1291 A.D. But, as H. H. Paper
kindly informs me, its grammar is not significantly different
from that of present-day colloquial Persian, so that my discussion
of it as if it were from the nineteenth century probably makes
for no serious distortions. In any case, this sample is later than
any of the ten non-Modern samples investigated by Greenberg
and me (with the possible exception of Classical Sanskrit). It
should be noted that the division between Middle and Modern
Persian is generally put in the eighth century A.D.
2. Fred W. Householder tells me that there is in fact good reason to
consider tha a separate word, and not a verbal prefix.
3. This suggestion is, of course, not new.
4. Cf. note 1.
S. Householder calls my attention to the existence of verbal com
pounds in Modern Greek of the type piyenoerkhome 'come-and
go', but 1 gather they are not common. Verbs derived from
compound nouns are of course common in many Indo-European
languages.
140
Warren Cowgill
6. Unless, as could conceivably happen, it has numerous and complex

compounds, and very restricted derivation and inflexion.
7. Householder has sent me M/W and syntactic indices calculated for
samples of Latin and Spanish by Heles Contreras. The samples
are: Caesar, De bello gallico 1.2, 1l.9, V.52; Gregory of Tours,
Historia Francorum 11.7, 111.3, 1f1.4; Peregrinatio od loco sancta
1.1, XIf.I, XXV.I; E. Relano, Historia dellenguaje (1953) p. 8;
R. Gallegos, Dona Barbara (1945) p. 96; and S. Reyes, Monica
Sanders (1951) p. 76. The figures are:
M/W
DIN
Pi/N
CoIN
.11
.73
.67
.68
.69
.16
.20
.23
.20
.69
.63
.68
.67
.17
.21
.20
.19
Caesar 1.2
Caesar 1I.9
Caesar V.52
Average
2.08
2.16
2.27
2.17
Gregory 11.7
Gregory IIL3
Gregory lIlA
Average
2.ll
2.11
2.11
.14
.16
.12
.14
Peregr. 1
Peregr'. XII
Peregr. XXV
Average
1.82
1.82
1.83
I.S2
.15
.IS
.19
.17
.62
.59
.61
.61
.23
.23
.20
.22
Relano
Gallegos
Reyes
Average
1.51
1.58
1.47
1.52
.42
.36
.44
.40
.39
.40
.30
.36
.19
.24
.26
.23
.\3
.09
. 11
141
of Old English-much later texts than Caesar, but impressionistically

of about equal morphologic complexity with Latin, and like Latin
typologically fairly close to ProtO-Indo-European. Then comes
Gregory from the sixth century A.D., doing his best to write in the
language of Caesar, and with an index of synthesis (2.11) only a little
lower than Caesar's. But the popular language had already changed
much more radically, as is shown by the Peregrinatio, a century or so
earlier than Gregory, but with an index of only 1.82, equal to that of
Modern Greek and lower than that of Modem Bengali. Apparently
by the fifth century A.D. spoken Romance had already become a
'modem' Indo-European language as far as number of morphemes
per word is concerned. Finally, the index of Modern Spanish, 1.52,
shows that the trend downward has continued, but is no lower than
that of Modern Persian, the least synthetic of the Indo-European
languages investigated by Greenberg and me .
The syntactic indices suggest that the proportion of concordial
nexuses has remained fairly stable-even rising a little-while the
proportion of isolating nexuses to pure inflectional nexuses has risen
considerably, especially during the last 1500 years. This is more of a
pattern than I have been able to see in the corresponding indices of the
other 14 languages.
!J;1
Householder's typescript lacks a figure for the M/W of the third

sample of Gregory.
Contreras' counts invite some comment. In the first place, it seems
noteworthy that the figures for different passages of the same Latin
author are fairly close together, as are also the M/W indices of the
three Spanish authors; but the syntactic indices for Spanish diverge
considerably. This suggests that of the figures derived by Greenberg
and me from counts of single loo-word passages, the M/W indices are
probably fairly reliable, but the syntactic indices may well not be.
The M/W indices fit fairly well into the pattern of my Table 2.
The highest index, 2.17, belongs to the oldest text, Caesar's polished
first-century D.C. prose. This puts Caesar's Latin, with regard to
synthesis, between my sample of Old Church Slavic and Greenberg's
f(
i
ON THE SEMANTIC STRUCTURE OF LANGUAGE
!'f
143
: !
CHAPTER 7

URIEL WEINREICH
Columbia University
1. The Nature of Semantic Universals
l.l. The state of our ignorance
,I
If challenged to summarize in a nutshell the universal semantic

properties of languages on which lingui~ts could agree, one would
probably list two:
);,
(a) All languages are information-conveying mechanisms of a
particular kind, different from other semiotic mechanisms which
are not language (cf. Hockett, 1960). Thus, we would rule out, as
nonlanguage, systems which use other than vocal sign-vehicles;
systems whose sign-vehicles are not composed of discrete recurring
units (phonemes); systems which have unrestricted combinability
of signs (Le., no grammar); systems whose signs are iconic;
perhaps even such systems-to add a pragmatic criterion-as are
not used for interpersonal communication.
(b) The semantic mapping of the universe by a language is,
in general, arbitrary, and the semantic "map" of each language is
different from those of all other languages.
Obviously this is not much to go on. If, in phonology, we had

only the two analogous statements-that all languages have pho
nemes, and that the particular phonological system is different in
every language-we would hardly have met for a conference on
142
phonological universals. Where shall we look for additional high

level generalizations about the semantic properties of language?
The following lines of inquiry, it seems to me, might be prof
itable:
(c) From the semiotic point of view, language is not a homo
geneous mechanism. What are the semiotic submechanisms

utilized in language? Are the several mechanisms analyzed by
Wittgenstein as "language games" (l958.77ff.) uniformly dis
tributed throughout the languages of the world? What formal
features of languages are correlated with their semiotic strata?
(d) What are the effects of sign combination on the meanings
of signs? In particular, how do the grammatical and phraseo
logical limitations on the freedom of combination affect the func
tioning of linguistic signs?
(e) Despite the basically arbitrary quality of semantic "map
ping" displayed by languages, there are nevertheless remarkable
parallelisms between both related and unrelated languages. How
are these parallelisms to be formulated and quantified?
(f) What generalizations can be made about any vocabulary as
a structured set, imperfect as the structuring may be? Can any
over-all structural characteristics of a particular vocabulary be
formulated, and if so, can the distribution of such characteristics
in the languages of the world be studied?
The scarcity of relevant data is in itself a major obstacle to
the elaboration of workable hypotheses. As the references scat
tered in the present discussion show, there is much to read, but
no obvious place to look things up. The most important works
on semantics, such as those by Ullmann (1951), Zvegincev (1957),
Regnell (1958), Ziff (1960), and Schaff (1960), are on the whole
preoccupied with the one semiotic process of naming, that is,
with the use of designators in theoretical isolation; they pay rela
tively little attention to the combinatory semiotics of connected
discourse. Linguistic facts are cited as anecdotal illustrations of
this or that segment of the theory, but no attempt is made to
sample a whole language representatively. The possibly unequal
distribution of particular semantic phenomena among the lan
~l
~!
!,
,."
1
144
Uriel Weinreich
guages of the world is generally not even considered. There

exists a fatal abyss between semantic theory and semantic de
scription (Weinreich, 1962), an abyss which dooms the former to
emptiness and the latter to atomization. Subtle philosophers of
language like Cassirer (1923) have indiscriminately mixed reliable
and unreliable evidence about languages, sometimes allowing
evolutionary prejudices to come into play; brilliant logicians have
shown a lack of curiosity about languages other than their own.
The most stimulating writer of all-Hans Reichenbach (1948)
samples human language only by reference to English, German,
Turkish, and occasionally French and Latin.
Except for some very brief remarks by the Aginskys (1948),
the only outright attempt to approach what might be classed as
the problem of semantic universals has been made by Ullmann.
In addition to his programmatic paper (1953), we have his ex
ploration of the semantic structure of one language, French
(1952). But his generalizations are, by and large, premature (cf.
Weinreich, 1955), and culturally restricted by their method
(see 4.1). Almost everything still remains to be done.
No reader of this paper will be so naive as to expect sensational
solutions to any of the outstanding problems of semantic analysis.
Fully to specify the conceptual framework which underlies the
following discussion would alone require a monograph. The
writer's only hope is that a critical discussion* of his memo
randum may help to put certain questions into researchable form.
In view of the state of semantic studies so far, even this would be
a memorable achievement for our Conference.
1.2. Some basic terms

It will be useful to adopt, as a basis of discussion, the scheme
of semantics as developed by Morris (1938). We will accordingly
It is a pleasure to acknowledge the criticisms of the following persons,
who read an earlier version of this paper: Robert Austerlitz, Ol'ga S.
Axmanova, Edward H. Bendix., Dwight L. Bolinger, Harold C. Conklin,
Joseph H. Greenberg, Charles F. Hockett, Fred W. Householder, Jr.,
Benjamin Hrushovski, Milka Ivic, Pavel Ivic, Roman Jakobson, Lawrence
Krader, John Lotz, Wita Ravid, Michael Riifaterre, Rulon Wells, and
Karl E. Zimmer.
145
say that a language is a repertory of signs, and that discourse

involves the use of these signs, seldom in isolation. The rules of
permitted sign combination (grammar) are formulated in terms
of classes of signs (grammatical classes). Languages contain signs
of two kinds: every sign is, in general, a designator or a formator 1
(cf. 2.2). A designator consists of a sign-vehicle and a designatum; 2
a formator consists of a sign-vehicle and an implicit instruction
for an operation, such as negation, generalization, and the like
(see further 2.2). A designatum may be said to constitute a set of
conditions; in a situation in which such conditions are actually
fulfilled, and the sign is used in reference to the situation, the token
of the sign may be said to denote (Morris, 1938 :24).3 Sometimes
sign-tokens are used with a claim to denotation, sometimes
without (cf. 2.2.1.1 and 3.1.4). All languages also have deictic
devices; these are signs used for referring without designation
(cf. 2.2.2). Furthermore, languages contain designators and for
mators for discourse about language (metalinguistic signs in
addition to object-language signs).
The analysis of semiotic devices available in a language for
designation, referring, shifting of levels, etc., constitutes its
semiotic description. The structure of the designata of the signs
of a language is the topic of its semantic description in the strict
sense; we may also speak of semantic description in the broad
sense as including semiotic description. The relation of semiotic
type and designatum of a sign to the form of the sign-vehicle
is, of course, by and large arbitrary; however, to the extent that
recurrent parallelisms can be found, such semantic-phonological
intersections are worth describing. (The important problem of
sound symbolism is beyond the scope of this paper; but cf.
note 65.) The relation between the semiotic type and designatum
of a sign and the syntactic class to which it belongs is, on the other
hand, often intimate; the intersections of semantics and grammar
require even more attention than semantic-phonological paral
lelisms, for any language and for language in general.
In the debate over the exclusion of semantic considerations
from grammatical description, Chomsky'S uncompromising stand
(1955, 1957: ch. 9) is, in our opinion, entirely correct. In this
.1"
::1:,
:,
i
"
,~
"ii
II
I
"
}
146
Uriel Weinreich
147
'I
paper, it will be assumed that the grammatical description of a

language is not only autonomous vis-a.-vis the semantic one, but
is also presupposed by it. We therefore propose to submit to
semantic analysis only utterances which are grammatical, and
which have a specified grammatical structure. Thus, we consider
it productive to ponder what is semantically unacceptable about
enter out, but not about into out, which is disqualified as ungram
matical. 4 Similarly, in analyzing the polysemy of a word like
fair into (0) 'not biased', (b) 'pretty good', etc., it is economical to
observe first thatfair (b) belongs, unlike fair (a), to a very special
subclass of adjectives (see 3.2.1).5
The proposed priority for grammatical over semantic descrip
tion raises the problem of the congruence between the units of
each type of description. Grammatical analysis operates with
meaningful elements (morphemes-segmental and suprasegmen
tal-as well as optional transformations), and some meaningless,
obligatory processes. Bloomfield (1933:162, 166) was satisfied to
posit a unit of meaning for each unit yielded by the grammar
('sememes' for morphemes, 'episememes' for tagmemes ~ op
tional transformations). Since then, the identification of gram
matical and semantic units has met with a number of objections
(e.g., Hjelmslev, 1953 :28f., and notably Bazell, 1954). The most
important are these: (a) Some morphs are meaningless ("empty
morphs," e.g., to with infinitives, the -0- of drunk-o-meter).
(b) It is unnatural to have sign-vehicles without segmental sub
stance, e.g., a meaningful word order. (c) There may be meaningful
submorphemic segments ("phonaesthemes," e.g., FL-OW, FL-it,
FL-y, FL-oat). (d) In "idioms," the semantic analysis must treat a
polymorphemicexpression as a whole. But none ofthese objections
seems sufficient. (a) "Empty" morphs are an artifact of an Item
and-Arrangement grammar; in an [P grammar, they are not
"empty," but are the segmental markers of a transformation
process. (b) From the point of view of semiotic theory, there is
nothing wrong with having a process as a sign-vehicle (e.g., my
raising one hand three times as a signal to a confederate). (c, d)
Such phenomena as "phonaestheme" and "idiom" are indeed
definable as many-to-one correspondences between grammatical
II.
"
Ii
and semantic units. But while identity between the two planes is
incomplete, it is a useful starting-point from which to describe the
lack of isomorphism actually found. (See also note 65, and cf.
Chomsky, 1957: 102f.).
1.3. Full-fledged, subdued, and enhanced semanticity of speech
In a remarkable passage, Sapir (1921 :13) likens language to a

dynamo capable of powering an elevator but ordinarily operating
to feed an electric doorbell. Language is used, more often than
not, in ways which do not draw upon its full semantic capacity.
In its "phatic" functions, when speech is used merely to signify
the presence of a sympathetic interlocutor, it easily becomes
"desemanticized" to a formidable extent. In its various cere
monial functions ("noncasual" language: cf. French, 1958) lan
guage may come to be desemanticized by still another mechanism.
In general, insofar as utterances are the automatic symptoms of
a speaker's state, insofar as they are interlaced with chains of high
associative probability, insofar, in short, as they are not subject to
the full voluntary control of speakers, they fail to represent the
language in its full capacity as a semantic instrument. Now, the
various "leakages" that in practice reduce the power of language
as a communicative instrument constitute a legitimate psycholog
ical problem, to the solution of which the linguist may have some
thing to contribute. But the more pressing task for linguistics, it
seems to me, is to explain the elevator, not the doorbell; i.e.,
avoiding samples of excessively casual or ceremonial speech, to
examine language under conditions of its full-fledged utilization
that is, under conditions where .no behavior but language would
fill the bill.
The use of language can also deviate from the norm in the op
posite direction, so that the language becomes, as it were, "hyper
semanticized." Such use of language is characteristic of much
good literature, although it can be found in workaday life as well.
There are at least two marks of hypersemanticization: (I) The
phonic vehicle of signs assumes an independent symbolic value
(whether "impressionistic"-sound-imitative-or "expression
istic," i.e., synaesthetic); a special semantic relation is imputed
148
Uriel Weinreich
to signs with similar vehicles (rhyme, etc.); in short, incipient

correlations between content and expression are exploited, in
contrast to the arbitrariness of this relation in semantically "nor
mal" uses of language. (2) Over the scope of a given text (poem,
etc.) meanings are imputed to some signs which are richer than, or
otherwise deviant from, the meanings of the same signs outside
the text. Whereas in the "standard" use of language the receiver
of a message must only decode it, not decipher it (crack the code),
in "hypersemanticized" language the common code is modified
ad hoc, and the favorably inclined receiver of the message must
guess the code modification before he can properly decode the
message. It would be uneconomical to dally with a semantic theory
which is too weak to account for these phenomena (cf. 3.1.2 and
the Postcript); but it is equally pointless to concentrate on these
special effects, as so many writers on "meaning" have done,
without first accounting for the semantic workings of language in
its more standard' uses.
2. Semiotic Stratification of Language
2.1. Logical basis oJ semiotic analysis
In the following discussion, the grammatical form of sentences
will be compared with their semiotic form. In particular, it will
be assumed that it is possible to describe all discourse as either
(a) having the semiotic form 'OJ(x)" or (b) deviating from it in
specified ways. In this formulation, 'x' stands for an argument
"something talked about"; '/' for a predicate-"something said
about x"; and '0' is a covering label for any of a number of
operations. More will be said about the inner structure of 'x', ,/"
and '0' in 2.2 and 3.1.
The investigation of discourse in its logical aspects is not a
fashionable pursuit, but it seems to be one of the most important
frontiers of linguistics for the decades ahead. It is a defensible
enterprise, I believe, provided certain cautions are observed.
First, there must not be a breath of normativism in it; 6 the de
scriptive linguist has no interest in making language usage "more
logical" than it is-on the contrary, he should explain, if possible,
149
why it is not, in effect, more logical (sec. 5). Second, it is useful to

keep in mind that only a very limited portion of logic is brought
into play; we are concerned mostly with rules of formation and
designation, that is, with limited aspects of a functional calculus,
which in logic are merely the preparatory steps for the study of
deduction, truth, etc. (cf., e.g., Carnap, 1942:24). Third, as men
tioned in 1.2, the study of the "logical" aspects of discourse, as part
of semantics, must remain autonomous of the grammatical
analysis so that their interrelation may be meaningfully com
pared. 7 Fourth, we must insist on a sufficiently versatile logic
and on a wide sampling of languages. 8 Finally, we must carefully
avoid the unjust claim that man cannot in his thinking transcend
the "logical mold" given by his languages; there is ample evidence
to the contrary-if not in Aristotelian logic, then certainly, let us
say, in the medieval doctrine of suppositions. But if these cautions
are observed, the investigation is both legitimate and promising.
For logic is congenial to language. Even experimental language
systems constructed by philosophers conform in many essentials
of logical structure to that of human language. 9
It would be considered naive today to attempt, as did Wegener
(1885), to describe the semiotic stratification of human language
with examples restricted to German, Greek, and Latin. But it
is remarkable how well Wegener's theory stands up now that
the range of our evidence has been vastly broadened. It takes
only a slightly more flexible calculus, I believe, to accommodate
all the varieties of semiotic structure evident in ordinary discourse.
2.2. Formators
Virtually every semantic theory operates with a dichotomy of
signs, corresponding to what we have called designators and
formators. 1o In most systems the formators, or "logical" signs,
are given by enumeration. In 1942, it was not yet known how the
distinction could be defined for semantics in general (Carnap,
1942:59). Reichenbach's attempted definition (1948:318-325) may
be objectionable on technical grounds, and further theoretical
investigations are needed. But for our purposes we can apply
Carnap's working definition of 'designator' (1947:6): "all those
':
150
Uriel Weinreich
expressions to which a semantical analysis of meaning is applied."

While there may be controversial cases, it would seem that a
rough distinction of designators (e.g., bread, smear, fast) and
formators (or, this) conforms to an intuitive classification.
. If we consider as designators those signs which can appear in
the place of,!' and 'x' in expressions of the form 'bf(x)', the com
plementary class of formators would include, roughly, the fol
lowing kinds (after Reichenbach, 1948:55-57): (I) "pragmatic
operators"; (2) indexical signs; (3) signs for propositional op
erations (not, or, same [?]); (4) quantifiers of various types;
(5) signs which organize the expression ("purely syntactic" signs).
There is, in principle, a possibility of mixed signs: those
which have both formative and designative components. But the
mechanisms of each type must first be analyzed separately.
The descriptive problems which may eventually yield universals
of language are basically of three types:
:1
Ii,
(a) With what degree of distinctness are conceivably separate

logical operations expressed, or expressible, in the language?
(b) To what extent do the formatorsappear as separate gram
matical units; and, contrariwise, to what extent are formator
components "built into" the designata of mixed signs?
(c) To what extent do formators or mixed signs have charac
teristic sign vehicles or characteristic grammatical properties? 11
2.2.1. PRAGMATIC OPERATORS. The field of "pragmatics" has

virtually no conventional content. 12 For the present discussion
we propose to include in it that paradigm of discourse features
which comprises assertion, and features incompatible with as
sertion and with each other: question, command, and attitudes
to the content of discourse, insofar as they are coded.
2.2.1.1. As a practical measure we take the assertive "mode"
for a standard of reference. It will be seen in 3.1.4 that in every
language most utterances contain at least one sign linkage in the
assertive mode, although they may also contain additional ones
in a "neutral" mode. Among the devices which a language has
for "neutralizing" the assertiveness of sentences, some are com
pletely specialized for this function; among such are nominalizing
151
transformations, marked, for example, by a case change in the

subject and a change in the verb to the "infinitive" or some other
subordinate mood. Sometimes, however, the neutralization of
assertiveness is at the same time "motivated" by an indication of
the speaker's uncertainty or a positive disclaimer ofresponsibility.
The German change from indicative to subjunctive (Er ist krank
vs. Er sei krank) is such a sign; we may also list the Turkish -mi.y,
the Hopi "quotative" (Whorf, 1956:119), the Bulgarian non
evidential (Jakobson, 1957:4f.).
In many languages, the suspension of assertion is obvious from
the subordination of a sentence to explicitly nonassertive con
junctions (e.g., if ... ); hence, the marking of suspended assertion
by the mood of the verb becomes redundant and may be elimi
nated, as it was eliminated in Modern English. It is unlikely that
there are languages which have a greater stock of assertion
suspending devices for independent sentences than for condi
tionals.
2.2.1.2. The indication of the imperative seems typically to
intersect with deictic categories (2.2.2) and to be more highly
developed for second person than for first or third, for future/
present tense than for past. The equivalents of the imperative
for nonsecond person are often grammatically more analytic and
are asymmetrical with the second-person expression (cr. Yiddish
gejn 'to go': 2. gejt, I. lomir gejn, 3. zoln zej gejn). In many lan
guages, the imperative is marked in the verb only; often the
second person subject is deleted. 13
2.2.1.3. Questions are a marked pragmatic mode incom
patible with assertion,14 although they are not usually expressed
by a form fitting the grammatical paradigm of verb moods. Sen
tence questions (yes-or-no) are almost universally indicated by
intonation changes and nearly as frequently, perhaps, by the
addition of a question "particle" (Russian Ii, Hopi pi", Chinese
.ma); in considerably fewer languages are there also changes in
word order. The hierarchy of the devices can probably be for
m ulated more rigorously.
In contrast to assertion-suspension and command, questions
(like attitudinal formators-2.2.1.4) constitute pragmatic op
152
Uriel Weinreich
erations applicable to parts of sentences as well as to wholes.

When applied to parts, typical intersections of the question op
erator with the "part of speech" occur in so-called completion
questions, or wh-questions. While there appear in general to be
special forms of interrogative words depending on the part or the
sentence whose completion is desired (cf. English what? vs.
when ?), there are interesting gaps. For example, verb interroga
tives (e.g., * Whatted he? = 'What did he do 1') sometimes occur
(cf. Sapir, 1921 :126, on Yana), but they are rather rare; adjective
interrogation in a language like English is accomplished peri
phrastically (what kind of? yet cf. Polish jaki, Yiddish vosar). It is
not clear whether there are any languages with prepositional
interrogatives, although it is easy to conceive them (e.g., English
*whep, meaning 'on or under or over or .. .', as in Wh-ep the table is
the book? 'Is the book on or under or inside r... etc.] the tableT). 15
More transparent is the reason why certain grammatical distinc
tions typical of a part of speech are neutralized when it is inter
rogativized: if what had separate singular and plural forms, we
would have to know the number of the answer before asking the
question. On the other hand, the grammatical specialization of
interrogative words is not correlated exclusively with the largest
parts-of-speech divisions; thus English distinguishes animate/
inanimate in the noun interrogative (who/what); adverbial inter
rogatives are particularly overdifferentiated (where/when/how and
i even why) in comparison with noninterrogative adverbs, whose
subcategories of place, time, and manner are entirely covert. It is
likely that such unusual distinctions, reflecting different dimen
sions of deixis (2.2.2), are typical of most languages.
2.2.1.4. Attitudes toward the content of discourse are always
present and form a subject for psycholinguistic research (Wein
reich, 1958); they are relevant linguistically insofar as they are
coded. The usual attitudes which find coded expression on the
subsentence level are approval and disapproval. Other systems are
clearly imaginable (e.g., a suffix indicating that the thing desig
nated by the word is feared, longed for, etc.), but it is not sure
whether they occur. Within the simple good-bad dimension, it
seems that hypochoristic forms are more common than pejorative
153
ones, and are implied by the latter within anyone language. It is

also clear that such "expressive derivation" is very unevenly
distributed among languages (Ullmann, 1953 :232); in the Euro
pean area, English is notoriously poor in this respect, Italian and
the Slavic languages (Stankiewicz, (954) are very rich; Yiddish is
the richest of all Germanic languages, probably as a result of
convergence with Slavic. As a formator, the expression of endear
ment seems to intersect with the designatum of smallness; even
if this is not a semantic universal, it is quite typical, although
theoretically it could have been the other way round. 16 Gram
matically, attitudinal formators seem to be distributed unequally
in any language which has them, for example, in Yiddish they are
standard for nOuns and adjectives but rare for adverbs and entirely
marginal for verbs (nursery talk only). It is doubtful whether in
any language the verbal class is more sensitive to such distinctions
than the nominal class. Attitudinal formators are sometimes
phonologically characterized, for example, by palatality in Yiddish
or by various consonantal modifications in Nootka (Sapir, 1915);
but even where such phonological characteristics are absent, or by
passed, certain paralinguistic devices (voice qualification) appear
in their place.
Attitudinal formators having whole sentences as their objects
are much more richly patterned than the good-bad qualification
for sentence parts. Grammatically, the two chief devices for
their expression seem to be affixal mood categories of the verb,
formed by affixes or auxiliaries (e.g., optative), and special
"modal" adverbs or particles (e.g., 'fortunately'). Thus, in Sierra
Miwok we find a volitional mood in the conjugation and a set of
adverbs meaning 'would that', 'dubitative' (Freeland, 195 I); in
Potawatomi we find particles meaning 'would that' and 'it is
doubtful that', combinable with the conjunct mode (Hockett,
1948:215). In the continental European languages, a particle of
"obviousness" seems very common (German ja, French done,
Russian ved' or -to, Polish przecie[i)). It appears that in many
languages such attitudinal formators share specific grammatical
and phonological features (monosyllabicity, unstressability, fixed
'
order, etc.; cf. Arndt, 1960).
154
: j'
Uriel Weinreich
Summarizing, we can say that formators of the pragmatic cat

egory are often combined with designative components into
mixed signs; that they tend to monopolize some types of sign
vehicles (intonation contours) and predominate among others
(order patterns, enclitic particles); and that they are quite un
evenly distributed among the parts of speech.
2.2.2. DEICTIC SIGNS. 17 These are signs (or components of
designata) which involve a reference to the act of speech in which
they are used. (See Casagrande, 10.3.) Among the factors of the
speech situation which are utilized in deixis are the following: the
utterer of the discourse (,first person') or the receiver ('second
person'); the time of discourse (tense) and its place (varieties of
demonstration); and the identity or nonidentity of the act of
discourse (anaphora, reflexiveness, obviation, etc.). That this
paradigm constitutes a striking universal of language can be
appreciated not only from its widespread distribution but also by
visualizing further factors of the speech situation which could be,
but do not seem to be, utilized in any language: the loudness of
speech, its speed, the certainty of the assertion. No language
seems to have "adverbs" meaning 'louder than I am now speak
ing', 'as slow as my speech now', or the like.
2.2.2.1. Person deixis occurs in highly asymmetrical structures
which would deserve a fresh cross-linguistic survey. Thus, many
languages have forms including speaker and hearer ("inclusive
first person"), but perhaps not those including first and third
persons (,not-you') or second and thitd ('not-'). Person deixis
often intersects unevenly also with nonperson distinctions of
gender and number (e.g., English you, undifferentiated for
number). With respect to parts of speech, person deixis is again
unevenly distributed. As a distinctive feature it seems to belong
characteristically to the noun category, whereas verbs display
only concord with the noun; it is not sure whether there are
languages with verbs corresponding to *to we = 'to be us', etc.
[n the Serbocroatian dialect of Gorski Kotar, according to
P. Ivic, the affirmative sentence-substitute da 'yes' may take per
son suffixes: da-m 'I yes yes, I do'; dd-s 'you yes = yes, you do'.
Within the noun class, the formators of person deixis seem to be
155
combinable with designators, namely status labels, only to a

limited extent. Thus there are languages with simplex mor
phemes corresponding to 'you, my superior' or 'I, your inferior',
but not to 'you, a teacher', 'we Americans': combinations of the
latter type are invariably complex, being either phrases (as in
English) or words (cf. the conjugated nouns of Miwok: miwHe'-y
'I am an Indian'; Freeland, 1951 :26; also in Hottentot, according
to Greenberg).
The distinction between honorific and non honorific signs,
limited in some languages to a minute place in the pronominal
system but cutting across a large part of the basic vocabulary in
other languages (Tibetan, Javanese), can be analyzed semantically
under different headings. Sometimes the honorific component of
meaning is not dependent on the speaker's evaluation and can
be considered on a par with any other designational feature (cf.
4.2); thus in Thai hat, bllt seem to mean 'royal hand, royal foot'
(as against mi, thao 'commoner's hand, commoner's foot')
regardless of who is talking to whom. If the choice of honorific
forms depends on the attitude of the speaker to the listener or to
the subject of discourse, the semantic component might best be
classed with other attitudinal operators (2.2.1.4); cf. the Tibetan
choice between u and go 'head', gongpa and sampa 'thought',
chhab and chhu 'water', etc. (Gleason, 1955:156), the first item
of each pair being honorific, the second ordinary. IS But where the
use of honorific terms for second person and ordinary, or depre
catory, terms for first person becomes standardized, we observe
an intersection between this attitudinal operator and person deixis.
Such seems to be the case in Chinese (Chao, 1956 :219), where
bih-chuh, literally 'dilapidated locality', in effect means 'my home
town'.
2.2.2.2. Time deixis is generally expressed by signs which
modify either the verbs or (as in Chinese) the sentence as a whole.
Time deixis seems to be independent of other forms of deixis but
yields syncretisms with certain quantifiers of verbs (omission of
iterative aspect in the present tense, etc.) and with certain prag
matic categories: fewer tenses may be distinguished in the impera~
tive than in the indicative, and it is quite usual for tenses to be
156
Uriel Weinreich
neutralized under nominalization. 19 Nonpresent tense is often

combined with suspension of assertion; cf. the use of past for
conditional in English, the use of nominalized sentences for distant
past in Sierra Miwok (Freeland, 1952:49), etc. There seems
universally to be equal or greater discrimination of time distinc
tions in past than in future. The criteria for degree of pastness vary
and deserve to be investigated. Often a language has a tense for
the period from morning of the same day to the time of discourse,
and a separate tense for time before the day of discourse. In
variably expressions like 'Sunday', 'in summer' refer to the Sunday
or summer nearest the speech act.
It is perhaps a universal that time-deictic "adverbs" are never
less differentiated than the tense systems (Le., there are not more
past tenses than distinctions of the type yesterday, .. , ago).
Time deixis seems to be most. typically associated with verb
forms, although it is a perfectly conceivable component of noun
designata as well (the former, quondam, present, future king, the
then king, the ex-king, the king-to-be). In a language like Tupi, as
Greenberg has pointed out, there is a conjugation of nouns for
tense. Tense formators tend to intersect with designata involving
absolute time-d. the synchronically simplex etmol 'yesterday',
silsom 'day before yesterday' in Hebrew-but not with other
designata.
2.2.2.3. "Place" deixis seems to be organized according to
distance from the first or second person (cf. Latin iste 'this one,
in relation to you'), visibility ('the one I see'), accessibility, or
perhaps also direction ('before, behind')-usually in relation to
first person. (For theoretical analyses of "indication," see Collin
son, 1937; Shwayder, 1961.) Where only one category of deixis
exists, it seems to indicate 'obviousness to first and second person'
(this; thus); the reference may be made precise by a coordinated
gesture. Place deixis (as shown by Householder), too, seems to be
compatible with designata, especially if related to motion; cf.
come vs. go, bring vs. take. With regard to parts of speech, "place"
deixis seems to show asymmetries very similar to those ofinterrog
ativity (cf. 2.2.1.3). In many European languages, we find deictic
nouns (inanimate this), adjectives (such), "overdifferentiated"
157
adverbs (place: here, time: now, manner: thus), but not deictic
prepositions or verbs (* to this 'to do this'); so contrary to some
semantic systems is this potential category that in Yiddish, for
example, the deictic verb (dosn) occurs only in slang and means
'to excrete'. 20 Again, while the nouns and adverbs of time and
place distinguish proximate and distal deixis on a binary principle
(this/that, now/then, here/there), such a distinction for adverbs of
manner and for adjectives is perhaps rarer; cf. Russian distal tak,
takoj-proximate etak, etak(ij), Serbocroatian ovakav/onakav
(Ivic), Chinese dzemma/nbnma (Hockett) with the English thus,
such, undifferentiated as to proximateness. The distribution of
such asymmetries requires cross-linguistic investigation.
"Place" deixis easily combines with absolute indications of
place, especially in languages which deal with a very narrow
geographical area, where 'higher', for example, may come to mean
'northward' because of the direction of the one dominant slope.
(On an orientation system of this type, cf. Haugen, 1957.)
2.2.2.4. The greatest variety is apparently found with respect
to the distinction between "the same" and "not the same" act of
speech. All languages have "pro-forms" such as he, which sub
stitute for other forms to avoid their repetition within a unit of
discourse considered as "the same." But pro-forms are on the
whole very unevenly distributed with respect to the parts of
speech. Perhaps all languages have pro-nouns but few have pro
verbs; English is perhaps unique among European languages in
having, in do, at least the rudiments of a verb-phrase substitute.
For a large number of languages pro-adjectives, pro-numerals,
and pro-adverbs of various types seem to be the unstressed forms
of the corresponding demonstratives (cf. German er hat solche
Haare 'hair of this kind' = 'hair of the mentioned kind'). But
for the pro-nouns and pro-adjectives (definite article), at least,
some languages distinguish between demonstrative deixis and
"within-the-discourse" deixis: he/the distinct from this, French lui
(if) and Ie distinct from ce, celui, -ci, (:a. German makes the dis
tinction (er, der vs. dieser) , but Yiddish has lost the adjective part
(der/dieser), falling back on the device of so many languages
stressed and unstressed demonstratives. It seems to be a universal
I, .
t-
If"!
IIi
II
~ ~
II
I,
158
Uriel Weinreich
that under nominalizations certain distinctions of discourse-deixis

are neutralized, e.g., Bill's books ( < (some) books? the books?).
Such neutralizations are probably more common than the mainte
nance of the distinction, as in English a friend of mine, Yiddish
(but not German) majnar a frajnd.
The act of discourse considered as a unit generally extends
backward from the moment of time, since the purpose of such
deixis is to utilize information already conveyed. But it is also
possible to have a certain amount of forward deixis, illustrated
by such pro-adjectives and pro-adverbs as the following, as follows,
(let me say) this.
In most languages there are also grammatical processes which
take as their scope a small, well-delimited part of discourse. This
may be the sentence, as in rules of concord (e.g., for animateness,
person, number) between subject and predicate, or the verb and
object noun; there may be concord between verb (tense) and
adverb of time within the sentence. The sentence is also the unit
of discourse in rules for reflexivity (designation of object noun
by special devices if it is the same as the subject of the same verb),
or of obviation (use of different "persons," e.g., for different noun
objects of the same verb). Apart from its grammaticalized aspect,
reflexivity of sentence scope seems also to be a component of
such words as English (one's) own, home (='one's own home'),
along (with the subject), enough (' ... for oneself'), etc.
Some utilization of these two scopes for the criterion "the same
discourse" seems to be universal.
2.2.3. PROPOSITIONAL OPERATIONS. Under this heading we con
sider the linguistic equivalents of certain semantic operations
applied to propositions. Such operations may be singulary (ne
gation) or binary (disjunction, conjunction, implication, equiv
alence).
It is faily clear that no language represents such operations
with the maximum economy. While it has been shown to be log
ically possible to define all propositional operations in terms of
two primitives (e.g., negation and conjunction) or even a single
primitive, it would be interesting to discover what redundancies
are practiced in ordinary language. To appreciate the possible
159
differences we need only consider the Latin distinction between

vel and aut (exclusive and inclusive disjunction), unmatched in
modern European languages until the rise of and/or; or the
alternatives to and not (e.g., except) and ifnot... then (e.g., unless)
in English. Greek has one negator with a component 'dependence'
(me) and another (ou) without (Seiler, 1958: 694f.). One of the
best logical analyses of certain "operators" of ordinary English
is the one given by Strawson (1952:78-92). Such logical operations
fuse with pragmatic components, e.g., but = 'and, surprisingly';
yet = 'and, very surprisingly'; p although q = 'q and surprisingly
p'; etc. For some operations, it may also be useful to compare
languages as to the definiteness and flexibility of the "scope."
Thus, in English, we cannot always distinguish unambiguously
between negation of the verb and negation of the sentence as a
whole; but most other parts of sentences can be negated sepa
rately, if not by not, then by un- or non-. 21 In some languages the
sign of negation is a member of a special small form class, as in
English; 22 in others, it is a member of a large class, typically the
verb (*to not, e.g., Finnish, Yana). A cross-linguistic study of
negation would certainly yield important results.
Negation is one formative component which combines very
easily with other designata to form paradigms of antonyms; cf.
over = 'not under': under 'not over'; well 'not sick'; French
ignorer 'ne pas savoir', etc. Negation occurs in many (most?)
languages combined with signs for variables, with various de
grees of grammatico-semantical isomorphism (cf. some: no,
somebody: nobody, sometimes: never for older ever: never; but not
somewhat: *nowhat).
An extremely frequent syntactic concomitant of propositional
operations is ellipsis; cf. 3.1.5.
2.2.4. QUANTIFIERS. The representation in ordinary language
of operations comparable to the binding of variables in logic
requires a highly specialized investigation. But even preliminary
reflection leads to a number of hypotheses. First, every ordinary
language is far more redundant in its representation of quantifi
cation than a logical system, which can define all "quantifiers" by
means of negation and one primitive. Most obviously, no lan
160
UrieI Weinreich
guage represents cardinal number in logistic terms, and a renewal

of cross-linguistic studies of numeral systems, from the point of
view of universal deviations from logic, would be quite opportune
(P. Ivic). Second, generalization is in many languages expressible
in a form analogous to the logical form' (x)f(x)' (e.g., whoever...),
but it is always also expressible approximately as if the univer
sal quantifier were a designator (all books, like interesting books).
Third, quantification is not, as in the simple functional calculus,
restricted to "argument-terms" (say, nouns), but is also com
binable with "function-terms" and deictic formators. Thus Eng
lish has universal quantifiers in noun function (whoever,' every
thing, everybody, all), in adjective function (whichever; every,
each, all), and in various adverb functions (wherever, every
where; whenever, a/ways; however,' but not *everyhow, despite
anyhow!). Yet it has no general verb (*to all 'to do everything').
As in the case of propositional operations (2.2.3), we find some
mixture with pragmatic factors, e.g., only = 'surprisingly, no
more than'. A widespread form of quantification is combined with
event names and concerns their frequency or completion. In
languages such formators have separate morphemic representa
tion (frequentative, perfective aspects), whereas in others they are
lexicalized, that is, combined with designata into mixed signs
(cf. English some: all:: to carry: tofetch; one: many :: to attend:
to frequent). For the quantifications of events, many languages
have special subsystems of signs, such as 'once', 'twice', 'n times',
'every time', 'nonce'.23 Simplex terms for the quantification of
spatiotemporal deixis ('ever before', 'once before', 'time t ago" etc.)
are more easily imagined than found.
Every language has signs for existential quantifiers. This
semiotic class intersects with grammatical divisions into parts of
speech and some of their subdivisions. It may be a universal that
the grammatical specialization of the signs for existential quanti
fiers corresponds to that for the interrogatives (2.2.1.3) and the
deictics (2.2.2.3). In English, for example, we have indefinite
pronouns (animate: somebody; inanimate: something); indefinite
pro-numerals, also serving as pro-adjectives (some, any, in
intricate interrelations), the indefinite pro-adverbs of manner,
161
place, and time (somehow, somewhere, sometime[sJ- but hardly

somewhen!), but not of cause (no *somewhy for why). It is to be
noted that there is no indefinite pro-verb (*to something = 'to do
something') and that even the existing names of variables are
grammatically complex and asymmetrically constructed. These
irregularities seem to be typical deviations from a logical model.
Variable names do not seem to combine easily with designata.
And yet, if we are permitted to contrast such lexical pairs as say
with talk = 'say something', await and wait = 'await something',
we find examples of mixture in the transitivity of verbs when the
object is not specified. In many languages such "expectancy" is
explicitly shown in the form of the verb; cf. Hungarian {rok '1
write', from 'I write the .. .'.
Perhaps all languages distinguish between "divided" and "un
divided" reference (Quine, 1960 :90ff.), that is, between nouns
which are quantified in the form 'some x, a little x, much x' and
those which are quantified in the form 'an x, one x, many x'. 24
But whereas in a language like English the specification of the
kind of reference, divided or nondivided, is obligatory for the
noun, in most languages it seems to be optionally marked. The
distinction also occurs among nonnouns, for example, divided
reference of verbs by means of punctual and iterative aspects; the
comparison of English with Russian suggests that the gram
maticalization of divided reference in one class of words does not
presuppose the same in any other class.
A further subclass among "divided-reference" terms are those
which denote individuals-often without any designative compo
nent: proper names. (On the semiotic nature of proper names,
cf. S0Tensen, 1958.) In English and many other languages, proper
names have a special grammar; we distinguish, for example,
proper Dolly from appellative the/a dolly, although the gram
matical machinery differs from one system to another. 25 House
holder and Hockett surmise that in every language proper names
are a semiotic type of sign with a grammatical mark of their own.
It seems that no language refers to individual constants of more
than one class, for example, by having "proper verbs" as well as
"proper nouns."
"
'i;
II t
,I
I., .
162
Uriel Weinreich
The West and Central European languages are perhaps atypical

in distinguishing between indefinite descriptions ('a so-and-so,
some so-and-so') and definite descriptions ('the so-and-so'), and
correspondingly between relative superlatives ('the sweetest') and
absolute superlatives ('a most sweeL.'). That this is a common
innovation in Europe is suggested by the further detail that all
these languages use for definite description the same form
(definite article) as for "within-discourse" deixis. Definite de
scriptions are applied only to nouns.
We have surveyed a number of semantic formators; the dis
cussion of purely syntactic formators 26 is postponed for 3.1.3.
2.3. Metalinguistic signs and operation$
:1
For ordinary purposes languages serve as their own metalan

guages. The effort expended by logicians since the Middle Ages
to disentangle the use of signs from their mention is in itself
evidence of how smoothly ordinary language blurs the distinction
between types. It may be useful for certain purposes to isolate
from the vocabulary of a given language those terms whose des
ignata are themselves aspects of language, such as word, say,
conjugate, mean, true (cf. Reichenbach, 1948 :58), but, on the
whole, these have characteristic features of neither grammar nor
phonology. What concerns us here is the question of devices
equivalent to quoting, that is, devices which may distinguish be
tween the use and mention of a sign of indefinite type. Many
cultures (including, according to Hockett, nonliterate societies)
use "vocal quotation marks," manifested by pause and occasional
intonational or voice-qualifying features. But the marking of
type-shift does not, it seems, become explicit and codified except
in writing systems, and even in writing traditions the use of
quotation marks is a relatively recent innovation, which semi
literates find difficult to use correctly.27 Many languages have
expressions like "so-called" or "to wit" ("say" = 'that' '" ),28 but
one wonders whether improvisation does not here prevail over
standardized features. Some languages are reported to have
special "quotative moods," but one should distinguish mere
163
suspension of assertion (2.2.1.1), which can serve numerous

functions, from specifically quotative mechanisms.
Another metalinguistic operation commonly performed in all
cultures is definition, for example, as answers to questions of the
type "What's an X?" So far, it would seem, no language has been
reported to mark defining statements as a semiotic class by any
overt grammatical means. 29
For every language, finally, stock must be taken of all meta
linguistic operators such as English true, real, so-called, strictly
speaking, German eigentlich, and the most powerful extrapolator
of all~like-which function as instructions for the 190se or strict
interpretation of designata.
3. Combinatorial Semantics
3.1. Semiotic structure of discourse

3.1.1. TYPES OF SIGN COMBINATION. If we consider the effect of
signs when they are grammatically combined to form discourse,
we detect two semiotic processes which are not reducible to each
other, and which might be called "linking" and "nesting." The
linking process has been the subject of a large philosophical
literature, but it would seem that it is nesting which offers far
greater theoretical difficulties.
Linking may be described as that effect of a grammatical con
junction of two signs 30 which yields a product of their designata.
Assuming, for example, thatflower has the designatum 'c1 C2 c/
(i.e., the conditions under which the flower denotes; cf. 1.2) and
yellow has the designatum 'c4 CD', then yellow flower, being a
grammatical expression in English, has the compound designatum
'c 1 C2, Ca . C 4 cli '. Similarly, ill (to) walk fast, the designatum
of the expression may be considered the sum of the designata of
walk and fast. The semiotic process is equivalent to Boolean class
conjunction. 31
It is possible to show that linking occurs on various levels. We
describe the effect of conjoining dark
yellow _ dark yellow as
a linking; similarly for yellow flower. Yet in dark yellowflower,
although there is one interpretation of the ambiguous phrase
l1riel Hleinreich
164
which may be described as an extended linking ('something which

is a flower and is dark and is yellow'), there is also another inter
pretation according to which there is something which is yellow
and a flower and "something" which is yellow and dark; but the
second "something" is not the same as the first: it is a color, that
is, a property of the flower. This effect is easily described in terms
of the so-called higher calculus of functions, in which it is possible
to speak of properties of properties. Such a calculus, in other
words, permits not only expressions like 'I(a)' or 'I(a, b)', but
also '~(f)" and even compound expressions like 'I(a)- ~(f), To
transcribe our example, we write 'F' for 'flower', 'Y' for 'yellow',
and 0' for 'dark', and formulate it (following Reichenbach, 1948)
as:
(3x) (3f) F(x) . f( x) . Y(f) . o(f)
The epistemological desirability of a higher functional calculus
is a matter of debate,32 but for describing, not criticizing, the
semiotic structure of discourse in ordinary language, no superior
method has yet been proposed.
If we consider next such expressions as buy flowers or under
water, we cannot say that the effect is an addition of designata at
all. It is as if the designata of buy and under contained open slots
which were harmoniously filled by flowers and water, respectively,
but in a nonadditive way. One of the differences between these and
the yel/ow flower examples stems from the fact that buy and (be)
under are two-place relations:
x buys Y = B(x, y)
w is under z = U(w, z)
But this qualification is insufficient, for some two-place relations,

such as 'resemble', 'be married to', can be explicated as linking.
For cases such as buy and (be) under, it is apparently necessary
to specify further that the two-place relation is asymmetrical, that
IS,
B(x, y) ::> B(y, x)
U(w, z)
::>
U(z, w)
ON THE SEMANTIC STRl1CTl1RE OF LANGl1AGE
165
It would seem that in asymmetrical relations, one argument
"links" semiotically with the function, while the other "nests." In

a semiotic theory involving designation such as was sketched in
1.2, the linking operation can be accommodated in an intuitively
acceptable way; the nesting operation must be introduced by a
special definitional stratagem. 33
We might adopt the convention that in a many-place asym
metrical relation, the first and only the first argument "links" with
the function; but since it is not usually obvious from the notation
whether the relation is symmetrical, it is preferable to introduce
a special mark for the nesting argument, such as 'B(x, y),.
Nesting, like linking, may involve a multiplicity of "levels."
Consider Jim observed the counting of votes. We have 0 = 'a
observes b' and C = 'c counts d'; writing Xl for 'Jim', YI for
'votes' and X for the omitted first argument of C, we have
O(Xb C) . C(x, YI)
In an example like Jim liked to observe the manufacture of lawn

mowers, we have four levels of nesting.
It is possible to interpret the operands of all operations repre
sented by formators (2.2) as being in a nesting relation to the
operators. The notational difficulties which arise in connection
with mixed signs do not seem to be insuperable; and it is hardly
the analyst's fault if language is complicated.
We may now test the theory of two kinds of sign combination
on some examples. Consider the English sentence, The three
bitterly crying children walked home fast. We find a three-place
function (walk) of level I which has as its arguments children,
home, and -ed ('time previous to the speech act'). Children links
with walk; the others nest. The function-name walk itself appears
as the linking argument of another function of level 2, fast.
Whatever is the argument of the O-level function [x is a] child is
also the argument of another O-level function, x cries. This O-level
function is in turn an argument for a level-l function, x is bitter.
We leave open the question whether three and -ed 'past' should be
analyzed as designator functions whose arguments are children
and walk, or as formators, that is, names of nondesignational
166
Uriel Weinreich
'1
:1,
"
IJi'f
i[
I."1
"1
t
"
semiotic operations to be applied to children and to walk. Cer

tainly the involves such an operation ('such children as have been
mentioned in this discourse'). Finally, home has a covert semantic
structure corresponding to 'the children's home', that is, reflecting
a nesting function x has a home and a proposition, the x who
walked is the x who had a home. We dispense with the technical
formulation of the analysis. 34
A simpler example is the Korean sentence, Kim-an s'E-cEk-al
p'a//i i1kat s'amnita 'Kim read the new book fast'. We have a
function,/, translatable as 'x readsy at time t'. The arguments are
x = Kim, y = tEk 'book', and t = -at 'time previous to the speech
act'. Here -an is a syntactic formator showing which is the x
argument; -al similarly shows which is the y-argument. The
functionf = i1kat 'read' is itself the argument of the function cp (f),
where cp = p'a//i 'fast'. The argument y of f(x, y, i) is also the
argument of another function, g(y), where g = s'!;: 'new'. In the
one-place functions cp and g, we find a linking effect; in the three
place function /, there is linking between f and x (and perhaps
with t; cf. the remarks on -ed in the English example) and a nesting
of y "in" f
Finally, we may analyze two famous Nootka sentences. The
expression lash-tskwiq-ista-ma 'select-result-in a canoe-asser
tion' is translated by Whorf (1956 :236) as "they are in a boat as a
crew of picked men." We seem to have a function of the form
f(x, y), in which the linked argument, x, is omittable, as it were:
f( ,y) or, in technical notation, f( X, y). The function f is lash
'select', and tskwiq, rendered by Whorf as 'result', is the marker
of y. But, interestingly enough, y here remains as a variable, and
is "bound" as the argument of another function, g = -ista- 'to be
in a boat', where it is again in a linking relation to the function.
The sentence thus has the form
1- g(y) fO:, y)
The sentence means, roughly: "It is asserted that the selected are
in a boat," or " ... that they, the selected, are in a boat"-whieh is
far more transparent than Whorf's tortured translation. A still
simpler case, involving no second-level function, is the Nootka
167
sentence: 'Aims-ja- ?is-ita-? i-'Ama 'boil-result-eat-agents-go for-he

does', to which Whorf (1956 :242) mystifyingly matches the
English "he invites people to a feast." Again we have a two-place
functionf(x, y), in whichf = ?i'A 'x goes for y', and x is omittable.
Another two-place function, E (x, y), is E = ?is'x eats y'. A
third two-place function, B (x, y), is B = 'Aims' x boils y', again
with x omittable; -ja- marks the preceding sign as being the y of
B, and -ita- marks the preceding as being the x of E. The final -ma
may be interpreted as an assertion operator. The whole sentence
then has this approximate form:
1- G(a, w) . E(w, z) . B(a, z).

We are now prepared to formulate an important hypothetical
universal:
In all languages a combination of signs takes the form of
either linking or nesting, and all languages use both patterns
in kernel sentences. No further patterns are introduced by
transformations. While the number oflevels is not theoretically
limited, linking on more than three and nesting on more than
four is very rare.
3.1.2. COMPATIBILITY OF DESIGNATA. It was assumed up to now,

for the sake of simplicity, that any designatum could be linked or
nested with any other. Yet the doubtful semantic acceptability of
expressions like yellows songs, sour rights, drink ice, etc., suggests
that the designata fall into various types which are not all equally
compatible with each other. 35
Our concrete knowledge of semantic systems is pitifully in
adequate for the formulation of any universals on this point. It
does appear that all languages have incompatible types of desig
nata. It appears further that some of the denotative bases of com
patibility, such as the sense by which something is perceived,
sensory perceptibility in general, spatiality and temporality,
number, etc., are very widely shared (cf. Cassirer, 1923) if not
universal. When two culturally very close languages such as
German and English are compared (Leisi, 1953), it turns out that
]68
Uriel Weinreich
while the compatibility of certain specific near-equivalents in the

two languages differ, the general domains of compatibility are
very similar: although German distinguishes giessen and schiltten
'to pour' according to the liquidity of the object, the distinction
appears in English in other contexts (e.g., eat/drink). But the lack
of data on this point is still abysmal.
No semantic theory would be complete without accounting for
the effects of combining "incompatible" designata. As B. Hru
shovski put it, the combination of otherwise incompatible desig
nata is a standard device of "hypersemanticized" discourse and
may be used by a writer/speaker to force the reader/hearer to
find some new, uncoded connection between the designata. 36 It
would be surprising if in any culture the improvised combination
of "incompatible" signs were unknown; but perhaps different cul
tures, like different literary periods within Western culture, differ
as to the matter-of-factness with which this semantic device is
regarded. In the urban cultures of Europe and America, the
unprecedented semantic experimentalism of modern poetry has
perhaps effected an atypical degree of tolerance for semantic
incompatibility.
3.1.3. MARKING OF SEMIOTIC ORGANIZATION. If our analysis of
sign combinations, or "syntagmatic semiotics," is correct, then it
must be completed by an account of a residue of signs anticipated
in 2.2(5), corresponding to what Reichenbach (l948:318ff.) called
"logical ter~s in a syntactic capacity." These are signs whose
functions is to organize the discourse by marking the argument
names and function-names, the linkings and nestings, the scopes
of pragmatic and semantic operations. These include certain
elements of word order, concord, certain aspects of conjugation
and inflection, as well as the covert, "cryptotypic" (Whorf, 1956:
92ff.) membership of designators in specialized grammatical
classes. But it will be noted that while the syntactic operators are
identified, in an enumeration such as this, by their grammatical
properties, they are not defined by these properties; for the
defining criterion is again a semantic one and agrees only in part
with the grammar (cf. 1.2). For example, in a declension some
cases serve only as syntactic operators, while others have desig-
169
native content. The nominative may be a sign of the subject, that is,
of the function-linked argument, while an illative that contrasts
with an elative combines an expression of nesting with a designa
tion. The semantic classification again intersects with the gram
matical one when we classify noun-verb Hnkings together with
certain verb-adverb linkings. The failure to distinguish these
criteria can lead only to frustration, which in the case of Sapir's
typology of languages (1921) reached truly magnificent propor
tions (cf. Most, 1949).
Among the most controversial problems in this connection is
that of so-called "grammatical meaning." There are those who
claim that the meaning of certain signs (~ formators?) is qual
itatively different from those of others
designators). To quote
but one sample out of scores: "Paarden ['horses'] indeed sym
bolizes 'more than one horse', but -en ['-es'] does not symbolize
'more than one'" (Reichling, 1935:353). In present-day Soviet
linguistics, too, the "Word-Paradigm" model of analysis (Hockett,
1954:90) holds a monopoly, and the possibility that the semantic
role of affixes and stems may be similar is considered an absurdity
(e.g., Budagov, 1958:5 and passim; Zvegincev, 1957:98f.; Sav
renko, 1959 :35ff.; Sendel's, 1959). The opposite view is that
there is no special kind of meaning such as "grammatical
meaning"; there are merely special signs which have the gram
matical (not semantic!) property of obligatoriness. 37 It is our
contention that only the latter position is tenable, as it is the only
one which conforms with the requirement that semantic and gram
matical criteria must be autonomous (1.2).38 The distinction
between material and formal meanings, which has dogged lin
guistics at least since Schleicher (Cassirer, 1923:164), is not only
ethnocentric, but is inapplicable even to Indo-European lan
guages, and should be scrapped. The distinction between auto
,categorematic and syncategorematic signs, in most of its very
numerous interpretations, covertly mixes grammatical with semi
otic criteria and is also totally untenable. This still leaves open the
question of what signs "belong to" the grammar, but whatever
the criteria may be-boundness, obligatoriness, etc.-they are
grammatical, nonsemantic criteria. 39
170
Uriel Weinreich
The specific grammatical properties of signs which impose on

discourse its semiotic organization are a vast subject in them
selves, and the topic cannot even be surveyed here. The one
problem that we wish to raise is the possibility of ambiguity in
semiotic organization.
:\.
\.
i
I,:
'.
1;1
(a) Summation versus linking. It was suggested in 3.1.1. that

linking is effected whenever signs are conjoined in a grammatical
combination. Languages, however, also have explicit linkage
markers, such as and: cf. cozy old houses = cozy and old houses.
But the sign of linkage, whether an and-word or mere con
junction, may be homonymous with a sign standing for another
semantic process equivalent to arithmetical summation; cf. four
hundred and twenty. (On the polysemy of 'and', cf. Biihler, 1934:
317f.; Hockett, 1958: 185f.) There may thus be ambiguities as to
whether linkage or summation is intended: cf. old and experienced
women: 'women who are old and experienced'? or 'some women
who are old and some who are experienced'? Where summation is
signified by mere conjoining, as in early literary Chinese, we also
get "paradoxes" like po mafe; rna '[a] white horse [is] not [a] horse'
'white [and] horse [is] not [a] horse'(Maspero, 1933 :52).
(b) Symmetry of linking. Whereas in a simple calculus of
functions the distinction between arguments and relations is cru
cial (Le., 'f(x)' is "grammatical" but 'x (f) , is ungrammatical),
in a higher calculus of functions, such as ordinary language, the
distinction is of very minor importanc~. If 'x(0)' ('this is an
x'), 'f(x)', '(j) (f)', etc., are all grammatical, it makes little difference
for one-place functions, whether we write 'f(x)' or 'x(f)'. To be
sure, most linguists believe, like Sapir (1921: 126), that "there must
be something to talk about and something must be said about
this subject of discourse once it is selected." But by intrasentence
criteria alone, we can only conclude that (barring minor sentence
forms; cf. 3.1.5) "there must be at least two things to be said about
each other." The determination of which of the "things" is the
"topic" or "theme," which the "comment" or "propos," seems to
depend on which is more surprisingly introduced in the context of
the preceding speech situation. 40
171
(c) Asymmetries of nesting. It is far different when we come to

nesting; here the specialization of roles for relation-name and
nesting argument-name is semiotically decisive. (We will call
'function status' the role of a sign as an argument-name or a
relation-name within a function.) The most usual pattern is ap
parently the specialized marking of arguments by syntactic for
mators, such as nominative vs. oblique case (Latin -us/-urn,
Korean -,m/-al), subject vs. object partide (Japanese -waf-gal, etc.;
fixed order is extremely common, even where it is partly re
dundant with segmental argument markers (e.g., they consider
stopping it vs. they stop considering it). Often a theoretically
possible ambiguity of organization is resolved by the semantic
absurdity of one of the alternative interpretations, or (as Hockett
has hinted) by proportionality with unambiguous portions of the
context. Thus, in the German sentence Die Birnen assen die
Kinder 'it""was the children who ate the pears', we conclude that
despite the unusual order, 'fry, x)', x = Kinder and y
Birnen.
We do so not only because of the absurdity of pears that eat
children, but also because the preceding context may contain, as
a clue, an unambiguous model, such as Der Vater ass die Kirschen,
und.... Many examples of place ambiguities seem to be due to
partial syncretism of grammatical categories, for example, be
tween dative and accusative euch and uns in German; hence,
er hat uns euch empfohlen ('he recommended us to you' or 'you to
us') is ambiguous, but in most instances the case distinction
would take care of discriminating the several nesting arguments of
empfehlen. But in their semiotic functions, grammars are not
100% efficient, and some unresolvable ambiguities do occur, rare
as they are. Chao (1959 :3f.) cites a Chinese sentence in which
a linking argument (equivalent to an English subject) is inter
changeable with various nesting arguments (time and place
specifications). In the Yiddish Hajnt iz sabas 'Today is Saturday',
one cannot tell whether hajnt is a subject or an adverb of time; in
s'kumt ajx a dolar 'you have a dollar coming to you', one cannot
tell from overt markers whether a dolar is the subject or the
object.
172
Uriel Weinreich
It is far more usual for certain distinctions of function-status

to become obscure when sentences are nominalized; cf. the
ambiguity ofLatin subjective and objective genitive (amor Dei < X
amat Deum? or < Deus amat X-um?) or of English visiting
relatives, derivable both from V(x,f) and from V(r, x). There also
occur ambiguities of the form j(a) . g(a)' vs. 'rp(f[a]'; cf. He
decided to leave immediately (= decided immediately? to leave
immediately?).
While the distinction between j(x)' and 'x(f)' in sentences with
one-level and one-place predicates is, as we have said, of minor
importance, the prevalence of functions of more than one place,
j(a, b, c .. .)', and of type higher than one, like j(a, b ... ) .rp(J,
g) .. .', imposes on the vast majority of sentences a determinate
semiotic structure. As pointed out in (b), semiotic considerations
alone would permit us to equate j(x)' with 'x(f)" or even with
'x x' or j 1'; but the productivity of expansion patterns of
sentences (Chomsky, 1957: ch. IV) endows even the <simplest
sentence with a grammatical structure similar to that of complex
ones, which in turn suggests for the simplest sentence a semiotic
analysis analogous to that of higher-level and higher-degree func
tions; it is only the virtually grammarless discourse of pictorial
writing (e.g., Fevrier, 1948 :40 illustrating Ojibwa incantations;
Voegelin, 1961 :85 on Delaware mnemonic pictography) or of the
gesture language of congenital deaf-mutes (cf. Spang-Thomsen,
1956) that resembles the form jlf. ..' in its semiotic "structure."
3.1.4. MAJOR FUNCTIONS AND THEIR BACKGROUNDlNG. It is a

further near-universal property of discourse that in sentences
expressing more than one function, whether homogeneous (Le.,
j(a) . g(a)') or heterogeneous (Le., j(a) . rp(f)'), one of the
functions is represented as the major function. The usual grammat
ical correlate of the major function is the subject-predicate con
struction, but in the verb-ph rase-sentences of polysynthetic lan
guages the semiotic cut is marked in other ways. 41 It seems, inci
dentally, that this universal feature of language is also transferred
to aU logical systems. 4.2 Each language has its own stock of gram
matical devices for "backgrounding" aU but the major proposition
173
of the sentence. It appears to be a universal, too, that in the back

grounding of a proposition some information is lost: the most
general loss is pragmatic-that is, the backgrounded proposition
is not fully "asserted"-but there may also be losses of tense and
subject-object distinctions. It is usual for every sentence to show
a major function, but some languages also have ways of back
grounding all functions in a sentence; cf. English There was a
beating of drums by the natives (see also below).
Every "major" proposition is perhaps capable of being back
grounded, at least as a nesting argument in a verbum dicendi
relation; it is much rarer, on the contrary, for a language to have
designators which cannot participate in a major function at aU,
and are condemned, as it were, to the background; in fact, it
seems plausible that such a sign would be not a designator but a
formator. 43
The high rigidity of j( x)' organization, reinforced by the
grammatical requirement to conform to an extremely low number
of sentence types within any one.language, is universally counter
balanced by the availability of grammatical devices for trans
forming f-signs into x-signs and vice versa. We are referring to
the semiotic effect of deriving verbals from nouns or noun-phrases,
nominals from verb-phrases and sentences-the "stativations"
and "verbations" so graphically sketched by Whorf (1956 :96ff.),
the "event splitting" analyzed by Reichenbach (1948:268f.).44
Related to these are the operations of abstracting a property from
a class, of "solving" functions for a particular argument (Reichen
bach, 1948:311f.)-semiotic processes expressed in language by
relative clauses and their analogs. These are by no means restricted
grammatically to the clause or phrase level; the whole process
may take place affixaUy. As an example we take the Fox sentence
(Sapir, 1921 :76), -kiwin-a-m-oht-ati-wachi 'they together kept (him y
in flight from them'. We introduce the notation a* for an argu
ment-name derived from a function. The Fox sentence has the
following semiotic form:
f
-kiwin- 'indefinite movement'; hence,f(x) 'x moves in an

indefinite way'
i'
I'
I
174
lIriel HVeinreich
cp
-a- 'flight'; hence, -kiwin-a-: cp (f) . f( x) 'x moves fleeingly

x flees'
1
Here we convert the two-level function to an argument:
[cp(f) . f(x)J = a*
g =
h =
g( )
y =
Z
'to cause'
'to be animate'
. h (x) = -m- 'to cause to an animate subject'
-wachi 'they, animate' (the 'causer")
} (manner
-oht- 'for the subject'

w -ati- 'several objects, one to the other' of causing)
'they, animate (wachi) cause a* for themselves
g(y, d*, i, w)
(oM-) to one another (-ati-)'
All together:
g(y, d *, Z,
. {a*
[cp(f) 'f(x)J} . h(x)
In many languages, a limited number of grammatical patterns

may be called on for changing the function-status of signs in a
multiplicity of ways. Consider the ambiguity of English His
dancing was surprising. If this means, 'the way he danced .. .', it is
of the form 'f(x) . cp(f)', where the first function is merely back
grounded; if it means, "the fact that he danced ... ", it has the
form '[f(x) = a*) . cp(a*)', where the first "proposition" as a whole
is converted to an argument. These matters would require a
specialized analysis. Yet it is useful at least to point out the
perhaps universal asymmetry of grammatical devices for nominali
zation and verbalization. Despite the exceptional structure of
Chinese, where the backgrounding.of chaau fann 'fry rice' to
'fried rice' involves no overt marking (Hockett, 1954: 102), it is
safe to say that in most languages conversion of a relation or a
proposition to an argument involves intricate grammatical pro
cesses and losses of information, whereas conversion of an
argument to a function may be accomplished by something as
simple as making the argument the complement of a verb or
particle 'to be'. In Miwok, for example, nominalization of a
sentence requires affix changes in subject and predicate, but any
J 75
ON THE SEMANTIC STRlICTlIRE OF LANGlIAGE
noun can become predicative either by being conjugated directly

or by being verbalized and conjugated as a verb (Freeland, 1951:
136). In English we can contrast the complexity of changes in
volved in the first and second conversion:
f(x)
x
-+
-+
a*
f ()
He often sent flowers -+ HIS FREQUENT

SENDING Of flowers ...
Three truly excellent wines
three truly excellent wines.
-+ ...
are
This grammatico-semantic asymmetry is also evident when we

compare derivationally related verbs and nouns of a language. No
matter how austerely the derivation is marked-even by zero, as
in English-the verb to X only exceptionally means 'to be an X'
(as in to soldier, to sire); much more usually to X means 'to treat
as an X', 'to cover by means of X', 'to perform X' (cf. to baby, to
mother, to people, to police, not to speak of verbs derived from
inanimate nouns). In other words, even "zero" derivation rarely
serves the purely syntactic role of converting an argument-name
to a relation-name (cf. Martinet, 1960: 140f.).
While all languages thus contain the means for overcoming the
specialization of particular designators for particular function
status (and this measure of convertibility may differ from lan
guage to language, achieving a peak in English and Chinese), it
is more than likely that such operations are learned rather late
in childhood; very young children may master 'f(x)' sentences and
very soon thereafter also 'fix) . cp(f)', but not 'cp(f)' as a major
f(x)]' ("it's
function ("the redness is surprising") or 'cp(a*) . [a*
funny for the eyes to be red"), which require special grammatical
transformations. Now the specialization of signs in argument
roles, relation-roles, and operator-roles naturally gives rise to a
powerful ontological metaphor (cf. Marcus, 1960). It is out of
such specialization that "class meanings" (Nida's "linguisemes,"
1953 :5) arise for nouns as "substance-names," verbs as "process
names," etc. Predication, a grammatical phenomenon, comes to
be correlated with one of its most typical, but certainly non
criterial semantic interpretations-"actor-action." 45 R. W. Brown
176
Uriel Weinreich
not only has given proof of the power of children's grammar

based ontology (1957), but has argued that as children grow older
they learn derivations and transformations whose semantic and
grammatico-semantic isomorphism decreases; that is, as they
learn to use higher-level functions as major functions and to
change function-roles of the signs, the foundations of their onto
logy crumble. This brilliant solution to an old impasse raises
fascinating prospects for the cross-cultural investigation of
juvenile ontologies and their possible blurring during adoles
cence. 46
3.1.5. MINOR SENTENCE TYPES. A conscientious separation of
semantic and grammatical criteria also allows us to give a precise
formulation to the old and treacherous problem of minor sentence
types and impersonal verbs. The mere review of the theories that
have been advanced concerning such expressions as Fire! or
It's raining would fill a good-sized book.
A reasonable solution should probably begin by distinguishing
ellipsis from minor sentence types proper. Ellipsis is to be defined
as a family of transformations, with precisely formulated scopes
and functions, yielding the isolation of a part of a sentence
against the background of a full source sentence. 41 (All languages
use ellipsis, under such typical conditions as replies to questions,
conjunction of similarly constructed expression by. and, etc.) A
second type are interjectional nominal expressions, always either
as vocatives, or as symptoms of emotional stress or its conven
tional or rhetorical simulation, of the form 'x!'. (In languages
which distinguish linking from nesting by overt segmental markers
even interjections may distinguish the forms '(x)!' and '(x)!'. Thus
the Roman beggar asking for bread probably said Panem! (accu
sative
'(i)!'), but if he found some bread unexpectedly, he
might have shouted Panis! (nominative = '(x)!').
In contrast to both elliptical and interjectional elements, we
encounter truly "stunted propositions" defined by having a form
like' (x, )' or 'f( , )', etc., in a system which not only permits, but
generally requires 'f(x, y),. Critical logic, finding such forms
inconvenient or conducive to metaphysical pseudoproblems (cf.
Reichenbach, 1948 :89f., 332), rejects these forms as "mean-
177
ingless" and prefers to write not '(tIf)' but '(tIy)f(y)'. But in lan
guages such forms do occur (Martinet, 1960:125[,).
What is thus semiotically "stunted" may receive very different
grammatical treatment, depending on the language. In English
and German, for example, stunted propositions require a dummy
subject it, or even a dummy subject plus is (It rained, It's a boy,
Es wird getanzt); but in other cases the stuntedness is marked by
the subjectless sentence [There was a raising ofeyebrows = 'f( ,yn
In Yiddish, a dummy subject is required only if no other term
occurs in the sentence: es regnt 'it's raining', but hajnt regnt 'it's
raining today'. In most languages (e.g., Latin, Russian, Hun
garian; but not English, Hebrew) the verb phrase alone can func
tion as a full-fledged sentence. Its semantic content does not
thereby lose its regular propositional form; the linking argument,
grammatically deleted, is then a 'he' or 'they' identified by dis
course-deixis. Such is the case in Latin Venit 'He is coming', and
probably also in many polysynthetic languages where the alleged
one-word sentences are really only one-word verb-phrases, func
tioning as a minor sentence type until a subject noun-phrase is
added. Finally, languages which use a copula for converting ar
gument names to relation names may have forms which are
"minor" both grammatically and semiotically; cf. Russian Vojna.
'It's war = There is a war on', Est' stol. 'There is a table', Hungari
an Asztal. 'It's a table', Asztal van. 'There is a table'; 48 in English
this pattern seems applicable only to evaluative adjectives (e.g.,
Excellent). In Chinese, according to Chao (1959:2), minor sen
tences "are more primary and relatively even more frequent" in
two-way conversation than in other languages; but all the
examples, including Feiji. '[It's an] airplane' and Yeoufeiji. 'There
is an airplane', are easily recognizable types. Many languages Seem
to lack grammatical distinction between certain major and minor
forms, such as Miwok soluku- 7 'a bow' = 'it is a bow' (Freeland,
1951 :36).
3.2. Contextual effects on designation
3.2.1. POLYSEMY AND HOMONYMY. We must now refine the
theory of designation to allow for certain contextual effects. In
178
Uriel Weinreich
contrast to the "monosemy" case formulated in 1.2, we now say

that a designatum may contain disjunctions between its com
ponents. Using A, B, ... as signs and (ci C2 ) as their designata,
we define:
Polysemy
A (ci V Cll)
A [c I
(c ll V cll)] etc. 4~
The polysemy of a sign may be resolved by the context,50 as

follows:
.
i'
Resolution
Given A{c l Cll' [C3 V (cll ' cs)]); B; C.

[f A + B, then A(c l cil c3)
If A + C, then A(c l Cll Cll cs).
In this presentation, the signs A, B, C ... need not be words or

even lesser segmental elements; grammatical processes, too, are
given to polysemy which is resoluble by the context of other
processes, for example, the English preterit: 'I. past, 2. (in con
ditions) counterfactual'. It will also be apparent that one of the
important types of polysemy and resolution involves compatibility
types in the sense of 3.1.2. Thus, blue and purple have color com
ponents in the context of signs for visible objects, but these are
replaced by "affective" values in such contexts as... music,
... prose.
It should be clear that by accepting a theory which permits
disjunctions within a designatum, we resolve the controversial
notion of Grundbedeutung or Hauptbedeutung (reviewed, e.g., by
Zvegincev, 1957:215ff., and rejected by him; cf. also the refuta
tion by Karolak, 1960:245-247) into clear-cut operational terms. 51
Before we can think of quantifying the incidence of polysemy
and idiomaticity in a language (cf. 4.3.1), we must also formalize
the distinction between vagueness and polysemy. In most standard
sources these are treated as a matter of degree (e.g., Ullmann,
1951 :119). Black (I949) has given an excellent account of vague
ness in the Peircian sense. 52 Some vagueness is inherent in every
sign, and the vagueness of different signs is not commensurable
I;
179
since vagueness is a pragmatic factor in denotation and hence

beyond the province of semantics as the study of designation.
Ambiguity, on the other hand, is a linguistic, semantic phenom
enon arising from the presence of disjunctions in a designatum. 53
These disjunctions are determinate results of the participation
of a sign in more than one paradigm; thus, taking coat as [cI '
'of arm's length, worn
(Cll V CS)], in which C l = 'garment', Cll
over shirt' and c3 = 'knee-length, worn as the outermost piece',
we have an ambiguity between coat 1 (Cl cll) and coat 2 (Cl CS),
since the classes of objects denotable by each are, in our culture,
discr~te. Similarly for arrange 'put in order' and arrange 'orches
trate'. We would like to propose the term "(synchronic) homon
ymy" for pairs or sets of signs having no element of their
designata in common, like cry 1 'shout' and cry 2 'weep'; 54 fair 1
'not foul' ,fair 2 'not biased', and fair 3 'pretty good'. But even short
of homonomy, polysemous designata differ in "smoothness" (cf.
Gove, 1957:12f.): the designatum [c1 Cll' Ca . (cll V cs)] is more
"smoothly" organized than [cl (c ll V c3 V c 4 V cs)]
In considering the effect of context on polysemy and homon
ymy, we find that signs in question behave very differently. For
coat 1 and coat 2, for example, the ambiguity probably remains unre
solved in most contexts; 55 for cry 1 and cry 2, on the contrary, it is
hard to think of ambiguous contexts in which the homonymy
would not be resolved. As for arrange 1 and arrange 2 or the various
fair's, it is possible to construct both unambiguous and ambig
uous contexts; an example of ambiguity would be: (Was the
weather good or bad?) It wasfair. .often the resolving context can
be specified in grammatical terms. Thus, if cry appears without a
direct object and without out, it is cry 2; if arrange appears without
a direct object, it is arrange 2; if fair appears in a negative
sentence, or modified by very, it is either fair l or fair 2 but not
fair 3 (Ravid, 1961; cf. also N. N. Amosova, SPLS, 1960: 16-18;
V. I. Perebejnos, SSM, 1961 :20-23). The grammatical speciali
zation of the disjunct parts of homonymous or polysemous desig
nata thus hints at differences in their semiotic form: for example,
cry 1 = f(x,;I) versus cry 2 = f(x). 56 In many other cases, however,
the resolving context cannot be stated in grammatical terms and
180
Uriel Weinreich
must be specified in terms of designators (e.g., fair judge implies

fair 2; fair weather implies fair 1 ; fair condition is indeterminate;
cry-baby implies cry 2 while cri-er implies, generally, cry 1).57 The
significant structural problem is to classify the resolving context
words by an analysis of the designata rather than by enumeration.
Thus, we might want to say thatfair impliesfair 2 ifit occurs in the
context of judge, game, decision, warning, etc., but a complete
analysis must find that Ci which is shared by the designata of
judge, game, etc. The widely practiced discrimination of polysemy
by "usage labels" (archaic. poetic; mining, zoology; etc.) involves
pragmatic or even nonsynchronic criteria of dialect mixture, and
-no matter how useful in itself-it is, from a semantic point of
view, beside the point (Zvegincev, 1957:235f.).
The reverse of contextual resolution of ambiguity, of equal
importance to all languages, consists in the capacity of a sign
to evoke a context. The limiting case is the unique constituent:
logan- necessarily implies -berry as runcible implies spoon and
shrift implies short. But highly limited leeway short of uniqueness
is also common. Thus addle, though it does not contain 'egg'
or 'heao' as an actual component of its designatum, nevertheless
implies a collocation with egg (Haugen, 1957 :459) or head, brain,
or pate. Similarly, to neigh implies horse as a subject; and so forth.
This may be called phraseological binding or cliche formation.
We could compute a coefficient of contextual density in a language
based on the incidence of contextual resolution of ambiguity and
cliche formation provided we had an adequate dictionary. Very
likely such a coefficient would hover fairly close to some mean for
all languages of the world. (The coefficient would be similar in
construction to a measure of information content for the average
morpheme or sign.)
3.2.2. DEPLETION. When we contemplate the variety of "mean
ings" which a word like take has in English (take offense, take
charge, take medicine, take notice, take effect, etc.), we come to
the conclusion that this is a case not of abnormally overdeveloped
polysemy of a word, but rather of its semantic near-emptiness. In
these contexts, take may be said to function as little more than
a verbalizer, not quite unlike -ize and other affixes. It is preferable
181
to consider the contextual effect illustrated here not as a resolution

of polysemy, but as a "depletion" of the designatum (Peirce, 1932:
428). Similarly, white in the context ______ wine is depleted, though
perhaps not so drastically, as a result of the limited contrasts of
color-adjectives possible in that frame. Depletion, then, may be
defined as a type of polysemy in which designata contain relatively
large optional parts whose actualization or nonactualization is
determined by precisely delimited contexts.
The phenomenon of depletion is surely a semiotic universal,
but perhaps its incidence varies in different languages. Ullmann,
for example (1952, 1953), without distinguishing it from polysemy,
argued that it is more common in French than in German and
English.
Perhaps every language has a portion of its vocabulary which
is given to depletion. Whether any universals can be formulated
here other than with reference to the high-frequency nature of
the "depletive" vocabulary is not clear. One is reminded of such
phenomena as the verb for 'give' functioning as a preposition
equivalent to 'for' (Mandarin, Thai, French Creole), 'say' as a
conjunction introducing quotations, 'body' or 'bone' as a mark
of the reflexive ('one's self'); 'son', 'eye', 'mouth' functioning as
depleted elements in compounds (Hebrew 'son-of-color' =
'nuance'; Malayo-Polynesian 'eye-of-day' = 'sun'; 'mouth-of-the
house'
'door' in various African languages), etc.
A limiting case of depletion would be that in which a given
context, E + -___ , causes A to lose its designatum altogether:
A is then completely predictable and meaningless (see note 39),
like the dative-case marker in a German noun-phrase after mit.
3.2.3. IDIOMATICITY. An idiom may be defined as a grammati.
cally complex expression A + B whose designatum is not com
pletely expressible in terms of the designata of A and B, respec
tively. 58 (The expression is nowadays often said to be semantically
exocentric, and its meaning is called a "macrosememe"; Nida,
1953.)
Given A (c1 c2 c3); B (c, . c6)
Idiom
Then (A + B) (c1 C, Cs . c7)
:1
182
,' "
.:
Uriel Weinreich
Examples: Finger-hut 'thimble' ("literally" 'finger-hat'), Hand

schuh 'glove' (,hand-shoe'), rub noses with 'be on familiar terms
with'. For any language possessing idioms-and this means every
language-the semantic description is not complete unless each
idiom, whether a compound or a phrase or an incompletely
productive "quasi-transformation" (Harris, 1957: 330f.; Smelev,
1960), appears in the appropriate semantic paradigms on a par
with morphological simplicia and productive transformations.
Thus rub belongs in a "field" with scratch, abrade, etc. ~ nose with
(ai:e, nostril, etc.; but rub noses with familiarity, intimacy, etc.
just as /m::lsruwm/ takes its place in the toadstool-fungus... "field"
regardless of the fact that /m';Js/ plays a separate role (pulp
pap-... ) and so does /ruwm/ (chamber-hall-...). It is often
useful to have a single term for idioms and grammatical simplicia;
"lexeme" is today the most widely used name (Goodenough, 1956;
Conklin, 1962; A. B. Dolgopol'skij uses "megasign" in SPLS
1960:35--42), even though "lexeme" has competing definitions.
In this paper we have been using "sign" to include lexemes and
their non segmental, processual analogs. It would be useful to have
statistics on the distibution of the rnorpheme-to-Iexeme ratio
(index of idiomaticity) in the languages of the world.
It is of great methodological importance to bear in mind the
complementarity of polysemy and idiomaticity.69 For if, having
formulated the designatum of A as (c1 Cj) and of B as (ca' c,,),
we find that A + B has the designatum (c1 c5 Ca . c,,), the re
sulting idiomaticity of A + B may be merely an artifact of our
failure to describe A more correctly as [c1 (c2 V c5 )], that is, our
failure to state that A contains a disjunction leading to polysemy
B. For example, if we
which is resolved in the context _ _
tentatively define charge as 'fill with energy-providing content'
(charge batteries, charge guns), and confront the definition with
the expression charge an account, we may either call charge an
account an idiom or revise the description of charge to show
polysemy: '1. fill ... , 2. burden'. The criteria for choosing solutions
for maximum economy in descriptive semantics have never been
explored, but it is reasonable to suppose that "unilateral idioms"
(e.g., charge an account) would wisely be avoided, whereas
183
"bilateral idioms" like rub noses should be permitted (cf. Mel'cuk,

1960:77f.; 1. S. Toropcev in SSM 1961 :50-54; N. L. Kame
neckajte, ibid.: 55-57).
Many languages seem to have specialized grammatical patterns
for idioms. In English, for example, a preposition plus a count
noun without an article (at hand, by heart) often signals idio
maticity.60 But it is also clear that it is not necessary for languages
to have their idioms grammatically marked. The relation of idiom
marking patterns to productive patterns in a grammar would be
worth investigating on a cross-linguistic basis.
3.2.4. DETERMINATION. When we compare a normal idiom with
the "source" expression in its nonidiomatic sense, we find that
elements of the component designata have dropped out of the
idiom. But there exists a special type of idiom formation which
we may call "determination"; in a sense it is the converse of de
pletion. In this pattern a sign which alone has a highly unspecific
or profoundly ambiguous designatum acquires a more deter
minate designatum in context. The effect is, more often than not,
bilateral.
Given A (c1); B (c 2)
Determination
Then (A
B ) (c1
C2
c.1 c,)
The existence of determination is likewise a universal, but again

languages differ strikingly in their degree of utilizing the device.
In English we find it especially in verb + adverb constructions:
make up, make over, get up, get over, etc. In cases like re-fer, re
ceive, con-/er, con-ceive it is hardly possible to give any designata
at all for the constituent parts. In some Sino-Tibetan languages,
on the other hand, the use of determination is highly developed:
cf. Chinese shih- 'lith-', which becomes determinate in context:
shih-t'ou 'stone', shih-yin 'lithography', etc.; 61 tao- 'road' and -lu
'road'-neither of them semantically determinate or grammatically
free-but tao-Iu 'road' (Maspero, 1933:55), 'success', lu-tao 'road',
lu-t'u 'road' (Sofronow, 1950:72, 76). Thai has biU 'I. cut; 2.
begging bowl; 3. noose; etc.' and bua'lj- 'noose' (grammatically
bound); but bUllrJ-biit 'noose' (free and determinate). In English
)
II
184
Uriel Weinreich
exact analogs have to be invented, such as *poly-mult for 'many',

*poli-urb for 'city'. Formulaically, the Sino-Tibetan type of process
turns out to be a combination of contextual resolution of homon
ymous ambiguity with idiom formation:
Given A (c1 V Cz V c;); B (c2
Then (A
C4
cs)
+ B) (cz).
4. Semantic Structure and Content of Vocabularies

4.1. Bases for comparison
There is hardly anything more tantalizing in the field of se

mantic universals than the question whether there are signs, or
more exactly, designata which are shared by all languages. (The
formators, which certainly show a high degree of universality,
were treated separately in 2.2.) In the practical problem of the
semantic structures of auxiliary international languages or inter
mediary languages for machine translation, the number ofrelevant
natural languages is small, and the amount of discoverable
"universality" is impressive. 6z But a modest amount of ethnolog
ical sophistication will persuade us that for the human race as a
whole, there are not very many universally shared designata. The
story of the shrinking word list of glottochronology (cf. Hymes,
1960 :4-7) shows that even experienced anthropologists may
overestimate the size of such a list, which has now, as a result of
constant reduction, shrunk to about one hundred items. 63
But for one who takes semantic description seriously, even the
items on the emaciated list are not strictly comparable. For can
we say that all languages share a word for 'eye' when, in one
language, the corresponding word involves polysemy with 'sight',
in another with 'middle', in a third with 'power', and so forth?
And even where there are parallelisms between the polysemy pat
terns, the conditions for their contextual resolutions will surely be
different for each language. For this reason, too, comparisons of
the "degree of motivatedness of signs" (the morpheme/sign ratio)
in sets of languages are feasible only where a high degree of inter-
185
translatability (a priori matching of designata) is assured; 84

Ullmann's survey (1953) of a few West European languages from
this standpoint is, contrary to his optimism, virtually useless for
world-wide typological purposes.
A more fruitful approach might therefore address itself to the
distribution, not of complete designata, but of their disjunctive
parts. In contrast to whole designata, the occurrence of their
monosemous parts will probably come much closer to universality.

Many of the items discarded from the glottochronological list
because of the noncomparability of designata as wholes could be
replaced by more easily compared disjunct parts of designata.
However, if semantic analysis is to be carried to its logical
conclusion, we cannot even stop there: we must base our ulti
mate comparison on the distribution of semantic components, or
distinctive features-the various c's, or conditions for deno
tation-which go to make up the formulation of a given designa
tum (1.2).
4.2. Componential structure
It is hardly necessary any more to analyze or to justify the
concept of semantic component. Since the appearance of the
tide-turning papers by Conklin (1955), Lounsbury (1956), and
Goodenough (1956), we have had an opportunity to see this con
cept applied effectively to a number of amenable fields. We pro
ceed here on the assumption that covert semantic components
are legitimate units of semantic description,65 and that, while
there may be no unfailing procedure for discovering such com
ponents, rational decision procedures can be established for
selecting between reasonable alternative descriptions. 66
Since most studies so far have concentrated on particularly
favorable fields such as kinship and color, it is important to
stress that the actual designata of languages, even apart from the
complications of polysemy (3.2.1), depart from the model (c1 '
Ct . C3 .... cn) in various ways. First, the commutability of the
several components is not always perfect, so that they may not
be as fully discrete as the canonic formula suggests. 67 Second,
man makes ample use of his innate capacity for "perceiving
,I
I
;;1
~\
\!,I
,!,;tl
l
I
Ii
!
ft"
1l
~~
"
,~ f
186
Uriel Weinreich
universals" and learns many designata by deriving a gestalt from

instances of denotata; hence the designatum coded in its canon
ical form is for many signs a scientific construct imposed with a
degree of artificiality which differs for various signs. Components
are also distinguished by the degree of their criteriality. Nor
mative terminologies are characterized by the fact that their
designata fully conform to the canonical form (cf. Budagov,
1958:23-29). In ordinary language only some areas are marked by
a degree of "terminologization." 68 An objective measure of termi
nologization for a set of signs might be given by the reliability of
informants' validation of proposed formulations of the designata. 69
[A third respect in which actual designata deviate from the
traditional model is by the presence of nesting within them; cf.
the Postscript.]
We may define as "immediate synonyms" any pair of terms,
A and A', such that their designata differ by one component. 70
(The "perfect" synonym which has been haunting contemporary
philosophy is of trivial importance to' ordinary language.) Where
as in the highly patterned or "terminologized" domains ofvocabu
lary, such as kinship or color, distinguishing components recur
in numerous sets of signs, the bulk of the vocabulary is, of course,
more loosely structured and is full of components unique to
single pairs, or small numbers of pairs, of synonyms. But the
componential structure as such is not impaired. One can therefore
anticipate excellent validations for analyses even of nonterminol
ogized lexical fields. A recent paper by Bendix (1961), in analyzing
a group of English synonyms for 'give', not only isolated a com
ponent common to the entire set (x gives y to z = 'x causes y to
have z')-a component which recurs in such pairs as show: see
drop : fall, make : be-but also separated out features of status
differential between giver and receiver, of casualness, etc., which
distinguish give from confer, grant, and the like. In an analysis of
a group of synonyms for 'shake', components of intensity (shake:
quake), possible voluntariness (shake,tremble), and others were
revealed which recur elsewhere in the vocabulary (cf. throw: hurl,
jump : fall). 71 Among the most impQrtant components to hunt
out in any vocabulary are those which define "dead metaphors":
187
given a designatum [cI C2 < Ca C, )] in which < Ca C4 ) is

a noncriterial component, a dead metaphor might be defined as
a sign having as its designatum (ca' c,). This describes, for example,
the relation of head' most important> top part of body above
the neck' to head 'most important part of...' in head of the table,
head of government, etc.
We may now introduce the notion of semantic continuity. A
semantic system is continuous if for every sign A (c I C2 cn)
/
there is a sign A' adequately defined as A (C l c 2 cn)-that is,
by changing one of the components of the designatum of A (Wein
reich, 1962). Contrariwise, there is a semantic discontinuity when
a change of a component CI to C' I fails to yield a designatum of
some sign in the given language. Clearly, there are in every lan
guage areas of greater and lesser semantic continuity; for example,
the color field is more continuous than the field of folk-zoological
nomenclature.
<
4.3. Applications
We could now put all this theoretical machinery to work ... if
only we had the data. Unfortunately, in the field of vocabulary
we have almost no critically compiled, commensurable data to
go on. As a matter of fact, the description of anyone vocabulary
is so vast a task-even for languages not so hypertrophied as the
West European ones (Weinreich, 1962)-that we must search for
suitable methods of lexical sampling for typological purposes.
For the time being, the best we can do is suggest some of the
variables for which we should plan to sample.
We may distinguish between general, quantitative coefficients
and special statements involving particular semantic components
and their combinations.
4.3.1.
GENERAL COEFFICIENTS
(a) For the average sign of a vocabulary, or a delimited lexical

domain, what is the degree of terminologization? What is the pro
portion of criterial to noncriterial components in the average
designatum?
I
:~
I,',
I
~i
j'
ih'!1:
~,
[':
1.\
"),1
,I
188
Uriel Weinreich
(b) For the average lexical set in a vocabulary, or in one of its

qelimited domains, what is the degree of semantic continuity?
: (c) For the average sign, etc., what is the incidence o.fpolysemy?
What is the incidence of homonymy? What is the average power
of contextual effects, such as ambiguity resolution, cliche for
m'ation, depletion, idiomaticity, and determination?
(d) Is there a typical absolute size to the stock of lexemes of
a language spoken by a preliterate community? Is there a universal
inverse proportion between the inventory of different words and
the number of idioms (Nida, 1958 :286)?
(e) For a vocabulary or a delimited lexical domain, how many
levels of contrast (Conklin, 1962) are there? In English we find
perhaps four and very often fewer. The unspecialized, laymen's
sectors of language simply do not have the depth of structure of
scientific zoological or botanical taxonomy. Is Wallace's hy
pothesis (1961b) correct in stating that regardless of cultural type
or level, institutionalized folk taxonomies do not contain more
than 64
2 6) entitities (with corresponding limitations on the
elaboration of vocabulary), and that this universal limit is
related to the human capacity for processing information? Is
there any connection between the low number of hierarchic levels
of contrast and the low number of types in language viewed as a
functional calculus (3.1.1)? Do languages have specific patterns
for forming superordinates? 12
(f) For a vocabulary, etc., what is the degree of its circularity?
Differently formulated, what is the efficiency of an ordinary lan
guage in serving as the metalanguage for its own semantic de
scription (Weinreich, 1962)7 What is the relation between this
measure of efficiency and the absolute size of a vocabulary?
4.3.2. SPECIAL CHARACTERISTICS

(a) What is the stock of semantic components of a given lan
guage? What are some of the universal components (e.g., 'gen
eration', 'sex', 'light' vs. 'dark', 'dry' vs. 'wet', 'young' vs. 'old',
'alive' vs. 'dead', 'incipiency' vs. 'steady state')?73
(b) What components typically or universally recur in com
bination? In other words, what are the "things" which have names
189
in most or aU languages? Is it not the case that, say, 'sex' and 'age',
'causing to perceive' and 'sense modality' typically appear together
(boy: man :: girl: woman; see: hear :: show: tell)?
What are the typical, recurrent patterns of polysemous
disjunction affecting particular components? Are there languages
which call 'seeing' and 'hearing', 'eye' and 'ear', 'hand' and 'foot',
'elbow' and 'knee' by the same name? Are we correct in assuming
that 'arm' and 'hand', 'leg' and 'foot', 'toe' and 'finger', 'smell'
and 'taste', 'cheek' and 'chin', 'tongue' and 'language', 'youngster'
and 'offspring', 'guts' and 'emotion', 'head' and 'importance',
'heavy', 'hard' and 'difficult' typically participate in polysemy?
Will an adequate sampling of languages confirm the findings
obtained in Europe that in polysemy among sensory terms, the
metaphoric transfer is always from sight to hearing (Ullmann,
1952:297), that space words are always extended to time notions
(e.g., long, short), and never vice versa?
(d) Is it true that among designations for man-made things,
the discreteness of semantic components in a designatum reflects
the definiteness of the cultural functions of the object?
(e) For a given lexical domain, do some languages show a
higher degree of terminologization in their vocabulary than other
languages, and is this related to differences in the attention paid
to the corresponding domain of "things" in the cultures1
(f) Is it true that "the vocabulary relating to the focus (or foci)
of the culture is proportionately more exhaustive than that
which refers to nonfocal features" (Nida, 1958:283)1 How is this
related to the specificity of designata (designatum/component
ratio), the degree of semantic continuity, and the degree of ter
minologization in the lexical domain concerned 1
(g) We may think of simplex signs as standing midway on a
scale between complex expressions, on the one hand, and factorial,
covert components of simplex signs, for the expression of a given
"meaning." Is there, for a given semantic domain, an optimal
level for simplex signs, related perhaps to the neurological and
psychological equipment of the human animal? Are there lan
guages, in other words, where 'round', 'bright', 'soft', and the like
are expressed not by simplex signs, but have to be rendered by
i~
'.
r
"
190
Uriel Weinreich
complex expressions or result only from the factoring out of

components from among more specific designata? 74
(h) Is there a way for finding objective support for the grand
characterization of the semantic "plan" of a language, its "cog
nitive style" (Hymes, 1961), to replace the highly impressionistic
procedures of Whorf and his disciples?
S. Conclusion
We have attempted to suggest a number of universals of lan

guage in the framework of a consistent and comprehensive
semiotic theory. At many points in the undertaking, especially
where designators were concerned, reference data were felt to be
so scant as to make the conclusions unattractively general and
modest in relation to the conceptual machinery. But certain
over-all patterns nevertheless emerge. Perhaps the most impressive
conclusion is that languages are universally less "logical," sym
metrical, and differentiated than they could be if the components
and devices contained somewhere in each system were uniformly
utilized throughout that system.
The greatest challenge arising from this finding of a property
of "limited sloppiness" in language is to determine what good it
does. Man demonstrates somewhere in every language that he is
capable of greater symmetry and discrimination than he employs
in the average discourse. We want to consider why this should
be so.
Very likely the answer will be found in the ratio between mem
ory capacity, attention span, accessibilty to recall, and effort of
discriminatory coding. We can imagine a'small office with a chair
to sit on and a desk to write on; its occupant may prefer, when
reaching for a book from a high shelf, to stand on a chair or even
to put a chair on top of the table, rather than further clutter up
the office with a ladder. A similar economy may account for the
unequal utilization of some semiotic potentials of language and
the overburdening of others.
But before such interrelations can be studied with precision,
we must have large amounts of empirical research. Above all,
191
there must be a clear-cut realization that the province of linguistic

semantics is the study not of denotation or reference, but of the
designational system proper to each language. 75
The distinction between denotation and designation, which is
at least of medieval origin, was prevalent in nineteenth-century
linguistics as the doctrine of "inner form" (cf. Funke, 1932;
Zvegincev, 1957: ch. VII) and reappears in (post-)Saussurean
linguistics as the distinction between content form ("valeur") and
content substance or purport (e.g., de Saussure, 1922: 158f.,
Hjelmslev, 1953 :30f., and, in brilliant practice, Lounsbury, 1956,
Goodenough, 1956). It also turns up in the modern philosophy
of language in many guises (J. S. Mill, Frege, Husserl, Marty;
Peirce, 1932: secs. 39Iff., Carnap, 1942,1947, Quine, 1953: ch. II).
Although it has a venerable tradition behind it, the emanci
pation of designation from denotation in our own time has come
under attack from various quarters. "Mechanistic" linguists, cap
tivated by early behaviorism, have protested that intensions as
psychic states are inaccessible to observation, and that de
scriptive semantics must wait until further progress in neurology
makes them accessible (e.g., Bloomfield, 1933:140); meanwhile,
all the linguist can do is observe "co-occurrence" between signs
and their assumed denotata (e.g., McQuown, 1956). In the face
of the difficulty of defining words by means of other words, lin
guists have been urged, in the spirit of Wittgenstein (cf. Wells,
1954), to look, not for the meaning of words, but for their "use"
in the language. (But "use" with respect to what?) Under the
influence of information theory, linguists have been urged to cal
culate the transitional probabilities between words and to con
sider these the "linguistic meaning" of the words (e.g., Joos, 1950:
356). Some philosophers have argued the need to eliminate
intension (designation) in the interests of ontological economy.
This may be a laudable critical proposal for constructing a lan
guage of science; still, the workings of ordinary language cannot
be described without intension. The philosophers themselves keep
running into the crude fact of structured designation. Frege and
Peirce faced it in connection with modal-logic problems; it keeps
cropping up in more recent literature under the headings of
192
Uriel Weinreich
indirect quotation, "oblique discourse," "referential opacity"

(Quine, 1960: 14 I ff.), "intensional structure" and "intensional
isomorphism" (Carnap, 1947 :56f.). All alternatives to the clas
sieal theory of language, when applied to ordinary language, turn
out either to evade or to obscure the important issues. 76 Bloom
field's neurological "reductionism," apart from its dependence on
potential discoveries which may never be made, misses the
properly linguistic, "autonomous" structuring of man-made se
mantic systems (cf. Wells, 1954:118-121); for "circumlocution"
is not, as Bloomfield thought, a "makeshift device" for stating
meanings, but the legitimate device par excellence. The slogans
of British philosophy, useful in sensitizing the linguist to certain
subtleties in the polysemy of folk-epistemological terms, hardly
compel us to abandon the semantic description of large trans
lucent segments of vocabulary.
Decades have been wasted. Linguistic semantics must free it
self from the paralysis imposed on it by a misguided positivism
insensitive to the specificities of language. Behavioral data? By
all means, let us have behaviorism rather than a new scllOlasticism
operating with inaccessible "dispositions to respond" (Morris,
1946; Carnap, 1947; Quine, 1960), let us have the observable,
publicly verifiable performance of human beings charged with the
metalinguistic task of manipulating signs for the disclosure of their
intensional structure.
Postscript 1965
The author's work on semantic problems undertaken since the
preceding paper was written in 1961 is reported in several recent
articles, references to which have been added to the bibliography
(Weinreich, 1964, 1965). The last-mentioned paper also takes issue
with some of the recent literature. A special paper (Weinreich,
1963) is devoted to Soviet lexicology.
In the treatment of "combinatorial semiotics," the present
chapter parted company with much of traditional as well as
modern logic by claiming that complex signs are constituted out
193
of simplex signs, not by a single process ("linking"), but by at

least one other irreducible mechanism ("nesting"). The author's
most recent paper (1965) not only pursues the differentiation of
semantic relations which hold between the components of a com
plex sign but takes the further step of claiming that many simplex
signs contain within their designata components that stand in
mutual relations other than linking. The more recent work thus
argues for a formal continuity between the definitional sentences
of a metalanguage and the sentences of the object language.
The 1965 paper also attempts explicitly to integrate a semantic
theory with a generative conception of syntax. It criticizes the
unsatisfactory, basically traditionalist approach to combinatorial
problems recently taken by Katz and Fodor (1963) and explores a
theory in which a sentence draws its semantic components not
from its lexical items alone but also from some of the syntactic cat
egories utilized in its formation. It thus suggests an abandonment
of the requirement stated in the present paper (sec. 1.2) to the
effect that a grammatical description be autonomous vis-a.-vis the
semantic description of a language. The recent version of the
author's theory also deals with the semantic interpretation of
deviant expressions; it seeks a way of overcoming the prejudices of
generative grammar against deviant expressions and its helpless
ness in dealing with the situation up to now.
The question of empirical validation of semantic analyses
against informants' reactions has recently been considered, along
with numerous other methodological matters, by Zimmer (1964)
and Bendix (1965). The latter study, which takes a significant
stride forward in the componential analysis of general vocabulary,
has succeeded in isolating a number of fairly abstract semantic
components which recur in several unrelated languages, and may
'yell be universal.
Notes
1. This particular term is adopted from Morris (1946), but withoutthe
pseudobehaviorist elements of its definition there.
194
Uriel Weinreich
2. The further "components" of the sign in the Peirce-Morris tradition

-"interpreter" and "interpretant"-are dispensable in the present
discussion.
3. We consider the distinction between denotation and designation
to be essential to any workable program in semantic research.
See also sec. 5. For a highly readable version of the theory of
conditions, see Ziff (1960), ch. lll.
4. The priority of grammatical over semantic description is conceived
of as a feature of the theory, not as a necessary sequence of dis
covery procedures. (Curiosa like "Jabberwocky" aside, linguists
in practice describe a grammar on the basis of texts which are, on
the whole, understood.) We observe that among the sentences (and
certain sentence sequences) generated by a grammar, some are
distinguished by being semantically unacceptable. The semantic
description of a language is adequate if it so formulates the
meanings of signs that we can pre'dict from an inspection of this
formulation that a sentence containing certain signs will be
semantically unacceptable. In Laxuti et al.'s approach (1959),
the distinction is between significant (osmys/ennye) and insignif
icant well-formed formulas, but for the analysis of real languages,
a stronger criterion of semantic unacceptability should be chosen,
e. g., literal absurdity, self-contradiction, and tautology. One of
the best tools of semantic analysis of a language is therefore a set
of skeletal sentences which, if their slots are incorrectly filled,
are especially likely to be semantiqdly unacceptable ( " - - is
a kind of---," "it's a - - - but.it's ---," etc.). The investi
gator can, of course, also get informants to perform explicitly
metalinguistic operations, e. g., arranging terms in semantic
paradigms (antonyms as well as multidimensional sets), ranking
synonyms for semantic distance, and even supplying (or evaluating)
definitions of terms. While the ideal form of a semantic rule is a
definition, many significant elements of a natural language
lend themselves to definition only partly or awkwardly
(cf. Weinreich, 1962); but a place in a semantic paradigm can be
formulated even for hard-to-defineor undefinable elements. But
this leads to problems of method which are beyond the scope of
this paper. [See Postscript.]
5. In Karolak's succinct formulation (I 960: 246), "the sememe ...
arises as the result of a given unit's entry into specific paradigmatic
functions, but the given unit is empirically a member of as many
paradigms as there are syntagmatic positions in which it is able
to function."
6. We can, for example, profit from Reichenbach's unexcelled "anal
ysis of conversational language" (1948: ch. Vll) without joining him
195
in his condemnation of the tendencies to "analogy" (p. 278) or

"equalization" (p. 263) of ordinary languages, or in his blame of
the German language for the "mistake" of deriving adverbs
from adjectives without an overt marker (p. 302). Logicians
have unfortunately shown little sophistication in distinguishing,
among the "defects" of ordinary language which they have been
seeking to overcome, those which are universal (e. g., those
which lead to the antinomies) and those which are specific to
particular languages. Thus the ambiguity of English is (class
membership vs. predication vs. identity) does not arise in the
same form in languages in which "adjectives" are "verbs." The
whole problem of definite descriptions would hardly have occurred
to logicians starting, let us say, from Russian or Latin or any
other language without articles. In order to keep the descriptive
and the critical enterprises distinct, we speak of the "semiotic
form" of expressions rather than their "logical form," lest we be
forced into the awkward conclusion that some logical forms are
illogical.
7. Reichenbach's most serious error, of course, is his desire to rewrite
grammar on logical principles (e. g., 1948: 255 and passim); but
his fulminations against "traditional grammar" should not blind
us to the incisiveness of his semantic analyses.
8. Thus, Sechehaye (1926) is tremendously disappointing, in view
of its title, since only French and a few other European languages
are sampled. Nevertheless, Sechehaye did see the equivalence
of the semantic relations adjective : noun :: adverb : verb (p. 64)
and came close to our own formulation of linking vs. nesting
(see 3.1.1) in terms of the distinction between "complement
intrinseque" and "extrinseque" (61-79, esp. 71f.). Schmidt's
logic (1959) is more subtle, but his material is drawn from one
language only (German).
9. As Morris puts it (1938: 21), " ... the formalized languages studied
in contemporary logic and mathematics clearly reveal themselves
to be the formal structure of actual and possible languages of the
type used in making statements about natural things; at point
after point they retlect significant features of language in actual
use." Whorf (1956) greatly exaggerated the cultural relativity
of logic by overlooking the most general patterns of sign combi
nation at the expense of the arbitrary structure of designators;
cf. also 3.1.3. A fascinating attempt to construct a logically eco
nomical language (Loglan) is described by J. C. Brown (1960).
On a logical language for structural organic chemistry, cf. Laxuti
et al. (1959); for linguistics, V. V. Ivanov in Pylannja ... (1960:
5-8); for geometry, Kuznecov et al. (1961).
196
Uriel Weinreich
197
tudinal components in mixed signs are made by Sapir (1915)

10. Carnap, for example, speaks (1947: 85) of "the customary distinc
and by Zvegincev (1957: ch. VII).
tion between logical and descriptive (nonlogical) signs." The
17. 'Deixis' seems to be the traditional linguistic term, used in the
various combinatory logics which strive to overcome this distinc
major relevant studies by Brugmann and by BUhler (1934).
tion (cf. Rosenbloom, 1950: l09ff.), whatever'their merit may be,
Morris (1938: 17f.) adopts the Peircian term 'index'. Jakobson
are clearly less similar to language than more traditional logistic
(1957)
took Jespersen's term, 'shifter'. Reichenbach (1948: 50)
systems.
speaks of 'token-reflexive words'. On the overriding importance
11. The designers of Loglan (cf. note 9) have assigned specialized
of deixis to communication, cf. Bar-Hillel (1954).
canonic forms to signs depending on their semiotic function
18. In Korean, as Hockett has pointed out to me, there is a six-way
in a manner which strikes the intuition as remarkably familiar
differentiation of forms (largely inflectional) depending on the
and acceptable (cf. J. C. Brown, 1960: 58f.).
relative
status of speaker and addressee; intersecting that, there
12. Of the three subdivisions of semiotics, pragmatics is the least
is a two-way differentiation said to depend on the relative status
well defined. A reading of Morris (1938, ch. IX) shows the lack of
of the speaker and what is spoken of.
clarity, and every writer since has made his own outline. Carnap
19. Not all languages are equally asymmetrical on this point. While
in 1955 relegated all descriptive semantics to pragmatics (1956:
English his canoe corresponds both to he has a canoe and he had a
233) and argued that "there is an urgent need for a system of
canoe, Potawatomi (according to a comment by Hockett) extends
theoretical pragmatics" (1956: 250). Lounsbury (1956: 189) and
the "preterit" inflection of verbs to that of possessed nouns. But
Jakobson (1957) outline more-or-Iess individual classifications.
probably no language distinguishes more tenses in nominalizations
The delimitation here is perhaps ad hoc, too, although it owes
than in kernel sentences.
much to Reichenbach (1948).
20. Nootka does have a verb stem qwis- 'to do thus' (Sapir, 1921: 181).
13. In languages where the imperative has no distinctive verb form
In some Serbocroatian dialects, according to P. Ivic, onoditi
and the deletion of the subject is its only overt mark, there is
'to that' is used (a) as a euphemism for "unmentionable" verbs,
ambiguity between assertions and commands in which the subject
(b) as a pause filler, or (c) as a pro-verb to avoid repetition. As
is "reinstated" for emphasis, as in English You eat this soup.
Hockett has remarked, in Chinese the addition of the "contin
A written phrase like PEDESTRIANS KEEP OUT is "pragmat
uative" -je also creates verblike forms from deictic adverbs:
ically" ambiguous at least until intonation shows whether pedes
dzemmaje 'to do it this way', nemmaje 'to do it that way'. But
trians is a separate clause.
such forms seem quite rare nevertheless; they suggest the one-way
14. An alternative formulation, suggested by Bolinger, is that asser
implication that no language marks more deictic distinctions in
tions and questions are poles on a gradient scale which also
the
"verb" than in verb and noun modifiers.
includes hesitant assertion in its middle ranges. Since "No" may
21. On affixal negation, see now Zimmer (1964).
be said in response not only to a question but also to an assertion,
22. Cf. Lees' class Preverb (Lees, 1960: 18ff.). Reichenbach's objection
it is not surprising that the maker of assertions sometimes allows
(1948: 308) that "the word 'not' ... is classified by many gram
his speech to reflect his anticipation of a negative response:
marians as an adverb; but it is a logical term" illustrates all the
15. Katz and Postal (1964: 152, note 29) disagree with the observation
defects of his book: in which language? by what kind of gram
that interrogative prepositions, although conceivable, do not seem
marians? and why cannot a sign be (grammatically) an adverb
to occur. They claim instead that such things cannot occur. Since
and (semantically) a logical term?
their theory contains no grounds for such a deduction, the
23. Thus Potawatomi has a verb-forming suffix -kuwunukut, which,
"disagreement" strikes me as spurious; the only scientific
added to a numeral root, yields a verb 'to be for ... years' (Hockett,
advance reflected in their formulation is an increased contempt
1948: 214).
for facts.
24. A special study is needed concerning grammatical devices used by
16. Thus, among children there may be the opposite association
languages to convert mass nouns to count nouns and for sub
between bigness and approval. Hockett has pointed out that
jecting
proper names to specification by definite description.
Potawatomi is actually a language with two productive dimin
There is no reason to assume that articles are utilized in the
utives-one of endearment on occasion, the other pejorative.
English way. For example, while in English the mass/count
A number of interesting remarks on the study of "covert" atti-
,i
198
:1
Urie/ Weinreich
difference is neutralized under the definite article (an/some iron:

the iron), in the NE dialect of Yiddish it is maintained by gender
differences (der ajzn 'the [piece of] iron' : di ajzn 'the [kind of]
iron'). In Hebrew, though there is a definite article, ha-, it cannot
be used for constructions like 'the Jerusalem which I remember':
a demonstrative has to be substituted (ota jerusa/aim ... ).
25. The following illustrations have been supplied by Householder:
In both Latin and Ancient Greek, place names have certain
special forms. Personal names have "natural" gender, so that,
e. g., G/ycerium is feminine (not neuter). In Greek the article is
optional with personal names in many environments where it is
obligatory with common nouns; Latin men's names have a
special three-part form, one part of which must be selected from
a very small closed list. In classical Greek the vast majority of
men's names are two-part compounds of a type which almost
never occurs except as a name. In Modern Greek many men's
names have a special vocative (in -0), and all are subject to pre
fixation by certain elements (barba-, kapitan-, kir-, ay-, etc.)
which cannot be attached to nonnames. In English, too (and in
many modern languages), there are morphemes like Mr., Mrs.,
Dr., Prof, etc., whose domain is complete personal names or
family names, and others (like Sir, Dame) whose domain is given
names. In many languages (not all), names form an infinitely
expandable set, such that any phonologically possible stretch
will be accepted without hesitation as being a name by all native
speakers. This is true of virtually no other class of morphemes
(though almost so for plant-names or the like in English). In
Turkish and Azerbaijani there are certain constructions from
which either personal names or all names are excluded (construc
tions involving indefiniteness; names are automatically definite,
like personal pronouns).
26. Carnap suggested to Reichenbach (1948: 325n.) that semantical
and syntactical formators be treated together, but we find it useful
here to maintain Reichenbach's original division.
27. American store windows are full of homemade signs like "HOT
COFFEE" TO TAKE OUT.
28. So Sanskrit iti and Turkish diye for all kinds of express or implicit
quotations, and Classical Greek has for many types of implied
and a few kinds of express quotations (Householder).
29. It has been suggested that definitions are sentences with an "equa
tional verb" representing a logical equivalence ('a _ b'), i. e., 'every
a is a b and every b is an a'. See, for example, Ziff (1960: 168ff.)
on the difficulties of this conceptiqn as applied to ordinary
language [cf. Postscript].
199
30. We need not take "conjunction" to imply temporal consecu

tiveness in the speech event. In a conjunction of signs'A + B', 'B'
may be a process "applied" to 'A'.
31. Ever since the logical phenomenon of predication was emancipated,
at considerable effort, from certain weB-known metaphysical

impasses, the semiotic process involved seems to have become
perfectly formulable in a theory of designation.
32. Concerning some typical philosophical objections to a higher
calculus, cf. Rosenbloom (1950: 87) and Smart (1949).
33. The procedure is roughly as foBows: 'Linking' and 'Nesting' are
taken as "metarelations" between a relation and an argument
(cf. Reichenbach, 1948: 229ff., 320). Writing 'Lk' for 'linking'
and 'Ns' for 'nesting', we define, for any formula '~(f, ...)',
Lk ("(.f) = DfC3z) ,,z) . f(z)
Ns ("(,f) = DfLk ("(.f).
It appears necessary that in a formula of the form '''f, g, h... )'.
at least one argument link with the function:

("() (f) (g) (h) ... Lk(,,(, f)
V Lk ("(, g) V Lk ("(,
h)
If the relation is symmetrical, there is no nesting:

("() (f) (g) "( (f, g) . "( (g, f)::::> Lk (,,(, f) . Lk ("(, g)
The linking operation is equivalent to the classical S-P operation.

In some respects, of course, the replacement of the Aristotelian
sentence formula, 's is P', by a logistic version which permits
polyadic relations; e.g., 'R(a, b)" has brought about, among other
technical benefits, a better approximation of the structure of
ordinary language; cf., e.g., Reichenbach (1948: 83), BUhler
(1934: 370). But to saY,as BUhler does, that "das logistische Schema
a Rb .,. [symbolisiert] zwei Relationsfundamente, die des S- und
P-Charakters entbehren," is accurate only for an uninterpreted
calculus; in ordinary language, either 'a' and 'R', or 'b' and 'R',
or both, remain very much in an S-P relation. This point has been
widely overlooked. Quine, for example (1960: 106f.), passes it by;
so does Sechehaye (1926: 72). Reichenbach (1948: 229ff., 320)
treats the "metarelation" between a relation and its argument as
a dispensable constant; but the constant is by no means dispensa
ble, since there are at least two noninterchangeable "metarela
tions."
A number of interesting problems arise in this connection which
have not yet been investigated. It seems, for example, that some
languages distinguish by grammatical devices a more permanent,
Uriel Weinreich
200
'I
,'I
'~!
'1
1
i
,t
"nomological" linking or nesting of argument with relations from

a more fleeting one; cf. Korean k'oc-a/1 tal-e p'imnita '[the] flower
grows [by nature] in [the] field' vs. k'oc-i... .... happens to grow';
Russian on bolen 'he is sick [now]' vs. on bol'noj 'he is [a] sick [per
son)'; English he is bumming vs. he is a bum; Polish 0/1 spiewak
'he is a singer' vs. on spiewa 'he sings' vs. on spiewa sobie 'he is
[casually] singing'; Chinese woo jia 'my [inalienable] family' VS.
woo de juolz 'my [alienable] table' (Hockett, 1958: 187). Defining
a change of a relation 'f(x, y)' to a derived relation '/'(y, xl', we
may study the use, in a given language, of overt and covert devices
for representing 'f" (overt: passive voice, e.g., 'x sees y' -+ 'y is seen
by x'; covert: antonyms, e. g., 'x gives y to z' -+ 'z receives y from
x'; 'x is under y' ...... 'y is over x'). Defining a change of a relation
'f(x, y)' to a derived relation with fewer places 'f"(x)', we may
again survey the overt and covert devices used by a language for
reducing the transitivity of relation signs (overt: 'x writes y'
'x writes', 'x is in y' -+ 'x is inside'; covert: 'x says y' -+ 'x talks').
Special treatment seems to be given to the semantic paradigm of
relation-names which, though asymmetrical, nevertheless link
with both argl,1ments (seem, constitute, form, resemble, etc.). It
also appears that when the designatum of an asymmetrical relation
name of language L is formulated in the metalanguage of semantic
description, ML, the corresponding sentence in ML must also
contain a nesting relation, although the converse is not true.
34. Reichenbach's formulas are often simplified in that the time
argument is omitted in the notation for verbs. The full analysis of
the present example would be roughly as follows. We write
'Walk' for '~ walks to y at [time] t; 'Fast' for '1 is fast at t;
'Child' for '~ is a child at
'Cry' for '~ cries at t; 'Bitter' for
'~is bitter at t; 'Home' for 'x' is a home at f'; 'Have' for 'x has
y at t; 'Mention' for 'x is mentioned in y' (where 'j? is an act of
speech), '3' for 'three'; '0' for 'this discourse'; 'Prec' for 'x pre
cedes y'; and' II)' for 'the time of 0'. We now have: (3X) (3y) (3/)
r;
(3J) (3g).
3 Child (x, I) . f (x) . Cry (f, t) . Bitter (f, t) Mention (x, 0)

g(x) . Walk (g, y, t) . Fast (g, t).
Home (y, t) . Have (x, y, t).
Prec (t, to)
[Cf. Postscript.]
35. I t is assumed here that the rules that are needed to exclude "seman
tically" unacceptable expressions are different in nature and in
form from grammatical rules; i. e., that a semantic description
201
of a language is an autonomous enterprise and not merely a
continuation of the grammatical description. [Cf. Postscript.]
36. Cf. Jakobson's demonstration (1959) of how a sentence cited for

its multiple and profound absurdity can be decoded by a sympa
thetic analyst through ad hoc modifications of the code.
37. Cf. Hjelmslev (1953: 28): "Thus we must not imagine, for example,
that a substantive is more meaningful than a preposition, or a
word more meaningful than a derivational or inflexional ending."
Similarly Chomsky (1957: 104f.). But we do not find it possible
to accept Hjelmslev's view that "in absolute isolation no sign has
any meaning." On the other hand, there remains the legitimate
problem of why most languages prefer "grammatical" to lexical
devices for the expression of certain meanings (P. Ivic).
38. Weinreich (1959: 335). In the formulation of this view I have
profited from stimulating discussions with Karl E. Zimmer.
If we have difficulty in seeing how a suffix (e. g., Dutch -en)
designates plurality, it is for two reasons only: (I) the meaning
of grammatical categories is often abstract (though 'many' in a
suffix is hardly more abstract than the word plurality, and in
entiti-es the base surely has a more abstract meaning than the
suffix); (2) the phonemically slight forms of affixes are conducive
to a high degree of homonymy; yet -s 'I. (with nouns) many;
2. (with verbs) one' is hardly more fractured than, say, the French
free form sa 'I. hundred; 2. without; 3. blood; etc.' It is to be
noted that in many languages, formators can be converted to
argument names; cf. dix _ dizaine, and _ addition, or _ alter

native, if - (an) if.
39. See Gleason (1962). Savcenko's attempt (1959: 43ff.) to replace

Sapir's grammatical criteria of parts-of-speech classification by
semantic ones seems belated; but his bold formulation of word
"
class universals, denied by Sapir, may be quite correct. In this
connection it is useful to refer to the relation between obligatoriness
and meaning. By a well-known principle from information
theory, what is completely predictable carries no information.
But while we may identify lack of information with meaning
lessness, we can identify presence of information only with
meaningfulness, not with meaning (Carnap and Bar-Hillel 1953).
It is in this sense that we can easily avoid Ziff's dilemma (1960:
182ff.) over the difference between 'having a meaning' and 'having
meaning'; it is striking that the negation-'having no meaning'
covers both a grammatical and a substantive syncretism of the
distinction. Insofar as a morpheme is completely redundant with
respect to some others (as Karolak, 1960: 246, put it, the morpheme
lacks paradigmatic function), we would call it meaningless (cf. also
202
"
!B,
,i'11i
.;
Uriel Weinreich
203
Zitf, 1960: 41). This would imply that the dative case in German,
Extrapolation from grammatical classes may, of course, create
for example, is meaningful when commutable with the accusative
not only abstract ontological types but also concrete designational
(e. g., after in) and meaningless otherwise (e. g., after mit). Though
classes, e. g., extrapolations from gender to sexuality.
this invalidates the search for Grundbedeutungen of cases, it is a
46. The increasing approximation of adult semantic behavior in
necessary consequence of the autonomy of grammar and semantics.
children of various ages has been the subject of recent studies by
40. A crude formulation of the greater surprise factor of the "com
Z. M. Istomina (~emjakin, 1960: 76-113), Flavell and Stedman

ment" or "propos" would be this: in a sentence of the form
(1961), and Ervin and Foster (1961).

'fl(xS, the "comment" is 'ft' if the preceding context was 'h,(Xt)',
47. For example, Lees (1960: 103f. and passim). A useful corollary
and the "comment" is 'Xl' if the preceding context was 'fl(xs)'.
would be the definition of a word as a minimum free form,

To determine which is the comment by intrasentence criteria
"free" implying not a vague "ability to be said alone" but the
would probably amount to a parasitic semantic interpretation
precise ability to be isolated by the stated ellipsis transformations
of a grammatical fact. (This note was stimulated by an objection
of the given grammar.
of Hockett's to an earlier formulation.)
48. All attempts to analyze such sentences in a binary way by means
41. We assume that in the Nootka examples analyzed earlier, there is a
of covert arguments must be adjudged abortive. If, for example,
distinctive hierarchy of structure, namely 'the selected are in a
a covert 'now' is a nesting argument of the 'rain' relation, it also
boat' and not 'those in a boat are selected' ; 'there is a going for
occurs in a backgrounded function in every other sentence using a
eaters of boiled stuff' and not 'there is a boiling of stuff for fetched
present-tense verb. If the 'rain' relation is linked with the covert
eaters'.
argument 'outdoors' or 'this', then every argument in every major
42. In Reichenbach's words (1948: 26), the expression 'a:::> b:::> c' is
sentence form may also be linked with some 'this' by a deictic
meaningless because it has no major operation. But is a semiotic
operation. As Householder has pointed out, some "weather
system without major functions inconceivable to human thought?
expressions" in some languages may nevertheless have a covert
43. This would be a reason to consider numerals in English, for
argument; cf. it is sunny, the weather is sunny; but only it rains,
example, as formators; note the awkwardness of the boys were
not *the weather rains.
five- a major function-as against five boys ...-a backgrounded
49. For a formal approach to polysemy as an intersection of semantic
"function" or a propositional operation. The differentiated
fields, see Laxuti et aJ. (1959: 217f.) and I. I. Revzin in SSM
"adverbial" affixes of polysynthetic languages might be interpreted
(1961: 17-19).
as designators restricted to minor function-cf. Nootka -ista
50. The fullest treatment appears to be by Zawadowski (1958; 1959).
'in a boat', Comanche pi"- 'with buttocks', ta'l- 'with the foot'
This well-worn idea (cf. Paul, 1880: 56; Wegener, 1885: 84;
(Casagrande, 1954: 148); but these suffixes may also be quasi
Breal, 1897: 141f.) has recently been proposed by Joos (1958) as
allomorphs of nouns for 'boat" 'buttOCkS', 'foot',
the foundation of a new science, "semology." It is being studied
44. Many interesting examples are adduced by Marcus (I 960), although
systematically by the Cambridge Language Research Unit
it is surprising to see a metaphysi~s erected on a linguistic basis
(c(. Masterman, 1959) and by Ju. D. Apresjan (1962).
where. human language is represented by a sample of one (German).
51. The problem of invariance under polysemy is approached from a
45. So greatly delayed has been our understanding of the relation
formal point of view by S. K. ~aumjan (SPLS 1960: 21-25) and
between grammar and combinatorial semiotics that even Bloom
by I. I. Revzin (SSM 1961: 17-19).
field, who would hardly have called an adjective a 'quality word',
52. "A proposition (or any other symbol) is vague when there are
did not mind mixing semantic and.grammatical spheres in calling
possible states of things concerning which it is intrinsically un
a predication the 'actor-action' construction. (Similar defects mar
certain whether, had they been contemplated by the speaker, he
Nida's treatment of episememes [1957: lOf.J.) For this mixture of
would have regarded them as excluded or alIowed by the proposi
metaphors, tolerable only so long as the autonomy of semantics is
tion ... " (Black, 1949: 30).
disregarded, Bloomfield has been charged with mentalism (!) in
53. Most dictionaries vastly exaggerate the incidence of polysemy at
an excellent article concerning the autonomy of grammatical
the expense of vagueness or generality, e. g., in listing separate
and semantic processes (Buyssens, 1950: 37; the same collection
meanings for fair as in fair chance and fair as in fair health. On
contains a most judicious statement of the problem by Larochette).
the fallacy of overspecification in semantic description, see
204
Uriel Weinreich
Benveniste (1954) and Zawadowski (1959); cf. also Zvegincev

(1957: 238-244) and Gove (1957).
54. Since it is essential to operate, not with meaning intuitions, but
with explicit meaning descriptions (Weinreich, 1962), the question
of whether shout and weep do or do not share a semantic compo
nent depends on the verbatim text of their definitions. One could
extract a common factor, such as 'emotional discomposure',
ad hoc; but would it be economical to carry it in the definitions
of each term?
55. But not in all; in rain--, it is obviously coat 2 (P. Ivic). The con
texts, in any case, to be legitimate must not involve any metalin
guistic operation. Cf. also note 76.
56. The best study of this problem seems to be Kotelova (1957); see
also Axmanova (1957: 104-165). Overlapping of grammatical
contexts, i. e., the possibility of ambiguity, as a necessary condi
tion for homonymy is stressed by Vinogradov (1960) and in
Avrorin et al. (1960), the most enlightening discussions of the
subject of homonymy. To the references from the older literature
cited by Ullmann (1951) should be added Richter (1926).
57. With regard to arrange, we find an extended ambiguity: if music!
means 'Musik' and music 2 means 'sheet music, Noten', then
arrange music means either arrange! music 2 or arrange 2 music!, but
as a whole it still remains ambiguous. It has been suggested
(notably by Godel, 1948; Fal'kovic, 1960) that divergent deriva
tional and compounding patterns can function as criteria for
resolving polysemy. But while such divergences may be frequent
concomitants of polysemy, an unbiased sampling of vocabulary
shows that they are not criterial (Kleiner, 1961). In principle,
such concomitances could serve only the lexicographic contempla
tion of the word, in isolation; it cannot help the hearer of living
speech. For how could the ambiguity of j'ai vu des voiles 'I have
seen veils/sails' be resolved, for the given act of speech, by the
fact that the singular of one alternative, 'veils', would be Ie voile,
while the singular of the other would be la voile? (The example is
from UIImann, 1953: 234.)
58. The novel conception of idioms offered by Hockett (1956) is
untenable, for reasons explained elsewhere (Weinreich, 1959).
59. Bar-Hillel (1955: 192) justly defines expressions as idiomatic not
only for a given language but also "with respect to a given ...
dictionary." On the problem of the relation between synonymy
and polysemy, see also Kurylowicz (1955); Saumjan in SPLS
1960: 21-25; and 1. 1. Revzin in SSM 1961: 17-19.
60. Cf. Axmanova (1957: 166-191) on grammatical properties of
certain Russian idioms; Smirnickij (1956: 203-230) on correspond-
205
ing patterns in English; and Ozegov's excellent study (1957) of

the general problem of idiom and cliche formation. Incidentally,
Hockett's specific suggestion (1956) that in English the pattern
A + N is a mark of an idiom is incorrect; actuaIly, A + N is
productively derived from X has a A N (e. g., yellow-belly) or
N is for ... A (e. g., plate for [keeping things] hot --+ hot plate).
61. The example is from DeFrancis (1950: 149), who issues useful
warnings against the exaggeration of Chinese peculiarities.
62. See Axmanova et al. (1961 : 24ff.). It is a mark of the timeliness of
the subject that the problem of semantic universals (semanticeskie
universalii) was independently raised by Axmanova at exactly the
same time as this paper was being written.
63. B. W. and E. G. Aginsky (1948: 170), under universals of concep
tualization, appear to list general semantic fields (including
'maturity', 'space', etc.) and not necessarily universal "Iexemes."
Elaborate thesaurus outlines, such as HaIlig and von Wartburg
(1952) and Voegelin and Voegelin (1957), are also not intended as
anything more than rough tools for eliciting vocabulary or for
classifying elicited items by gross topical domains. For a critical
discussion of the thesaurus approach to lexicography, see Wiister
(1959) and Hiorth (1960).
64. The best discussion of semantic calibration of vocabularies is
Becker (1948).
65. It is assumed that the complex expressions such as noun com
pounds, unless they are idiomatic, can be analyzed as kernel
constructions, or as meaning-preserving transforms of kernel
constructions, and that their meaning can be formulated in terms
of the meanings of the overt constituents and the relations of
linking, nesting, and backgrounding. It is only for the study of
the covert components of semantic simplicia (= grammatical
simplicia or idiomatic complex expressions) that componential
analysis is required. Some scholars, to be sure, are skeptical
about the objectivity of such analyses (e. g., N. F. Pelevina,
SSM 1961: 30-32) and outline instead a program of research
of "semi-idiomatic" complex expressions. This domain of semi
idiomatic, semicovert components, standing in a defective cne
to-one relation to morphemes, is certainly important, too. Such
defective relations appear in idioms and in all cases of sound
symbolism, whether impressionistic (grumble, hiss, sibilant) or
expressionistic (teeny; flit-float- ... ; Yiddish pejsax 'Passover'
kejsax 'Easter').
66. Modern logical semantics, which is still in its infancy, has already
provided us with a way of talking about combinatorial processes,
but it has had little to contribute so far to the componential
206
,I
Uriel Weinreich
analysis of designata. In Reichenbach's book, for example, we

find only casual treatment of 'complex' and 'descriptional'
functions (1948: 122, 311 f.). The roots of our analysis lie in
traditional logical (cf. Ziehen, 1920: 459-599) and psychological
(Bruner et al., 1956) models of the concept. The warning that
concepts cannot be identified with meanings, often voiced in
contemporary Soviet linguistics (e. g., Zvegincev, 1957: 147ff.),
seems to stem mainly from an appreciation of the fact, accom
modated in our theory, that the canonical form of the designatum
is only a limiting case, which fully fits only "terminologized"
vocabulary (Schaff, 1960: 389ff.). The psychological reality of
covert features can be demonstrated by psychophysical experi
ments (see now Luria and Vinogradova, 1959), but such proce
dures are probably unnecessarily circumstantial for the study of
specific vocabularies. Every semantic description of a sign, of
course, constitutes an analyst's hypothesis and is subject to tests
of consistency with the description of other signs, and to validation
by native informants (Wallace and Atkins, 1960: 78f.; Weinreich,
1962).
67. It may be useful to apply the notion of "family likeness" (Wittgen
stein, 1958: 17ff., 43f.) or "polytypic concepts" (Beckner, 1959:
22-25) to semantic analysis. Polysemy of a sign A would then be
defined as [(CI . cz) V (cz . ca) v (ca' c...) ... J. Budagov (1961 : 23f.)
has gone so far as to suggest that in some languages, depending
on the level of their cultural development, the "factorability" of
designata is more thoroughgoing than in others. Unfortunately,
his only evidence for an "underdeveloped" language is Sommer
felt's description of Aranta, which has been much criticized for
lack of anthropological refinement.
68. Just as structural semantic analysis is gathering momentum, some
scholars are looking ahead in cautioning that only a limited,
"terminologized" part of vocabulary lends itself to such analysis;
cf. A. A. Reformatskij (SSM 1961:13) and N. F. Pelevina
(ibid.: 30-32).
69. In an informal experiment, a graduate class of 17 students at
Columbia University was asked to match the eight terms bound,
hop,jump, leap, prance, skip, spring, and vault with their definitions,
taken from a much-used reference dictionary but slightly modified
so as to eliminate illegitimate clues. Out of 136 answers, only 54
(40%) were correct on a first run, and only 89 (65%) were correct
on a second run, when certain additional, contextual clues were
added. Treating dictionary definitions as a proposed description,
these low scores indicate a poor degree of reliability, i. e., a low
degree of terminologization in this lexical set. Surely the group
207
would have done better with eight kinship terms, which constitute
a highly terminologized field. For validation techniques in
semantics, cf. also Naess (1957, with references to his earlier
work) and Tennessen (1959).
70 A formal approach to synonymy in terms of set theory has been
sketched by V. V. Martynov in Pytannja ... (1960: 11-31).
71. On a componential analysis of a set of German "synonymous"
verbs, cf. M. V. Raevskij (SSM 1961: 39-41). On recurrent
components in nonterminologized vocabulary, see also Collinson
(1939).
72. Chinese, for example, has semantic devices of compounding anto
nyms for the formation of superordinates: lai 'come' + wang
'go' = lai-wang 'traffic'; shu 'lose' + ying 'win' = shu-ying
'result of the game'; zao 'early' + wan 'late' = zao-wan 'interval,
time'; xu 'false' + shi 'true' = xu-shi 'state of affairs', etc.
(Sofronow, 1960: 81ff.). The device certainly recurs in other
languages, though it is perhaps not so fully utilized; cf. Yiddish
tatp 'father' + mamp 'mother' = tatp-mamp 'parents'; gopl
'fork' + left 'spoon' = gopl-Mft 'cutlery'.
73. Various investigations now in progress [cf. Postscript] are intended
to throw light on the possibilities of componential description
of vocabulary. Of particular interest seems to be the problem of
whether objective distributional methods will yield results equiv
alent to intuitive-componential notions of meaning. Among the
relevant projects is the work of the Cambridge Language Research
Unit (Masterman, 1959); the work on a semantic calculus of
kinship terminology (Wallace and Atkins, 1960; Wallace, 196Ia);
and the experimental work on semantic analysis in the Machine
Translation Laboratory of Moscow's First State Pedagogical
Institute of Foreign Languages (1. R. Gal'perin et 01., SSM 1961:
5-8; A. K. Zolkovskij et al., SSM 1961: 6Of.; V. V. Ivanov,
SLTM 1961: 18-26). On universal "semes," cf. also A. B. 001
gopol'skij (SPLS 1960: 35-42). Concerning the universality of
affective meaning components, see Maclay and Ware (1961) and
Osgood (1961).
74. On the vocabulary of "abstract" superordinates in some languages
of primitive societies, cf. Moszyflski (1956).
75. The study of denotation or reference is, of course, an entirely
legitimate pursuit, both for the purpose of a general theory of
communication (e. g., Shwayder, 1961) and for the referential
"orientation" of certain primitive terms in a description of a
particular language as a semantic system (Laxuti et 01.,1959: 219).
In the CToss-cultural study of color naming, there has been a good
deal of progress recently; cf. esp. Ervin (1961) and Semjakin (1960).
Uriel Weinreich
208
76. Attempts to explicate ordinary language without recourse to

intensions are bound to lead to oversimplified conceptions. For
example, Goodman (1949) wishes to defend a purely extensional
theory of synonymy. This at first seems to lead to a difficulty;
for words like centaur and unicorn, the truth-value of a sentence
containing one of them is never changed by replacing it by the
other; hence, they would be synonymous. In order to save their
nonsynonymy, Goodman points out that there are contexts in
which they cannot be interchanged salva veritate, e. g., picture
0/ a _~. But it is necessary to point out that the differentiating
context is of a very special type, for these sentences are of a kind
called quasi-syntactical by Carnap (1937: 74). The words centaur
and unicorn would remain interchangeable in the contexts tail
0/ a .._ _, teeth 0/ a ..._ _, stomach of a ___, but not in
picture 0/ a ___ , poem about a ___, illusion 0/ a
etc.
The latter group of contexts involves metalinguistic operations
and may be said to demonstrate the reality of the very intensions
which the author sought to eliminate.
References
Aginsky, B. W., and E. G. (1948). "The Importance of Language
Universals," Word 4.168-172.
Apresjan, Ju. D. (1962). [Distributional Analysis of Meaning and
Structural Semantic Fields], Leksikograjiceskij sbornik 5. 52-72.
Moscow.
Arndt, W. (1960). "'Modal Particles' in Russian and German,"
Word 16. 323-336.
Avrorin, V. A. et al. (1960). {Discussion on Problems of Homonymy
...J, Leksikograjiceskij sbornik 4.35-102.
Axmanova, O. S. (1957). Ocerki po obscej i russkoj leksikologii.
Moscow.
_ _, I. A. Mel'cuk, E. V. Paduceva, and R. M. Frumkina (1961).
a tocnyx metodax issledovanija jazyka. Moscow.
Bar-Hillel, Y. (1954). "Indexical Expressions," Mind n. s. 63.359-379.
_ _ (1955). "Idioms," in W. N. Locke and A. D. Booth (eds.),
Machine Translation 0/ Languages, 183-193. New York and
London.
Bazell, C. E. (1954). "The Sememe," Litera 1. 17-31.
ON T HE SEMANTIC STRUCTURE OF LANGUAGE
209
Becker, H. (1948). Der Sprachbund. Berlin and Leipzig.
Beckner, M. (1959). The Biological Way o/Thought. New York.
Bendix, E. H. (1961). "Componential Analysis of an English Semantic
Field," unpubl. seminar paper, Columbia University.

- - - (1965). Componential Analysis 0/ General Vocabulary: the
Semantic Structure 0/ a Set 0/ Verbs in English, Hindi, and Japanese.
Ph. D. dissertation, Columbia University, to appear as a supple
ment to IJAL, April, 1966.
Benveniste, E. (1954). "Problemes semantiques de reconstruction,"
Word 10. 251-264.
Black, M. (1949). Language and Philosophy. Ithaca.
Bloomfield, L. (1933). Language. New York.
Breal, M. (1897). Essai de semantique. Paris. (Page ref. to English ed.,
1900.)
Brown, J. C. (1960). "Loglan," Scientific American 202 (June), 53-63.
Brown, R. W. (1957). "Linguistic Determinism and the Part of Speech,"
J. 0/ Abnormal and Soc. Psycho!. 55. 1-5.
Bruner, J. S., J. J. Goodnow, and G. A. Austin (1956). A Study 0/
Thinking. New York.
Budagov, R. A. (1958). Vvedenie v nauku 0 jazyke. Moscow.
- - - (1961). [Toward a Critique of Relativistic Theories of the
Word], Voprosy teor;; jazyka v sovremennoj zarubeZnoj lingvistike,
5-29, Moscow.
Biihler, K. (1934). Sprachtheorie. Jena.
Buyssens, E. (1950). "Conception fonctionnelle des faits linguistiques,"
Grammaire et psychologle, 35-51. Paris.
Carnap, R. (1937). The Logical Syntax 0/ Language. London.
(1942). Introduction to Semantics. Cambridge, Mass.
- - (1947). Meaning and Necessity. Chicago. (Ref. to 2nd ed., 1956.)
- - - , and Y. Bar-Hillel (1953). "Semantic Information," Brit.
J. Phi/os. 0/ Science 4. 147-157.
Casagrande, J. B. (1954). "Comanche Linguistic Acculturation,"
IJAL 20. 140-151,217-237; 21. 8-25 (1955).
Cassirer, E. (1923). Philosophie der symbollschen Formen, I. Berlin.
(Ref. to Eng. ed., New Haven, 1953.)
Chao, Y. R. (1956). "Chinese Terms of Address," Lg. 32. 217-241.
- - - (1959). "How Chinese Logic Operates," Anthrop. Linguistics
1. 1-8.
210
Uriel Weinreich
Chomsky, N. (1955). "Semantic Considerations in Grammar,"

Georgetown Univ. Monographs 8. 141-154.
- - (1957). Syntactic Structures. The Hague.
Collinson, W. E. (1937). Indication (Lg. Monographs 17).
- - (1939). "Comparative Synonymics," Trans. Philolog. Soc.,
54-77.
Conklin, H. C. (1955)... Hanun60 Color Categories," S WJAIl. 339
344.
- - (1962). "Lexicographical Treatinent of Folk Taxonomies," in
F. W. Householder, Jr., and Sol Saporta (eds.), Problems in
Lexicography (IUPAL, no. 21), 119-141.
DeFrands, J. (1950). Nationalism and Language Reform in China.
Princeton.
Ervin, S. M. (1961). "Semantic Shift in Bilingualism," American J. of

Psychology 74. 233-241.
- - , and G. Foster (1961). "The Development of Meaning in
Children's Descriptive Terms," J. of Abnormal and Soc. Psychol.
61. 271-275.
Fal'kovie, M. M. (1960). [On the Problem of Homonymy and Polys
emy], Vopr. jaz., no. 5, 85-88.
Fevrier, J. G. (1948). Histoire de l'ecriture. Paris.
;1
Flavell, J. H., and D. J. Stedman (1961). "A Developmental Study of

Judgments of Semantic Similarity," J. of Genetic Psychology
98. 279-293.'
Freeland, L. S. (1951). Language of the Sierra Miwok (IUPAL
Memoir 6).
French, D. (1958). "Cultural Matrices of Chinookan Non-Casual
Language," lJAL 24. 258-263.
Funke, O. (1932). Innere Sprachform. ReiChenberg.
Gleason, H. A., Jr. (1955). An Introduction to Descriptive Linguistics.
New York.
- - (I962). "The Relation of Lexicon to Grammar," in F. W.
Householder and Sol Saporta (eds.), Problems in Lexicography
(IUPAL, no. 21).
Godel, R. (1948). "Homonymie et identite," Cabiers F. de Saussure

7. 5-15.
Goodenough, W. (1956). "Componential Analysis and the Study of
Meaning," Lg. 32. 195-216.
ON T HE SEMANTIC STRUCTURE OF LANGUAGE
211
Goodman, N. (1949). "On L\keness of Meaning," Analysis 10. 1-7.

Gove, P. B. (1957). "Problems in Defining," in J. H. Shera et al. (eds.),
Information Systems in Documentation, 3-14. New York and
London.
Hallig, R., and W. von Wartburg (1952). Begriffssystem als Grundlage
fur die Lexikographie. Berlin.
Harris, Z. S. (1957). "Co-occurrence and Transformation in Lin
guistic Structure," g. 33. 283-340.
Haugen, E. (1957). "The Semantics of Icelandic Orientation," Word
13.447-459.
Hiorth, F. (1960). "Zur Ordnung des Wortschatzes," Studia Linguistica
14.65-84.
Hjelmslev, L. (1953). Prolegomena to a Theory of Language (IUPAL
Memoir 7).
Hockett, C. F. (1948). "Potawatomi IV," lJAL 14.213-225.
- , - (1954). "Two Models of Grammatical Description," Word
10. 210-233.
- - - (1956). "Idiom Formation," For Roman Jakobson, 222-229.
The Hague.
- - - (1958). A Course in Modern Linguistics. New York.
- - - (1960). "The Origin of Speech," Scientific American, Sept.
Hymes, D. H. (1960). "Lexicostatistics So Far," Current Anthropology
1.3-44.
- - (1961). "On Typology of Cognitive Styles in Language,"
Anthrop. Linguistics 3, no. 1,22-54.
Jakobson, R. (1957). Shifters, Verbal Categories, and the Russian
Verb. Harvard University.
- - - (1959). "Boas' View of Grammatical Meaning," American
Anthropologist 61, part 2, 139-145.
Joos, M. (1950). "Description of Language Design," repro from JASA
in Readings in Linguistics (1957), 349-356. Washington.
- - (1958). "Semology: a Linguistic Theory of Meaning." SIL
13. 53~70.
Karolak, S. (1960). [Some Notes on the Structure of the Semantic
Spectrum], Lingua posnaniensis 8. 243-253.
Katz, J. J., and J. A. Fodor (1963). "The Structure of a Semantic
Theory," Language 39. 170-210.
- - - and P. M. Postal (1964). An Integrated Theory of Linguistic
Descriptions. Cambridge, Mass.
Uriel Weinreich
212
Kleiner, E. (1961). "The Discrimination of Multiple Meaning in

English," unpubl. seminar paper, Columbia University.
Kotelova, N. Z. (1957). [Indications of Syntactic Relations ... as a
Means of Discriminating Semantic Distinctions], Leksikograjiceskij
sbornik l. 98-120.
Kurylowicz, J. (1955). [Notes on Word Meanings], Vopr. jaz., no. 3,
73-81.
213
Morris, C. W. (1938). Foundations of a Theory of Signs. Chicago.
- - - (1946). Signs, Language, and Behavior. New York.
Most, M. (1949). Comments in Actes du 6eme Congres international
des Linguistes (1948), 183-190. Paris.

Moszynski, K. (1956). [The Vocabulary of So-Called Primitive
Peoples1, Biuletyn Polskiego Towarzystwa
J~zykoznawczego
15.
Kuznecov, A. V., E. V. Paduceva, and N. M. Ermolaeva (1961).

[On an Informational Language for Geometry and an Algorithm
for Translating from Russian to the Informational Language],
Masinnyj perevod i prikladnaja lingvistika 5.3-21.
Larochette, J. (1950). "Les Deux oppositions verbo-nominales,"
Grammalre et psychologle, 107-118. Paris.
Laxuti, D. G., L I. Revzin, and V. K. Finn (1959). [On a Certain
Approach to Semantics], S.S.S.R., Ministerstvo vysSego obra
zovanija, Naucnye doklady vysJej 'Skoly; jilosofskie nauki, no. 1,
91-112.
Naess, A. (1957). "Synonymity as Revealed by Intuition," Philo
sophical Review, 66. 87-93.
Nida, E. A. (1951). "A System for the Description of Sema,ntic
Elements," Word 7. 1-14.
- - (1958). "Analysis of Meaning and Dictionary Making,"
/JAL 24. 279-292.
Osgood, C. E. (1961). "Studies on the Generality of Affective Meaning
Systems." Urbana, lll.: Institute of Communications Research,
207-219.
Lees, R. B. (1960). The Grammar of English Nominalizations. (/JAL
26, no. 2, part 2).
Leisi, E. (1953). Der Wortinhalt; seine Struktur 1m Deutschen und
mimeographed.
Ougov, S. L (1957). [On the Structure of Phraseology], Leksiko
grajiceskij sbornik 2.31-57.
Paul, H. (1880). Prinzipien der Sprachgeschichte. Halle. (Ref. to 5th ed.,
Englischen. Heidelberg.
Lounsbury, F. G. (1956). "A Semantic Analysis of Pawnee Kinship
Usage," Lg. 32. 158-194.
Luria, A. R., and O. S. Vinogradova (1959). "An Objective Investiga
tion of the Dynamics of Semantic Systems," Brit. J. Psychology
50.89-105.
Maclay, H., and E. E. Ware (1961). "Cross-Cultural Use of the

Semantic Differential," Behavioral Science 6. 185-190.
Marcus, H. (1960). Die Fundamente der Wirklichkeit als Regulatoren
der Sprache. Bonn.
Martinet, A. (1960). Elements de linguistique generale. Paris.
Maspero, H. (1933). "La Langue chinoise," Conferences de I'Institut de
Llnguistique de I' Universite de Paris, 33-70.
Masterman, M., ed. (1959). Essays on and in Machine T,anslation.
Cambridge.
McQuown, N. A. (1956). "Analysis of the Cultural Content of Lan
guage Materials," Language in Culture, 20-31. Chicago.
Mel'cuk, LA. (1960). [On the Terms 'Stability' and 'ldiomaticity'],
Vopr. jaz., no. 4, 73-80.
1920.)
Peirce, C. S. (1932). Collected Papers, II. Cambridge, Mass.

Pytannja ... (1960). Pytannja prikladnoji lingvistyky; tezisy dopovidej
miivuzovs'koji naukovoji konferenciji 22-28.9.1960. Cernivcy.
Quine, W. V. (1953). From a Logical Point of View. Cambridge, Mass.
- - (1960). Word and Object. Cambridge, Mass., New York.
Ravid, W. (1961). "The Grammatical Behavior of the Adjective 'Fair',"
unpubl. seminar paper, Columbia University.
Regnell, H. (1958). Semantik. Stockholm.
Reichenbach, H. (1948). Elements of Symbolic Logic. New York.
Reichling, A. (1935). Het Woord. Nijmegen.
Richter, E. (1926). "Ueber Homonymie," FS Paul Kretschmer, 167
201. Vienna.
Rosenbloom, P. (1950). The Elements of Mathematical Logic. New
York.
Sapir, E. (1915). "Abnormal Types of Speech in Nootka," reprinted
from Canada, Geological Survey, Memoir 62, in Selected Writings
ofEdward Sapir, ed. D. G. Mandelbaum, 1949.
- - (l921). Language. New York.
214
'I
I
I'i
Uriel Weinreich
Saussure. F. de (1922). Cours de linguistique generale. Paris and

Geneva, 2nd ed.
Savcenko, A. N. (1959). Casti reli i kategorii myslenija. Rostov-on-Don.
Schaff, A. (1960). Wst"p do semantyki. Warsaw.
Schmidt, F. (1959). Logik der Syntax. Berlin, 2nd ed.
Sechehaye, A. (1926). Essa; sur la structure logique de la phrase. Paris.
Seiler, H. (1958). Comments in Proceedings of the 8th International
Congress of Linguists, 692-695. Oslo.
Semjakin, EN., ed. (1960). Myslenie i ree: trudy Instituta psixologii
Izvestija Akademii pedagogiceskix nauk R. S. F. S. R., 113.
Moscow. [[ncludes six papers on color terminology in lan
guages of the Soviet Arctic and among preschool Russian children.]
~endel's, E.1. (1959). [On Grammatical Meanings on the Plane of
Content], in T. A. Degtereva, ed., Principy nauenogo analiza
.
jazyka, 45-63. Moscow.
Shwayder, D. S. (1961). Modes of Referring and the Problem of
Universals. Berkeley and Los Angeles.
SLTM (1961). Tezisy dokladov konferencii po strukturnoj Iingvistiki,
posvjaicennoj voprosam transformaclonnoj grammatiki. Moscow.
Smart, J. J. C. (1949). "Whitehead and Russell's Theory of Types,"
Analysis 10. 93-95.
Smelev, D. N. (1960). [On "Bound" Syntactic Constructions in
Russian], Vopr. jaz., no. 5, 41-60.
Smirnickij, A. L (1956). Leksikologija anglijskogo jazyka. Moscow.
Sofronow, M. W. (1960). "Die Methoden der Wortbildung in der
Sprache des Romans Shuihuzhuan,''' in P. Ratchnevsky, ed.,
Beitriige zum Problem des Wortes im Chinesischen, 11-94. Berlin.
S~rensen, H. S. (1958). Word-Classes in Modern English With Special
Reference to Proper Name~, With an Introductory Theory of
Grammar, Meaning, and Reference. Copenhagen.
Spang-Thomsen, B. (1956). Review of Strtlctureel analyse van vislteei
taalgebruik binnen een groep dove kinderen, by B. T. M. Tervoort,
Word 12.459-461.
SPLS (1960). Tezisy dokladov na 6-om plenarnom zasedanii [Slovarnoj]
komissii [0. L. Ja. A. N. S. S. S. R.] posvjaseennom sovremennoj
problematike leksikologii i semasiologii (19-21 oktjabrja 1960 g.).
Moscow.
SSM (1961). Tezisy dokladoll melvuzovskoj konferencii po prime
neniju strukturnyx i statisticeskix metodov issledo'lanija slovarnogo
sostavajazyka (21-25 nojabrja 1961 g.) Moscow.
215
Stankiewicz, E. (1954). "Expressive Derivation of Substantives in

Contemporary Russian and Polish," Word 10. 451-468.
Strawson, P. F. (1952). Introduction to Logical Theory. London and
New York.
Tennessen, H. (1959). Inquiry 2. 265-290.
Ullmann, S. (1951). Principles of Semantics. Glasgow. (2nd ed., /951.)
- - (1952). Precis de semantique fran~aise. Berne.
_ _ (1953). "Descriptive Semantics and Linguistic Typology,"
Word 9.225-240.
Voegelin, C. E, and F. R. Voegelin (1957). Hopi Domains (lUPAL
Memoir 14).
_ _' (1961). "Typological Classification of ... Alphabets," Anthrop.
Linguistics 3, no. I, 55-96.
Vinogradov, V. V. (1960). [On Homonymy and Cognate Phenomena],
Vopr. jaz., no. 5, 1-17.
Wallace, A. F. C. (1961a). "The Psychic Unity of Human Groups,"
in Bert Kaplan, ed., Studying Personality Cross-Culturally, 129
164. Evanston, IlL, and Elmsford, N. Y.
_ _ (l961b). "On Being Just Complicated Enough," Proceedings
of the National Academy of Sciences, 41. 458-464.
_ _, and J. Atkins (1960). "The Meaning of Kinship Terms,"
American Anthropologist 62. 58-80.
Wegener, P. (1885). Untersuchungen uber die Grundfragen des Sprach
lebens. Halle a. S.
Wells, R. (1954). "Meaning and Use," Word 10.115-130.
Weinreich, U. (1955). Review of Ullmann (1952), g. 31. 537-543.
_ _ (1958). "Travels in Semantic Space," Word 14.346-366.
- - (1959). Review of Hockett (1958), Romance Philology
320-341.
_ _ (1962). "Lexicographic Definition in Descriptive Semantics,"
in F. W. Householder, Jr., and Sol Saporta, eds., Problems of
Lexicography (lUPAL, no. 21), 25-43.
_ _ (1963). "[Soviet] Lexicology," in T. A. Sebeok, ed., Current
Trends in Linguistics, I, 60-93. The Hague.
_ _ (1964). "Webster's Third: A Critique of Its Semantics,"
lJAL 30.405-409. Russian version in Vopr. jaz. (1965), no. I,
128-132.
_ _ (1965). "Explorations in Semantic Theory," in T. A. Sebeok,
(ed.), Current Trends in Linguistics, III. The Hague, in press.
216
Uriel Weinreich
Whorf, B. L. (1956), Language, Thought, and Reality, Cambridge,

Mass., and New York.
Wittgenstein, L. (1958). Preliminary Studies ... the Blue and Brown
Books. Oxford.
Wi.ister, E. (1959). "Die Struktur der sprachlichen Begriffswelt und
ihre Darstellung in WorterbUchern," Studiumgenerale 12. 615-627.
Zawadowski, L. (1958). "La Signification des morphemes polysemes,"
CHAPTER 8
SEMANTIC UNIVERSALS
Biuletyn Polskiego Towarzystwa Juykoznawczego 17.67-95.

STEPHEN ULLMANN
(1959). "La Polysemie pretendue," ibid. 18. 11-49.
University of Leeds
Ziehen, T. (1920). Lehrbuch der Logik. Bonn.
Ziff, P. (1960). Semantic Analysis. Ithaca.
Zimmer, Karl E. (1964). Affixal Negation in English and Other Lan
guages: an Investigation of Restricted Productivity. (Supplement to

Word 20, no. 2.)
1. Introduction
Zvegincev, V. A. (1957). Semasiologija. Moscow.
The quest for universals has played a vital part in the develop
ment of semantic studies. The pioneers of modern semantics in
the last century saw in the discovery of general "laws" one of
the main objectives of the new science. As far back as the 1820's,
the German classicist C. C. Reisig had set up "semasiology"
as an independent division of linguistics, and had suggested that
it should investigate "the conditions governing the development
of meaning." 1 Half a century later, in 1883, Michel Breal was
even more categorical. In an article which introduced the term
semantics into linguistic terminology. he mentioned among the
tasks of the new discipline the study of the "laws which preside
over the transformation of meanings." 2 [n his Essai de seman
tique, which appeared fourteen years later, Breal showed how
this aim could be achieved, ana his example was followed by
other linguists who put forward a number of "laws" underlying
various types of semantic change. Among some dissenting voices
was Saussure, who warned that changes of meaning were often
due to unique causes and were no more than isolated accidents
in the history of language. 3 Yet the quest continued unabated.
It was accepted as axiomatic that, as one linguist, Jespersen,
put it, "there are universal laws of thought which are reflected
in the laws of change of meaning ... even if the Science of
217
218
Stephen Ullmann
Meaning ... has not yet made much advance towards discovering
them." 4 Even today there are some scholars who hold similar
views. Only a few years ago a leading Russian linguist criticized
contemporary semantics for having turned away from its prin
cipal task: the study of specific laws of linguistic development. fj
Since the early 1930's, however, there has been a significant shift
of emphasis in semantics, as in other branches of linguistics:
descriptive and structural problems have come to the forefront
of research, and the traditional study of changes of meaning,
though by no means abandoned, has been relegated to the
background. This shift of emphasis has had an important effect
on the search for semantic universals. There has been compara
tively little work of late on the orthodox type of semantic "law."
Instead, attention has been focused on synchronic features of
general validity, and also on the principles which determine the
structure of the vocabulary.
If one surveys the various semantic "laws" and other universals
which have been either implicity assumed or explicitly formulated
in the past, one finds that they have one thing in common:
nearly all of them are based on insufficient evidence. Only too
often have far-reaching conclusions been drawn from inadequate
data collected from a limited number of languages. The alleged
universals obtained in this way are in many cases quite plausible,
but plausibility is no proof unless the proposition is so self
evident that it becomes truistic and triviaL Besides, by the very
nature of things, most semantic universals are no more than
statistical probabilities, and the likeli~ood of their occurring in
a given language could be determined only if we possessed far
more extensive and representative data than we have at present.
What Leonard Bloomfield wrote about general grammar is
entirely applicable to semantics and deserves to be quoted in
full:
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 .... The fact that some features are, at any rate, wide
SEMANTIC UNIVERSALS
219
spread is worthy of notice and calls for an explanation; when

we have adequate data about many languages, we 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. II
Since it is one of the aims of this conference to provide precisely
such a wide factual basis for the setting up of universals, I shall
try in this paper to indicate certain semantic features and pro
cesses which might repay investigation on an interiinguistic
scale. First, however, it will be necessary to define more closely
the two terms "semantic" and "universal." Throughout this
paper, "semantic" will be used solely with reference to word
meaning. It has been customary since Aristotle to regard the
word as the smallest meaningful unit of speech. 7 We now know
that this is not so. The "smallest meaningful element in8 the
utterances of a language" is the morpheme, not the word. The
word itself is defined, in Bloomfield's classic formula, as a
"minimum free form" 9 which may consist of one or more mor
phemes. It follows that semantic problems will arise not only
at the word level but also below and above it: below it at the
level of bound morphemes (suffixes, prefixes, nonindependent
roots, etc.); and above it at the level of phrases and the higher
combinations into which they enter. No problems of meaning
below or above the word level will be considered in this paper;
nor shall I deal with the semantics of so-called "form-words"
pronouns, articles, conjunctions, prepositions, etc.-which,
though they behave like words in some respects, have a purely
grammatical function and do not therefore belong to the lexical
system of a language. 10
It should also be noted that the word itself is not a linguistic
universal in the absolute sense. In so-called "polysynthetic"
languages, where a whole series of bound forms is combined
into a single term, the word will obviously have a structure and
status entirely different from, say, English or Chinese, and
many of the tendencies discussed here are therefore inapplicable
to such languages.
220
Stephen Ullmann
With regard to the meaning and implications of the term

"universal," some flexibility will be needed when applying it to
semantic phenomena, which are often fluid, imprecise, and
subjective. From the point of view of their validity, the features
and processes discussed in this paper will fall into several broad
categories:
I. Some of them may turn out to be "unrestricted universals,"
or "universal implications," according to the terminology of the
Memorandum. Even these would be "unrestricted" only in the
sense that they occurred in all the languages examined in a
large-scale research program. We could, of course, never prove
conclusively that they are omnipresent--or "panchronic," as
Saussure would say ll-that they exist in every language at any
stage in its development.
2. Most semantic universals are likely to be of the statistical
variety: they will not be necessarily present in any given lan
guage, but one may to some extent predict the probability of
their occurrence. It should be added that certain semantic phe
nomena are not precise enough to be amenable to rigorously
statistical analysis, so that no more than a rough estimate of
probabilities can be expected.
3. There is yet another type which has some affinities with
universals but is far more limited in scope: parallel develop
ments which occur in a number of different languages but are
unknown elsewhere. Many types of metaphor and other forms
of semantic change fall within this category: they are too wide
spread to be due to mere chance, but not widespread enough to
be statistically significant. It is, of course, always possible that
such a tendency will turn out to be a statistical universal, and
will thus pass into the previous category, if the scope of the
inquiry is sufficiently widened.
4. In addition to these tendencies, attention will also have to
be paid to typological criteria, since, as the Memorandum
rightly points out, these have obvious connections with the prob
lem of universals. Very little work has been done so far on semantic
SEMANTIC UNIVERSALS
221
typology; 12 nevertheless, one or two criteria have already been
identified, and these will be discussed in the appropriate sections.
In semantics, as in other branches of linguistics, we may expect

to find two kinds of universals: synchronic features and diachronic
processes,13 though in practice it may not always be easy to
separate the two. It will also be expedient to distinguish a third
class of semantic universals: those which transcend individual
words and are bound up with the general structure of the
vocabulary.
2. Universal Features in Descriptive Semantics

2.1 Transparent and opaque words
The relation between sound and sense is a perennial moot

point in the philosophy of language. The Greeks were already
divided into two camps on this issue: the "naturalists," who argued
that words have their meaning "by nature" (physei), by virtue
of an intrinsic correspondence batween form and sense, and the
"conventionalists," who claimed that meaning is arbitrary, based
on a social convention (thesei). Saussure regarded "arbitrariness"
as one of the fundamental principles of language,14 whereas
linguists with a different temperament inclined toward the
naturalist view and emphasized the importance of onomato
poeia in the structure of our words. The old controversy flared
up again some twenty years ago, and during the ensuing debate,
the various aspects of the problem were usefully clarified. 15 It
seems clear, first of all, that no language is either completely
transparent or completely opaque. AU of them are likely to
contain both conventional and "motivated" terms in varying
proportions which will depend on a number of factors, some
linguistic, others cultural and social. The existence of two types
of words is in all probability a semantic universal. It is hard to
imagine a language which would have no onomatopoeic terms
and no transparent metaphors, and equally hard to conceive of
one which would consist solely of motivated words. This assump
i
J
I
,I
1
222
Stephen Ullmann
tion would, of course, have to be empirically tested, together

with other, more specific problems raised by motivation.
2. I. I. THREE TYPES OF MOTIVATION. In English and many
other languages, words can be motivated in three different ways.
The verbs swish, sizzle, and boom are phonetically motivated
because the sounds are a direct imitation of the sense. A com
pound like arm-chair and a derivative like thinker or retell are
morphologically motivated: whoever knows their components
will understand them at once. Finally, figurative expressions like
"the bonnet of a car" or "the pivot on which a question turns"
are semantically motivated: they are derived, by transparent
metaphor, from bonnet "head-dress" and pivot "short shaft or
pin on which something turns or oscillates." It should be noted
that morphological motivation is "relative" in the sense that,
while the words themselves are motivated, their ultimate con
stituents may be opaque, as they are in the preceding examples
arm, chair, think, tell, and the bound morphemes -er and re_. 16
The same may be said of semantic motivation: the metaphorical
usesof bonnet and pivot are transparent, but the words themselves,
in their literal meanings, are purely conventional.
Can these three types of motivation be regarded as semantic
universals? The first and the third types are likely to occur in
all languages; the morphological variety, however, will be more
restricted in scope since it will depend on the phonological and
morphological structure of each idiom. [n a language made up
entirely of monomorphemic words, there will be little room for
such motivation. On the other hand, the existence of infixes in
some languages will provide a form of motivation which English
does not possess.
2. I. 2. RELATIVE FREQUENCIES. The proportion of opaque and
transparent terms, and the relative frequency of the various
forms of motivation, may provide valuable criteria for linguistic
typology. This was adumbrated by Saussure when he distinguished
between two kinds of languages: the "Iexicological" type, where
conventionality is prevalent, and the "grammatical" type, which
SEMANTIC UNIVERSALS
223
prefers motivated words. From the examples he gave it is clear
that he was thinking primarily in terms of morphological motiva

tion. In his view, Chinese represents the extreme form of conven
tionality whereas Proto-Indo-European and Sanskrit tend
~i
Ii
toward the opposite pole; English is far less transparent than
German, while French, compared with Latin, shows a very con
siderable increase of the opaque element. 17
A glance at word-structure in English, French, and German
fully confirms Saussure's classification. There are numerous
cases where English and French have an opaque, unanalyzable
term corresponding to a transparent compound in German:
$.kate - patin - Schlittschuh; chive - cive - Schnittlauch;
glove - gant - Handschuh, etc. Elsewhere the same idea is
expressed in German by a compound, and in English and French
by a learned classical formation: hippopotamus - hippopotame
Nilpferd; phonetics - phonhique - Lautlehre; hydrogen
hydrogene - Wasserstoff, etc. German can also form derivatives
more freely than the other two languages. From the noun Stadt
it can derive the adjective stiidtisch, whereas English and French
have hybrid pairs: town - urban, ville - urbain. Similarly
bishop - episcopal, eveque - episcopal, Bischof - bischoflich;
language - linguistic, langue - linguistique, Sprache - sprach
lich, etc. It might be possible to devise some statistical test for
these relative frequencies. Such a test might be based on samples
from dictionaries, on a representative selection of texts, or on
,
both. Such isolated numerical data as are already available
:,
seem to be very suggestive. In Old English, for example, which
\~
was a more transparent language than Modern English, nearly
'j
fifty terms derived from heofon "heaven" have been counted,
including such picturesque formations as heofon-candel "sun,
J
moon, stars," and heofon-weard "Heaven's keeper, God." 18
,1J
Pending the collection of reliable statistics, the ease with which
.,.,]',
examples can be almost indefinitely multiplied is symptomatic
'!~
of the preferences of various languages. Naturally such preferences
are merely statistical, and an odd instance can always be found
where they do not work, as in the opaque German Enkel opposed
~
to the transparent English grandson and French petit-fils. 19
>~.
"~'I
.
",
J
'I
I'"
,~
,l
i1
,:.,:;
,.\.1
224
Stephen Ullmann
The distinction between these two types of word-structure has

far-reaching implications which can only be briefly mentioned
here. In the teaching and learning of languages it is obviously
of great importance whether a given vocabulary is relatively
transparent, with many self-explanatory terms and closely knit
associative systems, or whether a large proportion of the words
is opaque, containing no clue to their meaning. Within a speech
community, the presence of many learned classical terms may
create a "language bar" between people with different educational
backgrounds. 20 In forming new words, an idiom where com
pounds and derivatives are easily coined can rely extensively on
native resources, as Fichte already emphasized in his Speeches
to the German Nation: it can also provide ammunition for
purism and linguistic chauvinism. Morphological motivation
may even tempt philosophers to indulge in gratuitous etymolo
gizing in the vain hope that they may thus discover the "proper"
meaning of a word; some of Martin Heidegger's verbal acrobatics
can be traced back in no small measure to this factor. 21
The other types of motivation are less suitable for frequency
counts since they are more fluid and subjective than morpho
logical structure. It is commonly believed, for example, that
German is richer in onomatopoeic formations than French, but
it is hard to think of an objective test which might confirm or
disprove this impression.
It has also been suggested that there is a kind of equilibrium
between morphological and semantic motivation. Some lan
guages, it is claimed, will tend to fill gaps in vocabulary by
forming new words, whereas others add new meanings to existing
terms. 22 There may be a grain of truth in this suggestion, but
other factors are also involved in the process. If Modern English
and Modern French are far less transparent than their older
forms, this is due primarily to the introduction of countless foreign
words: French and Graeco-Latin into English, mainly Graeco
Latin into French. It would be difficult to prove that semantic
motivation, by metaphor or other means, has greatly benefited
by the decline of composition and derivation in these languages.
SEMANTIC UNIVERSALS
225
2. 1. 3. PATTERNS OF SOUND-SYMBOLISM. It is common knowl

edge that onomatopoeic terms, however conventionalized, often
show striking similarities in different idioms; they bear witness,
in Schuchardt's famous formula, not to historical connections
but to "elementary kinship." Here, then, is a strong prima facie
case for interlinguistic inquiries looking for universals. Since
a great deal has been written on the subject, it might be desirable
to start with an extensive inventory of what is already known
sorting out scientifically established facts from dilettantish
speculations which at times brought the whole matter into dis
repute. It will also be necessary to distinguish between "primary"
and "secondary" onomatopoeia. Of the two, the primary type,
the imitation of sound by sound, is far simpler and less con
troversial than the secondary type, where nonacoustic ex
periences-movement, size,. emotive overtones, etc.-are repre
sented by the sounds. It is not surprising that in many cases,
though by no means in all, the same noise should be perceived
and transcribed in much the same way in different languages.
The example of the "cuckoo" has often been quoted, and it is no
doubt significant that the bird should have closely similar and
distinctly onomatopoeic names not only in many Indo-European
languages (English cuckoo, French coucou, Spanish cue/iIIo,
Italian cuculo, Rumanian cucu, German Kuckuck, Greek k6kkyx,
Russian kukushka, etc.), but even in some Finno-Ugrian idioms
(Hungarian kakuk, Finnish kiiki, Zyrian kok).23 Similarly, it is
only natural that verbs for "snoring" should in many languages
contain an [rJ (English snore, German schnarchen, Dutch snorken,
Latin stertere, French ronjler, Spanish roncar, Russian chrapet',
Hungarian horkolni, etc.), and those for "whispering" an [sJ, UJ,
or [tIl (English whisper, German wispern andjliistern, Norwegian
hviske, Latin susurrare, French chuchoter, Spanish cuchichear,
Russian sheptdt', Hungarian sugni, susogni, suttogni, etc.). Such
correspondences are certainly interesting and worth studying
on a broader basis, though they are rather too obvious to throw
much light on the fundamental structure of language.
More significant and more delicate are problems raised by
secondary onomatopoeia. In this type the connection between
'.'
226
Stephen Ullmann
SEMANTIC UNIVERSALS
227
sound and sense is less evident than in the previous one; yet
The device is very old; it is already used in the Odyssey (Book I,
even here there exist extensive similarities between various
II.
56--57):
languages. A celebrated example is the "symbolic value" of
the vowel [i] as an expression of smallness. 24 This is found
aiel de ma/akoisi kal haimy/ioisi 16goisi the/geL
in a number of languages: English little, slim, thin, wee, teeny
(and ever with soft and wheedling words she beguiles him.)
weeny; French petit; Italian piccolo; Rumanian mic; Latin
It is interesting to find a very similar use of laterals in Finnish
minor, minimus; Greek mikros; Hungarian kis, kicsi, pici, etc.
and Hungarian poetry; 26 for example, in Sydiimeni Lau/u (My
To such adjectives may be added many nouns denoting small
Heart's Song), by Aleksis Kivi:
creatures or things, such as English kid, chit, imp, slip, midge,
Sie//' on lapsen lysti olla,
tit, bit, chip, chink,jijJy, pin, pip, tip, whit, and also such diminutive
Wan tullen tuuditella.
suffixes as the English -ie, -kin, and -ling. 25 By scrutinizing a
(It is pleasant for the child to be there,
wide variety of languages it might be possible to establish how
to swing when evening comes.)
general this feature is and whether it is at all formulable in
statistical terms. Even then we WOUld, of course, be left with
or in A wales; bdrdok (The Welsh Bards), by limos Arany:
some examples which run counter to the general tendency;
Ah! Lagyan kel az eji szel
indeed there are pairs of antonyms where the onomatopoeic
Milford abal fe/e.
pattern seems to be reversed, with the [i] sound occurring in the
(Oh! The night breeze rises softly toward Milford Haven.)
term for "large" while its opposite has an open vowel: English
Some at least of these onomatopoeic patterns appear to be
big - small; Russian velikH "great"
malen'kij "little, small."
deeply rooted in our modes of perception, as has been shown
The same may be said of German Riese "giant," Hungarian
recently by psychological experiments. 27
apro "minute," and Latin parvus, though in this last case it is
It is clear, then, that motivation in its various aspects can
perhaps significant that this adjective has not survived in Romance
suggest several promising lines of research which may well lead
and has been replaced by words whose phonetic structure was
better suited to the idea of smallness.
to the discovery of linguistic or stylistic universals.
Onomatopoeia is a popular device in poetry, and there is
2. 2. Particular and general terms
remarkable consistency in the way certain sound-patterns are
used for stylistic purposes in different languages. To cite but
Some languages are remarkably rich in words with specific
one example, a sequence of lateral consonants is particularly
meanings, while others utilize general terms and neglect un
well fitted to produce an impression of softness, as in Keats's
necessary details. French is usually regarded as a highly "ab
lines (Endymion, Book I, 157-158):
stract" language,28 whereas German is fond of "concrete,"
particular terms. It may be noted that "concrete" and "abstract"
Wild thyme, and valley-lilies whiter still
are
used in this context, not in their usual senses, but as syno
Than Leda's love, and cresses from the rill.
nyms of "particular" and "general." There are various symptoms

of this contrast between the two languages:
A famous line in Victor Hugo's poem Booz endormi is built on
the same pattern:
I. In some cases, German has three or four specific verbs
Les souffles de fa nuit flottaient sur Galga/a.
corresponding to one generic term in French: gehen, reiten,

fahren - aller; stehen, sitzen, liegen, hiingen -etre; ste/len, setzen,
228
Stephen Ullmann
legen, hiingen - mettre. The detailed particulars expressed by

the German verbs will often remain unformulated in French,
or will be indicated by the context - unless, of course, there is a
specific need to state them, in which case they will be added
as supplementary information: etre debout, aller a cheval, etc.
2. German, as we have seen, is a highly motivated language
which uses prefixes lavishly to specify every aspect of the action
expressed by the verb. These subsidiary shades of meaning will
normally be neglected in French: setzen, ansetzen - mettre;
schreiben, niederschreiben - ecrire; wachsen, heranwachsen
grandir. In English these nuances tend to be expressed by ad
verbial phrases: to put on, to write down, to grow up.
3. French will often use a derivative where German and
English have a more specific compound: cendrier - ashtray,
Aschenbecher; theiere - teapot, Teekanne; ramoneur -chimney
sweep, Schornsteinfeger.
4. Outside the lexical sphere proper, there are indications
of the same tendency in the German adverbial and prepositional
system, such as the distinction between herein and hinein, her
unter and hinunter, etc., according to the speaker's position,
and the accumulation of adverbs and prepositions to "trace the
whole trajectory" of an action: "Wir segelten vom Ufer her aber
den Fluss hin nach der Insel zu." 29 French and English would
leave most of the details unexpressed.
If a sufficient number of languages were examined from this
point of view, the relative frequency of particular and general
terms might become a useful criterion in linguistic typology,
even though it would be difficult to arrive at precise statistical
conclusions in this field.
Closely connected with this feature is a problem which has
exercised linguists and anthropologists for many years. It has
often been asserted that the languages of "primitive" races are
rich in specific and poor in generic terms. The Tasmanian ab
origines, for example, had no single word for "tree," only special
names for each variety of gum-tree and wattle-tree. The Zulus
have no word for "cow": they must specify whether they mean
a "red cow" or a "white cow," etc. 30 Unfortunately, these reports
SEMANTIC UNIVERSALS
229
were based only too often on inadequate evidence such as

observations by early missionaries, which were uncritically
accepted and reproduced by successive generations of scholars.
Only as late as 1952, for example, did an American linguist
explode the myth that there is no single term for "washing" in
Cherokee. 31 However, the general perpetuation of such assertions
has inevitably tended to bring discredit on the whole theory
of "prelogical mentality": at a symposium on the Sapir-Whorf
hypothesis, held in Chicago in 1953, a philosopher noted that
"everyone was apparently quite willing to talk about the primi
tiveness of a culture but most people were quite unwilling to talk
about the primitiveness of language." 32 One may wonder, how
ever, whether there is not at least a grain of truth in the old
theory. Certain facts in child psychology and in the history of
our own languages seem to suggest that there is. The case of
the Zulu speaker who has separate words for red cow and white
cow is strangely similar to that of a four-year-old Dutch boy
who had special terms for a cow with red spots and one with
black spots; it is true that he also possessed a general word for
"cow" tout court, which was probably due to the influence of his
mother tongue. 33 In the same way, the alleged lack of a word
for "tree" in the language of Tasmanian aborigines reminds one
of the history of Latin planta and its modern descendants. The
Latin word meant "sprout, slip, cutting." There was in Latin
no generic term for "plant" in the modern sense: arbor and
herba were the most comprehensive class-concepts in the botanical
field. According to a recent inquiry, the modern meaning of
"plant" is first found in Albertus Magnus in the thirteenth
century, whereas the French plante did not acquire this wider
sense until 300 years later. 34
It should also be borne in mind that what may seem to us a
plethora of specific terms may be due, not to faulty powers of
abstraction, but to the influence of climate and environment.
Thus it is only to be expected that the Eskimos and the Lapps
should require a variety of terms to distinguish between different
kinds of "snow." Similarly, "the Paiute, a desert people, speak
a language which permits the most detailed description of to po
230
Stephen Ullmann
graphical features, a necessity in a country where complex

directions may be required for the location of water holes." 36
In the words of Edward Sapir, "language is a complex inventory
of all the ideas, interests, and occupations that take up the
attention of the community." 36
In view of the great importance of the problem to linguists and
anthropologists alike, it would be most desirable to organize a
large-scale research project on the whole question of relations
between vocabulary and culture, with special reference to the
use of particular and generic terms at different levels of civiliza
tion and in different environments. Needless to say, the results
of such an inq uiry would be of direct relevance to the Sapir-Whorf
hypothesis and would throw valuable light on the influence of
language upon thought.
2. 3. Synonymy
In his Essai de semantique, Michel Breal put forward a lin

guistic law which he calJed "the law of distribution": words once
synonymous are subsequently differentiated in various ways and
thus cease to be interchangeable. 37 Bloomfield went even further
and argued that total synonymy is impossible in language:
"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." 38 In fact, it does
occasionally happen, in technical nomenclatures, that two syno
nyms which are completely interchangeable live on side by side
for some time, such as for example spirant and fricative in pho
netics, or caecitis and typhlitis in medicine, both of them denoting
an inflammation of the blind gut. 39 Yet it is perfectly true that
we automatically tend to discriminate between synonyms, that
we tend to assume that two or more words different in form
cannot mean exactly the same thing, or cannot mean it in exactly
the same manner. Differentiation may work in a variety of
ways: it may affect the actual content of the words involved,
SEMANTIC UNIVERSALS
231
their emotive overtones, social status, or stylistic register. One

linguist has counted no less than nine distinct ways in which
synonyms can be differentiated. 40 The "law of distribution"
describes an undoubtedly widespread tendency, but not neces
sarily a universal one. There is every reason to believe that
differentiation between synonyms is a sophisticated process
which appears relatively late in the development of a language.
In Old French, for example, a number of synonymous derivatives
could be formed from the verb livrer: Iivrage, /ivraison, livrance,
livre, Iivree, livrement, /ivreiire. Subsequently, this superabundance
was felt to be a mere embarras de richesse and was reduced to
one term: livraison. 41
Another general principle of synonymy is what might be called
the "law of synonymic attraction." It has often been found that
subjects prominent in the interests and activities of a community
tend to attract a large number of synonyms. Some significant
concentrations have, for instance, been discovered in Old English
literature. In Beowulf there are 37 words for "hero" or "prince"
and at least a dozen for "battle" and "fight." The same epic
contains 17 expressions for "sea," to which 13 more may be added
from other Old English poems. 42 An analysis of the vocabulary
of the twelfth-century French poet Benoit de Sainte-Maure tells a
very similar story: 13 verbs for "vanquish," 18 for "attack,"
37 nouns for "fight" and "battle," etc. 43 In slang there are charac
teristic clusters of synonyms, many among them jocular or
euphemistic, for the ideas of "stealing," "drunkenness," and
"death," whereas in French dialects there is a profusion of terms
for "horse," "rich," "poor," and especially for "mean, avari
cious"; the latter vice is described by nearly 200 different expres
sions, 9 of which are found within a single dialect. 44
A special form of attraction is the so-called "radiation of
synonyms," which was first noticed in French slang. 45 It was
found there that when a particular word was given a transferred
sense its synonyms tended to develop on parallel lines. Thus the
verb chiquer "beat" came to be used in the meaning of "deceive,"
whereupon other verbs for "beat"-torcher, taper, estamper,
toquer-received the same secondary sense. Such developments
232
Stephen Ullmann
are sometimes confined to two words: when the English verb

overlook acquired the transferred meaning of "deceive," its
synonym oversee underwent a parallel change. 46 It would be
interesting to find out how widespread these processes are in
different languages.
2. 4. Polysemy
This is the name given, since Breal, to the use of the same
word in two or more distinct meanings. Polysemy is in all prob
ability a semantic universal inherent in the fundamental structure
of language. The alternative to'it is quite unthinkable~ it would
mean that we would have to store in our brains a tremendous
stock of words, with separate names for any possible subject
we might wish to talk about; it would also mean that there
would be no metaphors and that language would thus be robbed
of much of its expressiveness and flexibility. As a philosopher,
W. M. Urban, rightly points out, "this double reference of
verbal signs ... is a basal d(fferentia of semantic meaning. The
fact that a sign can intend one thing without ceasing to intend
another, that, indeed, the very condition of its being an expressive
sign for the second is that it is also a sign for the first, is precisely
what makes language an instrument of knowing." 47
The frequency of polysemy in different languages is a variable
depending on a number of factors. The progress of civilization
will make it necessary not only to form new words but to add
fresh meanings to old ones; in Breal's formula, the more senses
a term has accumulated, the more diverse aspects of intellectual
and social activity it represents. 4R This is probably what Frederick
the Great meant when he saw in the multiplicity of meanings a
sign of the superior quality of the French language. 49 It would
be interesting to explore over a wider field the relation between
polysemy and cultural progress. Meanwhile, the frequency of
polysemy will also depend on purely linguistic factors. As al
ready noted, languages where derivation and composition are
sparingly used will tend to fill gaps in vocabulary by adding new
meanings to existing terms. Similarly, polysemy will arise more
often in generic words whose meaning varies according to
SEMANTIC UNIVERSALS
233
context than in specific terms whose sense is less subject to

variation. The relative frequency of polysemy in various lan
guages may thus provide a further criterion for semantic typology,
though once again it is hard to see how this feature could be
exactly measured.
There is, however, another aspect of polysemy which can be
more precisely quantified: its relation to word-frequency. By
systematically comparing the relative frequency of various
words with the number of senses in which they are used, the late
G. K. Zipf arrived at an interesting conclusion which he termed
the "principle of diversity of meanings." According to Zipf,
there is a "direct relationship between the number of different
meanings of a word and its relative frequency of occurrences." 50
He even tried to find a mathematical formula for this relationship:
his calculations suggested that "different meanings of a word
will tend to be equal to the square root of its relative frequency
(with the possible exception of the few dozen most frequent
words)."51 Put in a different way: m = F'it., where m stands for
the number of meanings and F for relative frequency. 52
Zipf's formula has the great advantage that it can be readily
tested in any language where figures for word-frequencies are
available. On the other hand, the method should be used with
extreme caution. Zipf's count of word-meanings was based on
dictionary material, and it is common knowledge that the lexi
cographer often has to proceed arbitrarily in sorting out the
different senses of a term. In numerous cases there are no sharp
demarcation lines between these senses; many of our concepts
have, as Wittgenstein put it, "blurred edges," 63 and it is im
possible to distinguish consistently between several shades of
the same meaning and several meanings of the same word.
Much will also depend on the comprehensiveness of the various
dictionaries, the extent to which they record technical and semi
technical usage. A count based on the Oxford Dictionary would
produce a very different result from one founded on a more
limited sample. It might even be wiser to refrain altogether from
overprecise formulas when dealing with such vague, subjective,
and unstable phenomena. The broader correlation between
234
"
Stephen UI/mann
polysemy and word-frequency is, however, more plausible and

deserves to be carefully tested in different languages. In fact, it has
always been clear that some of the commonest words in a lan
guage have a great diversity of meanings: in Littre's dictionary,
nearly 40 are listed under aller, nearly 50 under mettre, and some
80 under prendre and faire. 54
Polysemy is a fertile source of ambiguity in language. [n a
limited number of cases, two major meanings of the same word
are differentiated by formal means; for'example, gender (French
Ie pendule "pendulum" -Ia pendule "clock," German der Band
"volume" - das Band "ribbon"); flection (brothers brethren,
hanged - hung, German Worte "conh~cted speech" - Worter
"words"); word-order (ambassador extraordinary - extraordinary
ambassador, French une assertion vraie "a true statement"
un vrai diamant "a real, i. e., genuine, diamond"); spelling (discreet
- discrete, draft - draught, French dessin "drawing" - dessein
"design, plan, scheme"); etc. 55 In the vast majority of cases,
however, the context alone will suffice to exclude all irrelevant
senses. When all these safeguards break down, a conflict between
two or more incompatible meanings will ensue, and this may
lead to the disappearance of some of. these meanings, or even
to that of the word itself. In the present state of our knowledge
it is impossible to say whether there a~e any general tendencies
at work in these conflicts and in the way they are resolved.
A detailed monograph on polysemy in English adjectives sug
gests that such ambiguities will seldom result in the total eclipse
of a word; usually it is sufficient to eliminate one or more of
the conflicting meanings. Out 'of 120 adjectives investigated,
only 3-2~%-have disappeared altogether. 55 Further re
searches may show whether this is or is not a general tendency.
Linguistic geographers have also thrown some light on the con
ditions under which such conflicts are apt to arise. They have
found, for example, that "co-ordinated" meanings belonging to
the same sphere of thought are often an embarrassment, whereas
meanings from different spheres can coexist more easily; thus
it is inconvenient to have the same word for both "maize" and
"sorghum," but perfectly feasible for the same term to mean
SEMANTIC UNIVERSALS
235
"vine-shoot" and "end of a skein." Furthermore, the two mean

ings will not conflict if there is a clearly perceptible connection
between them, as, for instance, in the use of the same word for
"head" and, figuratively, for "nave of a wheeL" The situation
is further complicated by social factors such as the penetration of
the Received Standard into dialect areas. 57 When we have more
data available from various languages, we shall be in a position
to say which, if any, of these tendencies are of general validity.
2. 5. Homonymy
Unlike polysemy, homonymy is not necessarily an unrestricted
universal. Polysemy, as we have seen, is inherent in the very
structure of language, but one could easily imagine an idiom
without any homonyms; it would be, in fact, a more efficient
medium. Whether such an idiom actually exists could be revealed
only by empirical investigations. Even if it does, homonymy is
bound to be a statistical universal with a high degree of probability.
Some homonyms arise through diverging sense-development:
different meanings of the same word move so far away from
each other that they come to be regarded as two separate terms.
This happened, for example, in Englishfiower andfiour where the
difference in spelling underlines the fact that, from a synchronic
point of view, they are two distinct words even though historically
they have a common origin. Not aU cases are so clear-cut; some
times the lexicographer will hesitate whether he has to do with
one word or with two, with polysemy or homonymy. 58 The
great majority of homonyms arise, however, in a different way:
by converging sound-development. This leads to the coincidence
of two or more words which were phonetically distinct at an
earlier date; thus Old English mfte and metan have converged
and become homonymous in Modern English meat and to meet.
Now the chances of such coincidence will mainly depend on
two factors: word-length and word-structure. Languages where
short words abound will obviously have more homonyms than
those where longer words are prevalent. Hence the relative
frequency of homonymy in English and French, as compared,
for example, to German or Italian. Even more important than
236
Stephen Ullmann
length is the productivity of the various types of word-structure

in a particular language. For English we have some interesting
statistics compiled by 8. Trnka, a9 based on an analysis of words
included in the Pocket OxJord Dictionary oj Current English.
Trnka distinguishes 14 types of monosyllables, ranging from
words with one phoneme to those with 6. His tables show that
the commonest type is the eve sequence which, with 1343
monosyllables out of 3178, represents 42 % of the total figure.
The same category also contains the largest number of homo
nyms, 333. In some of the smaller classes, however, the proportion
of homonyms is relatively higher: in the ev combination, for
instance, there are 91 examples out of a total of 174 mono
syllables. In French, the general pattern of word-structure is very
different; there are, in particular, numerous monosyllables con
sisting of a single vowel or of a vowel preceded by one consonant.
Needless to say, the extreme simplicity of this type of word
structure produces a great profusion of homonyms. There are
sometimes as many as half a dozen words consisting of the same
vowel or consonant + vowel: au, aux, eau, haUl, oh, os; ceint,
cinq, sain, saint, sein, seing. 60 If comparable data could be
collected for a great many languages, we could find out whether
there are any universal, or at least widespread, tendencies in
this field; we would also gain a precise typological criterion for
determining the relative frequency of homonymy in general,
and that of its various types.
Homonyms, like several meanings of the same word, are
sometimes differentiated by formal means: gender (French Ie
poele "stove"
la poele "frying-pan," Ie vase "vase, vessel"
la vase "mud"); or flection (English ring, rang - ring, ringed;
German die KieJer "jaws"
die KieJern "firs"). In languages
like English and French, spelling is employed on a massive
scale to differentiate between homonyms, and this is often used
as an argument against spelling-reform. Bloomfield was skeptical
of the value of spelling as a safeguard against homonymy. "It
is wrong to suppose," he claimed, "that writing would be unin
telligible if homonyms (e. g., pear, pair, pare, or piece, peace)
were spelled alike; writing which reproduces the phonemes of
SEMANTIC UNIVERSALS
237
speech is as intelligible as speech." 111 This is doubtless true, but the
point is that writing should be in this respect more intelligible
than speech. English and French suggest that languages rich in
monosyllables, and therefore in homonyms, tend to retain a non
phonetic mode of spelling, and it would not be difficult to estab
lish whether this is a general tendency.
The most important safeguard against homonymic ambiguity

is, however, the influence of context. Many homonyms belong
to different word-classes; others are so diverse in meaning that
they could never occur in the same utterance. Even so, "homo
nymic clashes" happen fairly frequently, and they can be recon
structed with great precision from linguistic atlases. These
clashes, and the various ways they are resolved, have been studied
so thoroughly by Gillieron and other linguistic geographers 62 that
there is no need to discuss them here. In some cases, it is sufficient
to alter slightly the form of one of the homonyms: by giving
French heros a so-called "aspirate h," any possible confusion
between les heros and les zeros, "heroes" and "zeroes," is effect
ively obviated. Elsewhere, a substitute will have to be found;
this may be a derivative, a synonym, a term from the same sphere
or from a neighboring sphere, a borrowing from another lan
guage, or even a jocular metaphor; when, in part of Southwest
France, the words for "cock" and "cat" fell together, the cock
was renamed pheasant" and, more facetiously, "curate." When
we have more geographical and historical facts about such
conflicts in a number of languages, we shall be able to say how
common these various solutions are. It should be noted that
clashes between homonyms, or between different meanings of the
same word, are synchronic phenomena, whereas the changes to
which they give rise are diachronic processes. In this part of
linguistics, a rigid separation of descriptive and historical view
points would be entirely impracticable. The two must be com
bined, without being confused. 63
2. 6. Semantic typology
It will have been noticed that four of the five features examined
in this section-motivation, generic versus specific terms, poly
1:,1
,'i
~
238
Stephen Ullmann
~emy,
and homonymy-may, if studied on a suitable scale,

yield criteria for linguistic typology. All four criteria are statistical:
they are concerned with relative frequencies. The precision
with which they can be .determined will vary with the nature of
the phenomena themselves: it will be highest in homonymy and
lowest in the distinction between generic and specific words; in
the sphere of motivation and polysemy, at least certain aspects
of the problem may be amenable to numerical formulation. Two
further points are also worth noting. First, some of the preceding
features are interrelated: as we have seen, polysemy is closely
connected with motivation on the one hand, and with the use of
generic terms on the other. Second, all our typological criteria,
except, perhaps, motivation, have a direct bearing on the seman
tic autonomy of the word, the degree to which the hearer (or
reader) will depend on the context for understanding it. Obvious
ly, a generic term like French aller means less in itself, and is
therefore more "context-bound," than the more specific German
verbs gehen, reiten, fahren. Similarly, a word with many mean
ings will be highly ambiguous if encountered in isolation, without
any contextual support, as, for example, in a newspaper headline
or the title of a book or a play, whereas homonyms found in
the same isolated position will have no meaning at all. It follows
that languages where generic terms, pQlysemy, and homonymy
are prevalent will be relatively "context-bound"; French is a
classic example of this type of semantic structure, as I have
tried to show in my Precis de semantique franc:aise. Naturally,
the extent to which we have to rely on context in a given language
cannot be stated with any degree 'of precisiQn; yet it may emerge
fairly clearly from close scrutiny of the various factors involved.
3. Universal Processes in Historical Semantics
3. l. Metaphor
3. 1. 1. PARALLEL DEVELOPMENTS. Since metaphor is based on

the perception of similarities, it is only natural that, when an
analogy is obvious, it should give rise to the same metaphor in
various languages; hence, the wide currency of expressions like
SEMANTIC UNIVERSALS
239
the ''foot of a hill" or the "leg of a table." There are, however,

less obvious associations which are also remarkably widespread.
A well-known example is the figurative use of verbs for "holding"
and "grasping" in the sense of "understanding": English grasp,
catch; French comprendre (cf. prendre), saisir; Italian capire
(from Latin capere); German begreifen (cf. greffen), etc. 64 The
great difficulty about such correspondences is that they may not
be genuine cases of parallel development: the various languages
may simply have copied each other or some common model.
Thus, to take a very recent example, the close similarity between
English skyscraper, French graUe-del, Italian grattacielo, German
Wolkenkratzer, etc., is not due to a fundamental identity of
vision; the only spontaneous metaphor among them is skyscraper
which arose in America in the 1890's and was then translated
into other languages. 65 When dealing with earlier periods it will
often be impossible to distinguish systematically between loan
translation and genuine parallelism.
The only way of solving this difficulty is to collect instances
of the same metaphor from widely different languages which
cannot possibly have influenced each other. An impressive
example. of such an inquiry is C. TagUavini's article on the
names of the "pupil of the eye" in various idioms. 66 Among
other things, he has examined the metaphor underlying Latin
pupilla and its modern descendants, where the pupil is compared
to a small girl, or sometimes to a small boy, because of the vague
resemblance between a child and the minute figure reflected in
the eye. This analogy, which may at first seem farfetched, is
embodied in the words for "pupil" in various Indo-European
languages: Greek kore, Spanish nina (del ojo) , Portuguese
men ina (do 6Iho), and many others. But it is equally common in
other linguistic groups: Tagliavini has found examples in some
twenty non-Indo-European languages as remote from each
other as Swahili, Lapp, Chinese, and Samoan.
Such parallel developments are not confined to metaphor:
certain metonymic associations can be equally widespread.
Thus the use of the word for "tongue," the organ of speech, in
the sense of "language" is common to many Indo-European
240
Stephen Ullmann
idioms: English tongue, Latin lingua, Greek glOssa, Russian

jaryk, etc; it is also found in a number of Finno-Ugrian lan
guages, including not only Finnish and Hungarian but even
Zyrian, Cheremiss, and others. The same metonymy occurs also
in Turkish, in some African idioms, and elsewhere. 117 A collection
of such parallel metaphors and metonymies would be of out
standing value since the associations on which they are based seem
to be deeply rooted in human experience and largely independent
of culture and environment. Hence the importance of a project
announced at the London congress of linguists in 1952: the
compilation of a "dictionary of semantic parallels." 118
3. 1. 2. GENERAL TENDENCIES. Over and above these specific
developments, the general movement of metaphors appears to
be governed by some broad tendencies which are of great poten
tial interest not only to the linguist but also to the psychologist,
the literary critic, and others. Only four such tendencies can be
briefly mentioned here:
a. Nearly forty years ago, Hans Sperber put forward a
"semantic law" inspired by Freudian ideas. He started from the
assumption that if we are intensely interested in a subject, it will
provide us with analogies for the description of other expe
riences; in Sperber's terminology, it will become a center of
metaphorical "expansion." Thus the terrifying weapons of the
First World War suggested to French soldiers various jocular
metaphors: beans were described as shrapneis, and a woman
with many children was facetiously referred to as a machine gun
(mitrailleuse a gosses). Sperber summed up his "law" in the
following terms: "If at a certain time a complex of ideas is so
strongly charged with feeling that it causes one word to extend
its sphere and change its meaning, we may confidently expect
that other words belonging to the same emotional complex will
also shift their meaning." 69
Stated in these terms, Sperber's law is no more than a bold
generalization which would have to be extensively tested in
different languages and periods. There are certainly cases where
the principle is applicable. In sixteenth-century France, torn by
religious strife, there were numerous metaphors and similes
SEMANTIC UNIVERSALS
241
derived from the sphere of religion. 70 During the French Revolu

tion, analogies inspired by recent progress in physics and chem
istry were remarkably popular. 71 Subsequently, the introduction
of railways, the spread. of electricity and other technological
inventions enriched the metaphorical resources of the language. 72
It would seem, however, that Sperber's "law" is too categorical.
To take but one example, if there were an automatic connection
between emotion and metaphor, then one would expect our air
minded age to have far more images from aviation than are in
current use today. The same may be said of the application of
this principle to the imagery of a particular writer. While there
are some cases where the explanation works, there are others
where the major interests and preoccupations of an author
have left little or no trace in his metaphors, and an attempt
to reconstruct Shakespeare's "inner biography" from the sources
of his images has met with a very mixed reception. 73 Nevertheless,
there is obviously an element of truth in the theory, and its
implications are so interesting that it deserves to be carefully
investigated.
b. A very common form of metaphor in the most diverse
languages is the anthropomorphic type. This was already clearly
recognized by the eighteenth-century Italian philosopher Giam
battista Vico: "In all languages the majority of expressions
referring to inanimate objects are formed by transfers from the
human body and its parts, from human senses and human
passions.... Ignorant man makes himself the yardstick of the
universe." 74 Thus Vieo did not hesitate to regard anthropo
morphic metaphor as a linguistic universal. Modern linguists
will be more cautious, but there can be no doubt that such
expressions are extremely common in many languages. They can
describe both concrete and abstract experiences: we talk of the
neck of a bottle, the mouth of a river, the eye of a needle, the
brow of a hill, and also of the heart of the matter, the lungs of a
town, the sinews of war, etc. Side by side with these metaphors
from the human sphere, there are many others working in the
opposite direction, where parts of the body are named after
animals or inanimate objects: muscle (from the Latin musculus,
:\1
,J
242
Ij
~
11,
1'
;'~
',\
Stephen Ullmann
SEMANTIC UNIVERSALS
243
Homer and Aeschylus, and also in some ordinary expressions

literally "little mouse"), polypus, apple of the eye, spine, pelvis,
in Greek, such as barytone (from barys "heavy") and oxytone
and others. If wider investigations were to show that both types
(from
oxjs "sharp"); similarly in Latin gravis and aculUS, which
are universal, a further question would arise: which of the two
grave and acute accent. Commenting on such expressions,
our
gave
is the more frequent? A monograph published in 1948 by a
Aristotle
wrote in De Anima: "Acute and grave are here metaphors
Dutch linguist on the semantics of the body 75 suggests that
transferred from their proper sphere, namely, that of touch ....
transfers from the human sphere are more common than those
There seems to be a sort of parallelism between what is acute
directed toward it. In Sperber's terminology, our body is a center
and grave to hearing and what is sharp or blunt to touch." 77
of both metaphorical expansion and attraction, but it acts more
Synaesthetic metaphors have been found in China and Japan,
powerfully in the former than in the latter capacity.
India, Persia, Arabia, Egypt, Babylonia, and Palestine,78 and
c. From concrete to abstract. The fact that, as Bloomfield'
from the language of the K wakiutl Indians Franz Boas quotes the
put it, "refined and abstract meanings largely grow out of con
following powerful image: "the words of speech strike the guests,
crete meanings" 76 is perhaps too well known and too obvious
as a spear strikes the game or the rays of the sun strike the
to require detailed study. It would be most surprising to find a
earth." 79 Our own modern languages abound in such metaphors,
language where metaphors from abstract to concrete are more
some of them hardened into cliches: "cold voice," "piercing
common than those working the other way round. It might be

sound," "loud colors," French couleur criarde, Italian colore
more profitable to examine the extension of certain specific

stridente, and many more. 80 There is a rich literature on various
forms of metaphor within this catego,ry. One such form is the

aspects of synaesthesia, and by casting the net even wider it
wide use of images drawn from light and allied experiences, to

would not be too difficult to find out how general the phenom
denote intellectual and moral phenomena: "to throw light on,"

enon is, and whether it is, in fact, a semantic universal.
"to put in a favorable light," "leading lights," enlighten, iIIu~

Further investigations might also reveal that the movement
minating, brilliant, sparkling, dazzling, coruscating, beaming,

of
synaesthetic metaphors is not haphazard but conforms to a
radiant, etc. Another common pattern is the use of words denoting

basic
pattern. I have collected data for the sources and destina
sense-impressions to describe abstract experiences: "bitter

tions of such images in a dozen nineteenth-century poets, French,
feelings," "sweet disposition," "warm reception," "cold disdain,"

English, and American, and have found three tendencies which
"even temper," and others, To us these associations seem obvious

stood out very clearly: (1) transfers from the lower to the more
and trite; yet only empirical investigations could show how

differentiated senses were more frequent than those in the oppo
general they actually are.

site direction: over 80 % of a total of 2000 examples showed this
d. Synaesthesia. Somewhat akin to this last type are the so

"upward" trend; (2) touch was in each case the largest single
called synaesthetic metaphors where words are transferred from

81
source; and (3) sound was the largest single recipient. The
one sense to another: from touch to sound, from sound to sight,

same tendencies have been noted in some twentieth-century
etc. Since the advent of Symbolism, such transpositions have been

Hungarian poets, 82 and it is interesting to learn that the first
erected into an aesthetic doctrine. Baudelaire proclaimed that

and most important among them, the "hierarchical" principle,
"Ies parfums, les couleurs et les sons se repondent" (Correspon

agrees with the findings of experimental psychology. 83 Naturally,
dances), and Rimbaud wrote a sonnet on the color of vowels

the inquiry will have to be considerably broadened, and extended
(Voyelles). But the modern vogue of synaesthesia should not

from literary style to ordinary language, before we can begin to
obscure the fact that this is an ancient and widespread, and quite
generalize; it should also be borne in mind that these tendencies
possibly a universal, form of metaphor. It is found already in
244
Stephen Ullmann
are purely statistical, and there are bound to be deviations from

them in particular instances. I myself have found such deviations
in the poetry of Victor Hugo, where there are so many syn
aesthetic metaphors derived from the visual sphere that only
the third of the three tendencies is valid: sound is still the main
recipient, but sight takes the place of touch as the chief source
of transpositions, and there is no significant difference between
"upward" and "downward" transfers. 84
3. 2. Extension and restriction oj meaning
Ever since the early days of modern semantics it has been
known that two opposite tendencies are at work in the develop
ment of words: some terms tend to widen their meaning, others
to narrow it. The English bird has extended its range since Old
English times when it was used only in the sense of "young
bird." As the logicians would say, its "extension" has been
increased, and its "intension" has been reduced: it is now applic
able to more things but tells us less about them. On the other
hand, an old synonym of bird,fowl, has developed in the opposite
direction: originally it meant "bird" in general (cf. German
Vogel), as it still does in the Bible: "Behold the Jowls of the air."
Subsequently its range was narrowed down to its present meaning,
which is more distinctive and less comprehensive than the older
sense. 85
Both extension and restriction can result from a variety of
causes, some purely linguistic, others psychological or social.
Nevertheless, several linguists have suggested that restriction
of meaning is on the whole more common than extension. 86 This
has recently been confirmed by some psychological experiments
conducted by Heinz Werner,87 according to whom there are two
main reasons for the disparity. "One is that the predominant
developmental trend is in the direction of differentiation rather
than of synthesis. A second reason, related to the first, is that
the formation of general concepts from specific terms is of lesser
importance in non-scientific communication though it is rather
a characteristic of scientific endeavor. In other words, language
in everyday life is directed toward the concrete and specific
SEMANTIC UNIVERSALS
245
rather than toward the abstract and general." The problem is

of great interest, but we shall need many more facts from dif
ferent languages before we can set up the predominance of
restriction as a semantic universal.
3.3. Taboo
The term taboo is of Polynesian origin, and the very fact that
we use such an exotic word to denote a phenomenon which is
very common in our own culture is symptomatic of the univer
sality of taboo. Here we are concerned only with the linguistic
side of the problem. There is a voluminous literature on the sub
ject, and, as in the case of onomatopoeia, any future research
project could best be started by compiling a critical inventory
of what is already known. Language taboos seem to spring
from three main causes. First, there are those inspired by Jear,
or "holy dread," as Freud preferred to call it: 88 religious restric
tions on the use of the name of God, and also superstitious
avoidance of any direct reference to the dead, to the devil, and
to evil spirits, and the widespread and varied taboos on animals.
A second group is dictated by a sense of delicacy: when we
have to talk of such unpleasant topics as illness and death,
physical or mental deficiencies, and such criminal acts as cheating,
stealing, or killing, we often have recourse to euphemisms, and
this can permanently affect the meaning of the latter: instead of
veiling a tabooed subject, the euphemism will become indis
solubly linked with it, as has happened with undertaker, disease,
imbecile (from Latin imbecillus 9r imbecillis, "weak, feeble"),
and other similar terms. Third, taboo bans may result from a
sense of decency and propriety: references to sex, names of
certain parts and functions of the body, and swear-words are
particularly subject to this form of taboo. While all three types
are of wide currency, none of them is an unrestricted universal
since they are governed by social and cultural factors and will
arise only in certain environments. The first type is bound to
become rarer with the progress of civilization, though it will
not disappear altogether. The second and especially the third
type, on the other hand, will be encouraged, up to a point, by
246
Stephen Ullmann
the development of higher moral standards and more refined

forms of social behavior, though some of these more sophistic
ated taboos may be subsequently rejected as prudish and hypo
critical: we no longer say limbs or benders instead of legs, or
waist instead of body, as did some Boston ladies a hundred years
ago. 89 The growth and decay of the various forms of taboo, in
relation to social and cultural development, could be systemat
ically studied in various languages. Many data are already
scattered in linguistic, anthropological, and psychQlogical
treatises, but they would have to be broadened, classified, and
reinterpreted before definitive conclusions could be reached.
Apart frQm these general tendencies, SQme specific patterns
of taboo and euphemism WQuid alsO' be WQrth lQoking intO'.
Perhaps the most striking feature is the frequency and diversity
Qf taboO's Qn names of animals. A recent mQnograph Qn the
subject, by a Brazilian linguist, 90 cites no less than 24 animals
whQse names have been subjected to such bans in variQus lan
guages. They range frQm ants, bees, and worms to' bears, 01
tigers, and liQns--even butterflies and squirrels appear in the list.
One Qf the mQst remarkable cases is that of the weasel. The fear
inspired by this animal has given rise to' a multiplicity Qf prQpitia
tory euphemisms which are very similar in different languages:
in SQme of them it is described as a "little woman" (Italian
donnola, Portuguese doninha) Qr as a "pretty little WQman"
(French belette, diminutive of belle, Swedish !ilia sniilla), While
elsewhere a pretense is made of including it within the family
by turning it intO' a "bride," a "daughter-in-law," or a "sister
in-law." 92 In other forms of tabQo, toO', there are some interesting
parallel developments; thus the same mixture of euphemism and
irony which gave imbecile its present sense lies at the root of
similar changes in the same sphere: French cretin is a dialectal
form of chretien; benet comes from benedictus "blessed"; English
silly once meant "happy, blessed" (cf. German selig), whereas
idiot goes back to a Greek word meaning "private person,
layman."
As some of these examples show, euphemism, or ironical
"pseudo-euphemism," will Qften lead to' a permanent depreciation
SEMANTIC UNIVERSALS
247
of meaning. The frequency of so-called pejorative sense-develop

ment was nQticed by many early semanticists; 93 some saw in it
a symptom Qf a fundamental streak of pessimism or cynicism in
the human mind. Yet, as Bn!al rightly pointed out, "this alleged
pejorative tendency is the result of a very human disPQsition
to veil and disguise awkward, Qffensive, or repulsive subjects." 94
Thus the notorious deteriQration which has affected variQus
words fQr "girl" or "woman," such as English hussy, quean,
French fllle, garee, or German Dime, was no doubt due to
genuine Qr pseudo-euphemism rather than to' any antifeminine
bias. These and other types of pejorative sense-change-those
arising frQm national or social prejudice Qr frQm a simple asso
ciation of ideas-are sufficiently widespread to' be worth in
vestigatingon a broad interlinguistic basis. Side by side with
these pejQrative changes there are alsO' ameliorative ones 95
where an unpleasant meaning is either weakened Qr even turned
into a favorable one. An example of weakening is the English
blame, which is histQrically the same word as blaspheme,- a case
of positive improvement is English nice from Latin nescius
"ignorant." One has the impression that such ameliorative
changes are less CQmmon than pejorative ones, perhaps because
the ranks of the latter are swelled by euphemisms and pseudo
euphemisms; but this would have to be confirmed by wider
investigatiQns. Another problem which it WQuid be interesting
to explore is the develQpment of neutral terms, "voces mediae,"
which often tend to' specialize either in a favorable or in an
unfavorable meaning. Thus both luck and Jate are in themselves
neutral, ambivalent wQrds, but the adjectives lucky and Jatal
have become PQlarized, the fQrmer in a positive, the latter in
a negative, sense. One wQnders whether there is any predominant
trend of development in one direction or anQther.
"
,i,I
,\j'
'I
(I
~il
'~
1\
,~
~
ij,l
~
~
I.
.,1,\.
3. 4. Implications Jor linguistic reconstruction
The processes discussed in this section, to which several

others could be added, are of direct relevance to etymology and
comparative linguistics. CQmmenting on the traditional study
,~
'I
d
i
~
'!i
,I
248
Stephen Ullmann
of semantic changes, Bloomfield stated that, "aside from its

extralinguistic interest, [it] gives us some measure of probability
by which we can judge of etymologic comparisons." 96 Such
probability would be very considerably increased if some of the
tendencies involved turned out to be semantic universals. This
would help the etymologist and the comparatist in two ways.
First, it would tell him what kind of changes to expect, and
whether a particular change suggested by his data would be
common or infrequent, normal or exceptional. Second, it would
enable him to choose between alternative explanations. Let us
assume, for instance, that the preponderance of synaesthetic
metaphors from the lower to the higher senses was to be shown
by further research to be a semantic universal. Let us also assume
that an etymologist was faced with two early meanings of a
given word, one related to touch, the other to sound. When
deciding as to which of the two meanings came first, it would be
logical to surmise that the tactile sense preceded the acoustic
one, since transfers from touch to sound are far more common
than thQse from sound to touch. It is true, of course, that these
tendencies are purely statistical, and it is perfectly possible that
in a particular case the process worked the other way round.
Nevertheless, a hypothesis which was in harmony with the
general tendency would have a better chance of being correct
than the alternative explanation; it might even be possible to
. calculate the margin of error, which might be large in some
cases and negligible in others.
4. Universal Principles in the Structure of the Vocabulary
During the last three decades, structuralist methods have been
introduced into semantics, and there has been a shift of interest
from single words toward higher lexical units. The importance
of this new orientation can be seen from the fact that "structural
semantics" was on the agenda of the last international congress
of linguists, held in Oslo in 1957. 97 The new science is still in
its infancy, and it is faced with considerable difficulties. While
no one would seriously maintain that the vocabulary is without
SEMANTIC UNIVERSALS
249
any organization, it is clear that the methods of structural

analysis which have been successfully applied to other branches
of linguistics are not immediately applicable to semantics; it
is sufficient to remember that, as the Memorandum points out,
the number of phonemes in any language does not exceed 70,
whereas the Oxford Dictionary is said to contain over 400, 000
words. 98 Despite these difficulties, some encouraging results
have already been obtained,99 and the interest of research
workers is turning more and more toward these matters. These
studies have already raised several problems with universal
implications, three of which will be briefly discussed here. They
arise at three distinct levels of linguistic analysis: at the level of
single words, that of conceptual spheres, and lastly that of the
vocabulary as a whole.
4. I. Lexical constants
A comparison of a' wide variety of languages would quickly
show whether there is such a thing as a "lexical constant": an
object, event, or other feature of such fundamental importance
that it must somehow be expressed in any language; 100 whether
it is expressed by a nonindependent root, a simple word, a
compound word, or even a phrase is of secondary significance.
Even if the evidence for such constants were so overwhelming
that we could set them up as unrestricted universals, we would
still have to allow for differences between various languages.
Assuming, for example, that the idea of fatherhood is a lexical
constant, we find that in Latin there were two words for "father":
genitor for the physiological relationship, and pater, which
carried social connotations (cf. paterfamilias).lOl But this does
not really affect the status of a lexical constant; it merely means
that its various aspects may be expressed by separate words in
some languages.
If a list of lexical constants could be established-whether
as unrestricted universals or as statistical ones with a high degree
of probability-this would be of great interest to comparative
linguistics. When studying the vocabulary of Proto-Indo
250
Stephen Ullmann
European or any other extinct language, we could safely aSSume

that it had some word or other element for the expression of
such constants. In some cases these basic words have survived in
the idioms descended from the proto language, as in English
mother, Latin mater, Greek meter, Sanskrit matar-, etc. Else
where they have been replaced by other terms for a variety of
reasons. Taboo in particular has often disturbed the pattern of
correspondences. Thus the "left hand" is quite possibly a lexical
constant, yet there are different words for it in various Indo
European languages. Some have, in fact, been borrowed from a
foreign source: French gauche from Germanic, Spanish izquierdo
from Basque. This diversity is obviously connected with the
superstitions and taboos which have developed around the left
hand in many countries. Another possible lexical constant, the
"moon," has also been the object of many superstitions which
are still faintly noticeable in our terms lunatic and lunacy. As
Bloomfield points out, "the Indo-European languages use the
most varied words for 'moon'; it is notable that Russian has
borrowed Latin ['Iu: na] as [lu'na], though otherwise it makes
scarcely any but highly learned borrowings from Latin." 102
When the name of a lexical constant is struck by a taboo ban or
drops out of use for some other reason, a replacement has to be
found, and this may lead to the borrowing of words which
would not normally pass from one language to another.
4. 2. Lexical fields
One of the most fruitful concepts evolved so far in structural

semantics is that of the "lexical field," closely associated with
Jost Trier and his school. So much has been written of late on
this subject 103 that it is unnecessary to go into details. It will
be sufficient to recall that lexical fields are highly organized
and integrated conceptual spheres whose elements mutually
delimit each other and derive their significance from the system
as a whole. In each field a sphere of experience, concrete or
abstract, is analyzed, divided up, and classified in a unique way
which embodies a scale of values and a peculiar vision of the
SEMANTIC UNIVERSALS
251
world. Examples of lexical fields are: the system of colors, the

network of family relations; or, among abstract experiences,
the terms for intellectual qualities, ethical and aesthetic values,
religious and mystical experiences.
The numerous articles and monographs which have recently
been published on these problems have all tended to emphasize
the differences between these fields in various languages; they
have concentrated on what is distinctive and idiosyncratic in
them rather than on what they have in common. Yet, beneath
all the diversity, there is likely to be an underlying unity which
a systematic comparison of these fields would no doubt reveal.
Thus we are told of striking differences between the number and
nature of color distinctions: 104 there was no single term for
"brown" or "gray" in Latin; Russian has two words for "blue"
sfnij "dark blue" and goluboj "sky-blue"; the Navaho "have
two terms corresponding to 'black', one denoting the black of
darkness, the other the black of such objects as coal. Our 'gray'
and 'brown', however, correspond to a single term in their
language and likewise our 'blue' and 'green'." 105 These dif
ferences are highly significant, but it would be equally interesting
to know whether there are any elements common to all classi
fications of colors, any distinctions which have to be expressed
everywhere and which could therefore rank as lexical constants.
The same point is even more clearly noticeable in another
closely organized field which has been extensively studied in
various languages: the nomenclature of kinship terms. Take,
for instance, the words for "brother" and "sister." These two
concepts seem so fundamental to us that we find it difficult to
imagine any language that could do without them. Yet a glance
at other idioms will show that they are not in any sense lexical
constants. In Hungarian, there was no single term either for
"brother" or for "sister" until well into the nineteenth century; 106
instead, there were, and still are, two pairs of separate words
for "elder" and "younger brother" and "elder" and "younger
sister." In Malay, on the other hand, there is one collective term
for "sibling," which can also mean "cousin." In his report on
structural semantics to the Oslo congress of linguists, Professor
Stephen Ullmann
252
Hjelmslev summed up the difference between the three solutions

in the following table: 107
Hungarian
"elder brother"
"younger brother"
"elder sister"
"younger sister"
Mtya
ocs
nene
hug
English
Malay
brother
saudara
sister
The three arrangements, though very different, have one thing

in common: the general relationship of "siblings" (children of
the same parent or parents) is expressible in each of them, either
in' itself or combined with other criteria. A comparison of the
same field in a number of languages would reveal whether this
relationship is a semantic universal. It would also show how
many ways there are of "structuring" this part of the field, and
how frequent these various solutions actually are. The same
method could then be applied to other sections of the field. Even
languages belonging to the same family and culture will some
times show remarkable discrepancies. Thus there is no single
term for "grandfather" or "grandmother" in Swedish: a distinc
tion is made between farfar, the father's father, and morfar,
the mother's father, and similarly between farmor and mormor.
Latin had no single word for "uncle" or "aunt": it distinguished
between the father's and the mother's brother (patruus
avunculus) and between the father's and the mother's sister
(amita - matertera); only the two middle terms have survived
English uncle and aunt. In languages with a different social
and cultural background, these discrepancies will be even more
marked. In Dravidian, for example, there is an intricate hierarchy
of kinship terms based on four sets of distinctions: sex, generation,
alliance, and age, of which the third, the only nonbiological one,
is the most important. 108 In Malay, the collective term saudara
"sibling or cousin," which has already been mentioned, can be
subdivided into "younger" and "elder sibling or cousin," and the
latter again into "elder sister or female cousin" and "elder
brother or male cousin." 109
SEMANTIC UNIVERSALS
253
It may be noted in passing that the theory of lexical fields has
certain affinities with the Sapir- Whorf hypothesis. Trier and his
followers would readily agree with Whorf that each language
contains a "hidden metaphysics" and that "we dissect nature
along lines laid down by our native languages." 110 There are,
however, two important differences between the two schools:
(1) lexical fields have so far been explored mainly in the best
known European languages, whereas Whorf deliberately turned
away from "Standard Average European" and concentrated on
totally different linguistic systems, notably the American Indian
ones; (2) the theory of lexical fields is focused on vocabulary,
while Whorf's most impressive successes were obtained in the
grammatical sphere. It would seem, then, that the two approaches,
which have developed independently of each other, 111 could
usefully supplement each other, and the time may come when
they can be combined into a unified theory.
4. 3. The classification of concepts
Experiments have been made time and again to devise a more

rational arrangement for dictionaries than the customary one in
alphabetical order. Roget's Thesaurus was an early attempt in
this direction. In recent years, some closely reasoned schemes
for conceptual dictionaries have been put forward,112 and the
subject has become so topical that it was placed on the agenda
of the London congress of linguists in 1952.113 At that congress,
Professor von Wartburg presented an even more ambitious
project which he and R. Hallig had worked out over a number of
years: a broad classification of concepts applicable to any
vocabulary.ll4 Under three main headings: "The Universe,"
"Man," and "Man and the Universe," concepts are classified
and arranged in such a way that they form an articulate structure
of interdependent elements. The aim of the project is practical:
if a series of monographs on the vocabulary of different languages,
or different periods of the same language, could, within reason,
all conform to the same pattern, the results could be easily com
pared and any differences quickly noticed. Before and since the
i
il
it
254
Stephen Ullmann
publication of the Hallig-Wartburg scheme, a number of studies

on the vocabulary of French writers from various periods have
been based on this system: 1111 it has also been applied to a text
in Romansh. 116 While no one would claim special virtues for
this particular classification, it would be an important step
forward if a system of concepts could be generally accepted as a
uniform yet flexible framework for further lexical studies. 117
S. Conclusion
The list of topics discussed in this paper is not meant to be
exhaustive in any way. I have merely tried to suggest some direc
tions in which we may look for universals. or, more modestly,
for general tendencies in semantics. If a co-ordinated research
program could be organized to explore some of these problems,
then we would have to establish a rough order of priorities,
starting with relatively simple questions and gradually working
our way toward more complex ones. From this purely practical
point of view, the subjects listed previously fall into four broad
categories:
1. It would be best to begin with some clearly defined problems
which could be formulated in precise numerical terms. Such
problems are, for instance, the relation between polysemy and
word-frequency; connections between homonymy and word
structure; the sources and destinations of synaesthetic metaphors;
the number and nature of lexical constants.
2. In the next phase of the program we could proceed to the
study of certain phenomena which are more complicated in
themselves, but about which extensive data are already available
from many languages. Onomatopoeia, taboo, and parallel
metaphors would belong to this category.
3. At a later stage we might be ready to tackle such intricate
matters as the ratio of transparent and opaque words; the pre
ponderance of specific or generic terms; the frequency of pejora
tive and ameliorative sense-change and of extension and restric
tion; the structure of the same lexical field in various languages.
255
SEMANTIC UNIVERSALS
4. Finally, some important research projects will have to wait

until we have the means of collecting the necessary data. Thus,
if there are any general tendencies behind the conflicts caused
by polysemy and homonymy, we shall be in a better position to
discover them when linguistic atlases are available for many
more languages than at present.
If, in the course of such a program, some semantic universals
could be precisely identified, this would be of great significance
not only for linguistics but also for other branches of study.
While some of the problems discussed earlier are of purely
linguistic interest, others clearly have wider implications. To
mention but a few, the distinction between transparent and opaque
words raises important educational issues; onomatopoeic and
metaphorical patterns are of direct relevance to stylistics; synaes
thesia is basically a psychological phenomenon, with wide rami
fications in language and literature. Such problems as taboo
and lexical fields could best be attacked by a concerted effort of
linguists, anthropologists, ethnologists, psychologists, and sociol
ogists. The study of lexical fields, and of the structure of the
vocabulary in general, would also supplement the Sapir-Whorf
hypothesis and throw light on the impact of language upon
thinking, which is one of the main themes of contemporary
philosophy. Among all branches of linguistics, semantics un
doubtedly has the most varied and most intimate contacts with
other disciplines, and the discovery of universals in this field
would have far-reaching repercussions in neighboring spheres.
Acknowledgments
I am indebted to the following colleagues for information and
critical remarks: Professors Harold C. Conklin, Charles F. Hockett,
Fred W. Householder, and Dell H. Hymes.
Notes
1. See H. Kronasser, Handbuch der Semasi%gie, Heidelberg, 1952,
p. 29, and K. Baldinger, Die Semasi%gie, Berlin, 1957, pp. 4f.
256
Stephen Ullmann
2. M. Breal, "Les lois intellectuelles du langage," published in

L' Annuaire de l' Association pour I' encouragement des etudes
grecques en France; quoted by A. W. Read, "An Account of
the Word Semantics," Word, IV (1948), 78-97.
3. F. de Saussure, Couts delinguistique gent!rale, 4th ed., Paris, 1949,
p.132.
4. O. Jespersen, Mankind, Nation, and Individual from a Linguistic
Point of View, Oslo, 1925, p. 212.
5. V. A. Zvegintsev, Semasiologija, Moscow, 1957, p. 46.
6. L. Bloomfield, Language, New York, 1933, p. 20.
7. See R. H. Robins, Ancient and Mediaeval Grammatical Theory in
Europe, London, 1951, pp. 20f.
8. C. F. Hockett, A Course in Modern Linguistics, New York, 1958,
p.123.
9. Bloomfield, op. cit., p. 178.
10. Cf. S. Ullmann, Principles of Semantics, 2nd ed., Glasgow and
Oxford, repro 1959, pp. 58f.
11. Saussure, op. cit., pp. 134f.
12. Cf. Ullmann, "Descriptive Semantics and Linguistic Typology,"
Word, IX (1953), 225-240, and Dell H. Hymes's comments in
Anthropological Linguistics, III (1961), p. 27.
13. Cf. A. Sommerfelt, "Points de vue diachronique, synchronique,
et panchronique en Iinguistique generale," Norsk Tidsskrift for
Sprogvidenskap, IX (1938), 240-249.
14. Saussure, op. cit., p. 100.
15. For references, see Ullmann, Principles of Semantics, pp. 83f., 305.
16. cr. Saussure, op. cit., pp. 180ff. On this problem see, recently,
L. Zawadowski, "The So-called Relative Motivation in Lan
guage," Omagiu lui Iorgu Iordan, Bucharest, 1958, pp. 927-937.
17. Saussure, op. cit., pp. 183f.
18. See V. Grove, The Language Bar, London, 1949, pp. 45f.
19. Cf. U. Weinreich, Language, XXXI (1955), 538. On the need for
statistical data, see ibid. and G. Mounin, Bulletin de la Societe de
Linguistique de Paris, LV (1960), 50.
20. See Grove, op. cit.
21. See M. Wandruszka, "Etymologie und Philosophie," Etymologlca.
Walther von Wartburg zum 70. Geburtstag, Tlibingen, 1958,
pp. 857-871: pp. 858ff. On Fichte see ibid., pp. 866f.
22. See C. Bally, Linguistique generale et linguistique franraise, 3rd
ed.; Berne, 1950, p. 343.
23. See Z. Gombocz, Jelentestan ("Semantics"), Pecs, 1926, p. 12.
24. Cf., recently, M. Chastaing, "Le symbolisme des voyelles. Signifi
cations des I," Journal de Psycho[ogie, LV (1958), 403-423,
461-481.
SEMANTIC UNIVERSALS
257
25. Cf. Jespersen, Language: Its Nature, Development, and Origin,

London, repro 1934, p. 402.
26. See L Fonagy, A kalto; nye[v hangtantib6[ ("From the Phonetics of
the Language of Poetry"), Budapest, 1959, pp. 24ff., 71.
27. See H. Wissemann, Untersuchungen zur Onomatopoiie, I, Heidel
berg, 1954.
28. V. Br0ndal, Le fran~ais, langue abstraite, Copenhagen, 1936;
Bally, op. cit., pp. 346ff.; J. Orr, Words and Sounds in En
glish and French, Oxford, 1953, ch. VIII. Cf. on these matters
C. F. Hockett, "Chinese versus English," in H. Hoijer, ed., Lan
guage in Culture, Chicago, 1954, pp. 106-123.
29. See Bally, op. cit., p. 350.
30. See Jespersen, Language, p. 429.
31. A. A. Hill, "A Note on Primitive Languages," International
Journal of American Linguistics, XVIII (1952), 172-177.
32. A. Kaplan in Language in Culture, op. cit., p. 219.
33. W. Kaper, Kindersprachforschung mit Hilfe des Kindes, Groningen,
1959, p. 11.
34. W. von Wartburg, quoted by K. Baldinger, "L'etymologie hier et
aujourd'hui," Cahiers de I'Association Internationale des Etudes
Franc;aises, XI (1959), 233-264; p. 259 cited.
35. P. Henle, ed., Language, Thought, and Culture, Ann Arbor, 1958,
p. 5.
36. Ibid.
37. M. Breal, Essai de semantiC/ue, 6th ed., Paris, 1924, p. 26.
39. Cf. C. Schick, II Linguaggio, Turin, 1960, p. 188.
40. W. E. Collinson, "Comparative Synonymics," Transactions of the
Philological Society, 1939, pp. 54-77.
41. See F. Brunot and C. Bruneau, Precis de grammaire historique
de la langue franc;aise, 3rd ed., Paris, 1949, p. 172.
42. See O. Jespersen, Growth and Structure of the English Language,
6th ed., Leipzig, 1930, p. 48.
43. See W. von Wartburg, Problemes et methodes de fa linguistique,
Paris, 1946, pp. 175f.
44. Ibid., p. 135.
45. See M. Schwob and G. Guieysse, "Etudes sur I'argot fram;ais,"
Memoires de la Societe de Linguistique de Paris, VII (1892), 33-56,
and more recently B. Migliorini, "Calco e irradiazione sinonimica,"
Boletin del Instituto Caro y Cuervo, IV (1948), 3-17, repro in
Saggi Linguistici, Florence, 1957. As Professor Hockett rightly
points out, "radiation of synonyms" is a special form of analogy.
46. See S. Kroesch, "Analogy as a Factor in Semantic Change,"
Language, II (1926), 35-45.
258
Stephen Ullmann
SEMANTIC UNIVERSALS
259
London, 1937, ch. III. On homonymic clashes in English, see

47. W. M. Urban, Language and Reality, London ed., 1939, pp. 1I2f.
R. J. Menner, "The Conflict of Homonyms in English," Language,
48. Breal, Essai de semantique, p. 144.
XII (I936), 229-244, and E. R. Williams, The Conflict of Homo
49. Ibid.
nyms in English, Yale Studies in English, 100 (1944).
50. G. K. Zipf. 'The Repetition of Words. Time-Perspective, and
63. Cf. Wartburg, loc. cit., and Ullmann, Principles of Semantics,
Semantic Balance," The Journal of General Psychology, XXXIl
pp. I44ff.
(1945), 127-148; p. 144 cited. Cf. also the same author's Human
64. See Gombocz, op. cit., pp. 6f. On parallel metaphors see, recently,
Behavior and the Principle of Least Effort, Cambridge. Mass.,
A. Sauvageot, "A propos des changements semantiques," Journal
1949.
de Psychologle, XLVI (1953),465-472.
51. "The Meaning-Frequency Relationship of Words," The Journal
65. See B. Migliorini, "Grattacielo," Lingua t! cultura, Rome, 1948,
of General Psychology. XXXIII (1945), 251-256; p. 255 cited.
pp.283f.
52. J. Whatmough, Language, A Modern Synthesis, London ed., 1956,
66. C. Tagliavini, "Di alcune denominazioni della pupilla," Annali
p.73.
dell' Istituto Unlversitario di Napoli, N. S., HI (1949), 341-378,
53. L. Wittgenstein. Philosophical Investigations. Macmillan. 1953,
esp. pp. 363ff.
p.34.
67. See Gombocz, op. cit., p. 94; B. CoIlinder, Fenno-Ugric Vocab
54. See Kr. Nyrop. Grammaire historique de la langue frar/raise,
ulary, Stockholm, 1955, pp. 25 (s. v. Finnish kieli), 43 (s. v.
IV: Semantique, Copenhagen. 1913, p. 26.
Hungarian nyelv); G. Revesz, The Origin and Prehistory of
55. Cf. A. Rudskoger, "Fair, Foul, Nice, Proper": A Contribution
Language, London, 1956, pp. 56f.
to the Study of Polysemy, Stockholm, 1952, pp. 473ff.
68. See J. Schropfer, "Wozu ein vergleichendes Worterbuch des Sinn
56. Ibid., p. 439. Cf. also R. J. Menner, "Multiple Meaning and Change
wandels? (Ein Worterbuch semasiologischer Parallelen?)," Pro
of Meaning in English," Language, XXI (1945), 59-76.
ceedings of the Seventh International Congress of Linguistics,
51. See K. Jaberg, Aspects geographiques du langage, Paris, 1936, p. 64.
London, 1956, pp. 366-371.
58. On the boundary between polysemy and homonymy, see R. Godel,
69. H. Sperber, Ein/iihrung in die Bedeutungslehre, Bonn and Leipzig,
"Homonymie et identite," Cahiers Ferdinand de Saussure, VII
1923. p. 61. The English translation is the one given by W. E. Col
(1948), pp. 5-15; P. Diaconescu, "Omonimia ~i polisemia,"
linson in Modern Language Review, XX (1925), 106.
Probleme de Lingvisticii Generalii, I, Bucharest, 1959, pp. 133
70. See E. Huguet, Le langage figure au XV/e sieele, Paris, 1933,
153; M. M. Falkovich, "K voprosu ob omonimij i polisemij,"
pp. 1-18.
Voprosy Jazykoznanija, 1960, no. 5, pp. 85:-88. Cf. also Weinreich,
7 I. See F. Brunot. Histoire de la langue fran~aise, X, 1, pp. 64ff.
Language, XXXI (1955), 541f.
72. Cf. Ullmann, The Image in the Modern French Novel, Cambridge,
59. B. Trnka, A Phonological Analysis of Present-Day Standard
1960, pp. 1400.
English, Prague, 1935, pp. 57-93. See also Jespersen, "Mono
73. On this problem see Wellek and Warren, Theory of Literature.
syllabism in English," Linguistica, Copenhagen and London,
London ed., repro 1954, pp. 214ff., and Ullmann, Style in the
1933, pp. 384-408. As Professor Hockett points out, Trnka's
French Novel, Cambridge, 1957, pp. 31ff.
figures would have to be revised in the light of current methods of
74. Quoted by Gombocz, op. cit., p. 13.
phonemic analysis.
75. J. J. de Witte. De Betekeniswereld van het Iichaam, Nijmegen, 1948.
60. See Bally, op. cit., pp. 269f.; L. C. Harmer, The French Language
76. Bloomfield, op. cit. p. 429.
Today, London, 1954, ch. IV; A. Schonhage, Zur Strukwr des
77. Quoted by W. B. Stanford, Greek Metaphor, Oxford, 1936, p. 49;
franzosischen Wortschatzes. Der franzosische Einsilber, Bonn,
cf. ibid., pp. 53 and 57.
1948, unpubl. thesis reviewed by G. Gougenheim in Le Franrais
78. See A. Wellek, "Das Doppelempfinden im abendlandischen
Moderne, XX (1952), 66--68. Cf. P. Miron, "Recherches sur la
Altertum und Mittelalter," Archiv fur die gesamte Psychologie.
typologie des langues romanes," Atti dell'VIII. Congresso Inter
LXXX (1931), 120-166.
nazionale di Studi Romanzi, Florence, 1960, vol. II, pp. 693-697.
79. F. Boas, "Metaphorical Expressions in the Language of the
.
Kwakiutl Indians," Donum Natalicium Schrijnen, Nijmegen and
62. See esp. Wartburg, op. cit., ch. III; Orr, op. cit., chs. XII-XIII;
Utrecht, 1929, pp. 147-153: p. 148.
I. Iordan and J. Orr, An Introduction to Romance Linguistics,
260
Stephen Ullmann
SEMANTIC UNIVERSALS
261
80. Cf. Gombocz, op. cit., p. 7.

100. Cf. H. J. Pos, "The Foundation of Word-Meanings: Different
Approaches," Lingua, I (1947-1948), 281-292: pp. 289ff.
81. See Ullmann, Principles of Semantics, pp. 266ff.

101. See MeiIlet, op. cit., vol. I, p. 41.
82. See A. H. Whitney, "Synaesthesia in Twentieth-Century Hungarian
Poetry," The Slavonic and East European Review, XXX (1951
103. For a general survey see S. Ohman, "Theories of the 'Linguistic
1952), 444-464.
83. Cf. H. Werner, Language, XXVIII (1952), 256.
Field'," Word, IX (1953), 123-134, and her book, Wortinhalt und
84. See Ullmann, "La transposition dans la poesie lyrique de Hugo,"
Weltbild, Stockholm, 1951. Various aspects of the field theory
Le Fran{:ais Moderne, XIX (1951), 277-295: pp. 287f.
are discussed in Sprache-Schlussel zur Welt. Festschrift fur Leo
85. See H. Schreuder, "On Some Cases of Restriction of Meaning,"
Weisgerber, Dusseldorf, 1959. Cf. also Ullmann, Principles of
*
Semantics, pp. 152ff. and 309ff.
English Studies, XXXVII (l956), 117-124.
104. See esp. I. Meyerson, ed., Prob!emes de la couleur, Paris, 1957.
86. Breal, Essai de semantique, p. 107; Bloomfield, op. cit., p. 151;
105. Henle, op. cit., p. 7.
J. Vendryes, Le langage, Paris, 1921, p. 237.
87. H. Werner, "Change of Meaning: a Study of Semantic Processes
106. The word fiver "brother" and nover "sister" are neologisms
through the Experimental Method," The Journal of General

formed in the late 1830's and early 1840's; see G. Barczi, Magyar
Szdfejto Sz6tar ("Hungarian Etymological Dictionary"), Budapest,
Psychology, L (1954), 181-208; p. 203 cited.

1941 (s. v. fiu and no, ne).
88. S. Freud, Totem and Taboo, London, Pelican Books, repro 1940,
107. Hjelmslev, "Pour une semantique structurale," repro in Essais
p.37.
Iinguistiques, Copenhagen, 1959, pp. 96--113; p. 104 cited. I have
89. Jespersen, Growth and Structure, p. 226.
90. R. F. Mansur Guerios, Tabus Lingiifsticos, Rio de Janeiro, 1956,
replaced the French terms by English ones and the Malay form
..~
sudarii by saudara, as suggested by Professor Conklin.
ch. XVI1l.
91. Cf. A. Meillet, "Quelques hypotheses sur des interdictions de
108. See L. Dumont, "The Dravidian Kinship Terminology as an
,~
vocabulaire dans les langues indo-europeennes," Linguistique
Expression of Marriage," Man, LIII (1953), 34-39; "Hierarchy
historique et linguistique generale, 2 vols., Paris, 1921-1938, vol. I,
and Marriage Alliance in South Indian Kinship," Occasional
pp. 281-291, and M. B. Emeneau, "Taboos on Animal Names,"
Papers of the Royal Anthropological Institute of Great Britain
Language, XXIV (1948), 56--63.
and Ireland, no. 12 (1957).
92. Mansur Guerios, op. cit., pp. 152ff.; Nyrop, op. cit., pp. 275f.
109. I am indebted to Professor Harold C. Conklin for the Malay
93. On these processes see esp. H. Schreuder, Pejorative Sense
data. Cf. also H. Gaiton, Zeitschrift fur Ethn%gie, LXXXII
Development in English, I, Groningen (1929), and K. Jaberg,
(1957), 12/-138; W. H. Goodenough, Language, XXXII (1956),
"Pejorative Bedeutungsentwicklung im Franzosischen," in Zeit
195-216; F. G. Lounsbury, ibid., pp. 158-194; O. N. Trubachov,
schrift fur romanische Philologie, xv (1901), XVII (1903), and
Istorija slavianskich terminov rodstva, Moscow, 1959; L. Weis
XIX (1905).
gerber, Vom Weltbild der deutschen Sprache, 2 vols., 2nd ed.,
94. Breal, Essai de semantique, p. 100.
Dusseldorf, 1953-1954: vol. I, pp. 59ff., and vol. n, pp. 81f.
95. See G. A. van Dongen, Amelioratives in English, I, Rotterdam, 1933.
110. Language, Thought, and Reality, Selected Writings of Benjamin
96. Bloomfield,op. cit., p. 430. On these problems see esp. G. Bonfante,
Lee Whor/. ed. J. B. Carroll, Cambridge, Mass, and New York,
1956, pp. 212f.
"On Reconstruction and Linguistic Method," Word, I (1945),
111. Cf. Weisgerber, op. cit., vol. II, pp. 255ff.
132-161; E. Benveniste, "Problemes semantiques de la reconstruc
tion," Word, X (1954), 251-264; H. M. Hoenigswald, Language
112. See esp. F. Dornseiff, Der deutsche Wortschatz nach Sachgruppen,
Change and Linguistic Reconstruction, Chicago, 1960.
5th ed., Berlin, 1959. See also two articles by K. Baldinger:
97. See the Proceedings of the Eighth International Congress of Lin
"Die Gestaltung des wissenschaftlichen Wtirterbuchs," Romani
guists, Oslo, 1958, pp. 636-704. It appears again on the agenda
stisches Jahrbuch, V (1952), 65-94, and "Alphabetisches oder
of the next congress, to be held in Cambridge, Mass., in 1962.
begrifflich gegliedertes Worterbuch?", Zeitschrifi fur romanische
98. S. Potter, Modern Linguistics, London, 1957, p. 101.
Philologie, LXXVI (1960), 521-536.
99. For a brief account of these results, see the Supplement to the
113. See the Proceedings of the Seventh International Congress of
second ed. of Ullmann, Principles of Semantics.
Linguistics, pp. 77-89, and 343-373.
262
Stephen Ullmann
114. R. Hallig and W. von Wartburg, Begrij}ssystem als Grundlage Jur

die Lexikographie. Versuch eines Ordnungsschemas, Abhandlungen
der deutschen Akademie der Wissenschaften zu Berlin, Klasse
filr Sprachen, Literatur und Kunst, Heft 4, 1952.
115. See the works listed in Wartburg, Problernes et methodes, p. 161,
and H. E. Keller. Etude descriptive sur Ie vocabulaire de Wace,
Berlin. 1953.
116. M. H. J. Fermin, Le vocabulaire de Bi/run dans sa traduction
des quatre Evangiles, Amsterdam. 1954.
117. Cf. Language ill Culture, p. 193.
CHAPTER 9
IMPLICATIONS OF LANGUAGE UNIVERSALS FOR
LINGUISTICS
ROMAN JAKOBSON
Harvard University, Massachusetts Institute of Technology
No doubt, the linguists who are present have responded to the

scientific gain of this stimulating conference with a feeling of
joyful relief. It has often been said that linguistics is a bridge
between the sciences and the humanities, but it was a long time
before the unity of linguistics with the exact sciences became
definitely consolidated.
Hermann Helmholtz (II, 25f. 3)* predicted that "students will
find themselves compelled to go through a stricter course of
training than grammar is in a position to supply." This great
German scientist of the last century was aghast to find evidence
of a "certain indolence and vagueness of thought" in his compa
triot students of grammar, and particularly to note their "laxity
in the application of strictly universal laws. The grammatical
rules, in which they have been exercised, are for the most part
followed by long lists of exceptions; accordingly they are not in
the habit of relying implicitly on the certainty of a legitimate
deduction from a strictly universal law." According to Helmholtz,
the best remedy for these defects "is to be found in mathematics,
where there is absolute certainty in the reasoning, and no author
ity is recognized but that of one's own intelligence."
Our century has witnessed the gradual stages of a spectacular
rapprochement between linguistic and mathematical thought. The
* References are cited by number of appearance in the bibliography.

263
264
Roman Jakobson
gratifying concept of invariance, which in synchronic linguistics

had been first applied for an intralingual comparison of variable
contexts, was finally expanded to an interlingual comparison.
Typological confrontation of diverse languages reveals universal
invariants; or-to quote the inaugural chart of the present
conference, the "Memorandum Concerning Language Univer
sals" prepared by J. H. Greenberg, C. Osgood, and J. Jenkins
"Amid infinite diversity, all languages are, as it were, cut from
the same pattern." We see emerging ever new, unforeseen, but
henceforth perfectly discernible "uniformities of universal scope,"
and we are happy to recognize that the languages of the world
can actually be approached as manifold variations of one world
wide theme-human language.
This outlook is particularly agreeable after the stern opposition
to any TYPOLOGICAL comparison of languages which was current
among American linguists during the 1940's and mutatis mutandis
corresponded to the simultaneous Soviet Russian ban on com
parative HISTORICAL studies by the then dictatorial Marrist
dogma.
The tension between two polar trends-parochial particu
larism and all-embracing solidarity-which Saussure observed
in language (205ff. 15), is true for linguistics as well: "individual
language-oriented definitions" and concentration on differen
tials alone alternate here with a search for common denom
inators. Thus among scholastic theoreticians of language the
renowned Paris savant of the twelfth century, Pierre Helie, declared
that there are as many kinds of grammar as there are languages;
whereas in the thirteenth century, grammatica universalis was
considered indispensable to give grammar a scientific status.
Roger Bacon taught: "Grammatica una et eadem est secundum
substantiam in omnibus linguis, licet accidentaliter varietur"
(43 17). Only today, however, does linguistics have at its disposal
the necessary methodological prerequisites for constructing an
adequate universal model.
The strictly relational, topological character of the cross
language invariants under study has been repeatedly pointed out
in the course of our deliberations. Previous endeavors to define
LANGUAGE UNIVERSALS AND LINGUISTICS
265
the interlingual invariants in ABSOLUTE metrical terms could only

fail. There is an inventory of simple relations common to all
tongues of the world. Such relations pertain both to the early
acquisitions of children's language and to the most stable verbal
properties in those types of aphasic regress which display a
mirror picture of infants' development. This repertory (484ff. 6)
may be exemplified in phonemics by such simple relations as
compact/diffuse (universally displayed in vocalism, and for most
languages also in consonantism), grave/acute (universally dis
played in consonantism and/or in vocalism, in the former almost
universally), and nasal/nonnasal (near-universal in consonantism).
To instance simple relations among grammatical universals, we
may cite the difference between the classes of nouns and verbs
(which assign to their referents the roles of 'existents' and
'occurrents' respectively, as Sapir used to call them: p. 1 13,
p. 123 14). This difference is correlated but never merges with
the likewise universal difference of two syntactic functions
subject and predicate. A few more examples: the particular
class, pronouns (or in Charles Peirce's terms, 'indexical sym
bois': 275ff. 10); the category of number, with its basic distinction
between singular and plural; and the category of person, with
its opposition of impersonal ('third person') and personal forms,
which in turn include an opposition of addressee ('second person')
and addresser (,first person'): the two numbers and the three
persons are universally displayed by pronouns, as J. H. Greenberg
states.
Another and much richer inventory of universals consists of
implicational rules which set a compulsory connection between
two different relational properties of language. Thus in phonemics
the combinability of distinctive features into bundles or sequences
is restricted and determined by a considerable number of universal
implicational rules. For instance, the concurrence of nasality
with the vocalic feature implies its concurrence with the conso
nantal feature. A compact nasal consonant (lp/ or /Tj/) implies
the presence of two diffuse consonants, one acute (In/) and the
other grave (1m/). The acute/grave opposition of compact nasal
consonants Un! vs. /Tj/) implies an identical opposition of com
;i
I
C!
266
Roman Jakobson
pact oral stops (lcl vs. Ik/). Any further tonality opposition of
nasal consonants implies a corresponding opposition of oral
consonants; and any opposition of nasal vowels implies a
corresponding opposition of oral vowels (cf. C. A. Ferguson's
"Assumptions about Nasals").
The present-day inquiry into the hierarchical arrangement of
phonemic systems enables us to uncover the basis for each of
the implicational rules stated. The more complex a phonemic
entity, the less susceptible it is of further fissions. The important
role assigned by the late Viggo Br0ndal to the laws of com
pensation in the grammatical structure of languages (105ff. 1) is
perhaps even more significant for their phonemic patterning
(49 I ff. 6). For example, the marked character of nasals in their
relation to orals results in the lower combinability of nasality
with further features. The marked character of compactness in the
diffuse/compact opposition of consonants explains the near
universal character of compact nasals and the limited spread
of their diffuse counterparts. Inversely, the marked character
of diffuseness in the diffuse/nondiffuse opposition of vowels
explains why there are fewer diffuse than nondiffuse phonemes
among the nasal vowels of the world (cf. Issatschenko 6). On
the other hand, of the two oppositions--':grave/acute and com
pact/diffuse-the former takes primary place in the phonemic
stratification of the consonantal pattern; therefore the com
pact/diffuse opposition of nasals implies their grave/acute
opposition, as shown earlier (cf. Greenberg's forcible conclusions
relevant to the distinctions present in an unmarked morphological
category but neutralized in its marked counterpart).
The grounds for phonemic universals invariably lie in the rela
tional structure of the sound pattern. Thus, for instance, in
languages without the opposition of stops and corresponding
continuants, the obstruents are always implemented eXClusively
or primarily as stops, because it is precisely the stops which stand
in maximum contrast to vowels.
When we examine the few ultimate oppositions which underlie
the whole phonemic structure of language and deal with the laws
of their interrelation, we necessarily resort in the search for
267
interlingual invariants to the same isomorphic principle as in

eliciting intralingual invariants, and thus easily proceed in tracing
the typology of existing phonemic patterns and their universal
foundations. The tenacious belief that maintains the diversity of
languages to be wider in phonemics than in grammar proves
at variance with the facts observed.
The 'logical operations' which H. J. Pos, the outstanding
Dutch theoretician of language, apprehended in the binary
oppositions of distinctive features 12 do indeed give the purely
formal bases for a precise investigation of language typology
and universals. Sol Saporta's segregation of references to vowels,
as "a class defined in formal terms," from references to nasals,
as a "class of phenomena defined in substance terms," is ground
less, because any distributional definition of vowels presupposes
that we identify phonemes in a given position as those possessing
one common oppositive feature, vocality, just as the nasal
phonemes are for us those which carry the oppositive feature
of nasality. In both cases we must deal with relational concepts
superimposed on sensuous data.
The distinction of phonemic entities "universally present by
definition, i.e., universally necessary," like the phoneme, from
those "universally present by empirical observation," like the
syllable, makes no sense whatever. Saporta affirms that "in a
language in which all syllables are exactly one phoneme long,
the distinction between syllable and phoneme disappears"; but
such a language is absolutely impossible, because the only
form of syllable universally admitted is the sequence "consonant
plus vowel." Saporta's assumption is as aimless and arbitrary
as if he referred to some imaginary language where all words
were one phoneme long, or where each phoneme contained but
one feature. The hierarchy of universal linguistic units, from
the utterance to the distinctive feature, must be a formal definition
applicable to world-wide verbal experience. We are faced with
the question of general laws which govern the relations between
linguistic units differing in their rank. Thus, with regard to both
phoneme and word, the smaller the number of phonemes and
their combinations and the shorter the word pattern in a given
""'f-,
268
Roman lakobson
language, the higher is the functional load carried by the pho

nemes. According to J. Kramsky,7 the higher the percentage of
consonants in the code, the lower is the rate of their occurrence
in the corpus. Should this affirmation prove correct, it would
mean that distinctive features tend toward a universally constant
frequency in the corpus.
On the grammatical level, J. H. Greenberg's list of 45 im
plicational universals is an impressive achievement. Even if
advancing research somewhat reduces the number of exception
less universals and increases the sum of near-universals, these
data will remain invaluable and indispensable preliminaries to
a new typology of languages and to a systematic outline of the
universal laws of grammatical stratification. Skeptical reminders
of numerous as-yet-unexplored languages are hardly convincing.
First, the number of languages analyzed or available to analysis
is enormous, and, second, even if there may possibly occur
a further increase of near-universals to the corresponding dimi
nution of exceptionless universals, this result cannot shake the
momentous interest of the inquiry. Statistical uniformities with
a probability slightly less than one are no less significant than
uniformities with probability of one. We may expect, however,
that with the progress of this search and with the refinement of
its methods there will be discovered many new grammatical
universals along with new near-universals.
Greenberg's statements on universals in the "order of meaning
ful elements" rightly put forward the notion of a "dominant"
order. We are reminded that the idea of dominance is not based
on the more frequent occurrence of a given order: actually what
is here introduced into the "order typology" by the notion of
dominance is a stylistic criterion. For example, of the six mathe
matically possible relative orders of nominal subject, verb,
and nominal object-SVO, SOY, VSO, VOS, OSV, and OVS-all
six occur in Russian: The sentence, "Lenin cites Marx," can be
rendered as SVO (Lenin citiruet Marksa), SOY (Lenin Marksa
citiruet), VSO (Citiruet Lenin Marksa), VOS (Citiruet Marksa
Lenin), OSV (Marksa Lenin citiruet), and finally OVS (Marksa
citiruet Lenin) .. yet only the order SVO is stylistically neutral,
269
'I
while all the 'recessive alternatives' are experienced by native

speakers and listeners as diverse emphatic shifts. SVO is the only
word order initially used by Russian children; and in a sentence
like Mama ljubit papu 'Mama loves papa', if the order of words is
inverted-Papu ljubit mama, small children are prone to mis
interpret it: "Papa loves mama." as if one had said, Papa ljubit
mamu. Correspondingly, Greenberg's first universal could be
restated as follows: In declarative sentences with nominal
subject and object, the only or neutral (unmarked) order is almost
always one in which the subject precedes the object. If in a lan
guage like Russian the nominal subject and object are not distin
guished by morphological means, the relative order SO is com
pulsory-Mat' ljubit doc', 'Mother loves daughter'; inversion
of the nouns would mean 'The daughter loves the mother'. In
languages without distinctive characteristics of object and subject,
the order SO is the only one admissible.
The cardinal task of deducing empirical universals "from as
small a number ofgeneral principles as possible"-alreadyachiev
able by and large in phonemics-has been courageously ap
proached by Greenberg on the grammatical level with more than
promising conclusions. Particularly fruitful are his remarks on
what we would call, in Charles Peirce's terminology (I. c.).
the 'iconic' aspect of word order: "The order of elements in
language parallels that in physical experience or the order of
knowledge." The initial position of a word in unemphatic speech
can reflect not only precedence in time but also priority in rank
(the sequence "President and Secretary of State" is far more
usual than the reverse), or it may reflect a primary, irremovable
role within the given message. In the sentences Lenin citiruet
Marksa 'Lenin cites Marx' and Marks citiruetsja Leninym
'Marx is cited by Lenin' (with the recessive alternatives Marks
Leninym citiruetsja, Citiruetsja Marks Leninym, Citiruetsja
Leninym Marks, Leninym Marks citiruetsja, and Leninym
citiruetsja Marks-each variety with its own stylistic shade),
only the first of the two nouns, the subject, is unomissible, but
the oblique term, the accusative Marksa and instrumental
Leninym may be left out. The nearly universal precedence of the
,, ,
j,
fj!
'1 1
'l~
270
Roman Jakobson
subject with regard to the object, at least in unmarked construc

tions, points to a hierarchy in focusing. It is not by chance that
Greenberg's paper treats the universals of grammar "with
particular reference to the order of meaningful elements" (syn
tactic or morphological constituents).
In general, the 'iconic symbols' of language display a par
ticularly clear-cut universalistic propensity. Thus, within a
grammatical correlation the zero affix cannot be steadily assigned
to the marked category and a 'nonzero' (real) affix to the un
marked category. For example, according to Greenberg, "There
is no language in which the plural does not have some nonzero
allomorphs, whereas there are languages in which the singular
is expressed only by zero. The dual and trial are almost never
expressed only by zero." In a declensional pattern, the zero
case ("which includes among its meanings that of the subject of
the intransitive verb") is treated like the singular in respect to the
other numbers. Briefly, language tends to avoid any chiasmus
between pairs of unmarked/marked categories, on the one hand,
and pairs of zero/nonzero affixes (or of simple/compound gram
matical forms), on the other hand.
Phonemic experience may yield some useful stimuli for the
investigation and interpretation of grammatical universals. In
particular, one may expect the order of children's acquisitions
and of aphasic losses to throw new light on the stratification of
morphological and syntactic systems.
As we have already observed, the unaccountable fear of a slip

into the phonetic substance may hamper the phonemic typology
of languages and the discovery of general phonological laws.
Likewise, the exclusion of semantic considerations (which has
been a tantalizing experiment in grammatical descriptions)
would be, with respect to typology, a flat contradiction in terms.
One must agree with Greenberg that it would be impossible to
identify grammatical phenomena in languages of differing
structure without "employing semantic criteria." Morphological
and syntactic typology and universal grammar as its ground
work deal primarily with 'grammatical concepts', by Sapir's
designation. It is obvious that in grammar there is no conceptual
271
opposition without a corresponding formal distinction, but

neither on the intralingual nor on the interlingual level is this
distinction supposed to use one and the same 'grammatical
process'. Thus in English the one opposition singular/plural is
expressed either by suffixation or by vocalic alternation (boy: boys;
man: men), If one language expresses this opposition by suffixa
tion only and another only by vowel alternation, the basic distinc
tion of two grammatical numbers nonetheless proves to be
common to both languages.
Not only grammatical concepts but also their interconnection
with grammatical processes (exemplified earlier by the analysis
of word order), and, finally, the structural principles of such
processes, call for an extraction of implicational universals,
Fortunately, in his quest for the universals of grammar Green
berg does not share the whimsical prejudice against "semantics
oriented definitions," which, strange as it seems, may have
filtered even into our Conference on Language Universals. One
must fully approve Uriel Weinreich's witty remark, that if in
phonology we had only a couple of commonplace statements on
all-language properties, "we would hardly be meeting for a
conference on phonological universals," and, again, that isolated
truisms about the universal semantic properties of languages
offer "not much to go on," A realistic approach to this field,
however, opens an ever-widening prospect for new high-level
generalizations. A conditio sine qua non of such inquiry is the
consistent distinction between grammatical and lexical mean
ings (or, in Fortunatov's terms, the formal and the real meanings:
see ch. 7 11), which, despite methodological itineraries traced
particularly by the outstanding 'American and Russian path
finders in linguistics, still bewilders and confuses some students
of language. Some of them even seem to be non pi used by rudi
mentary questions: What, for example, does the plural or the
past tense or the inanimate gender actually mean in a verbal
code? and does it possess ingeneral any meaning?
A cautious and unremitting search for the intralingual and
therewith interlingual semantic invariants in the correlations of
such grammatical categories as, for example, verbal aspects,
~~,
~
t'i
11
!l
;1,
272
"
f~
Ii"
:\;
rll
ii
\1
Roman lakobson
tenses, voices, and moods becomes indeed an imperative and

perfectly attainable goal in present-day linguistic science. This
inquiry will enable us to identify equivalent grammatical opposi
tions within "languages of differing structure" and to seek the
universal rules of implication which connect some of these
oppositions with one another. The great mathematician A. Kol
mogorov, an expert also in the science of language, has judiciously
defined grammatical cases as those classes of nouns which
express "wholly equivalent states" in regard to their referent
(absoljutno ekvivalentnye sostojanija otnositel'no dannogo pred
meta 18). We analyze a grammatical case into its componential
semantic properties and treat these componential properties just
as we do distinctive features in phonemics: that is, we define
both as terms of INVARIANT oppositions and, correspondingly, as
VARIABLES, dependent on different contexts or on different sub
codes (styles of language). Incidentally, though it happens that in
certain contexts the use of a given case is compulsory and that
in this instance its meaning turns out to be redundant, this cir
cumstance does not allow us to equate even so predictable a
meaning with meaninglessness. It would be a sheer misunder
standing to imagine that these occasional redundancies might
invalidate to any extent the search for the general meanings of
grammatical cases. It is true that the Russian preposition k 'to'
implies the dative case subsequent, but the Russian dative does
not imply an antecedent preposition k and thus preserves its own
general meaning of 'direction toward', just as the Russian noun
xleb 'bread' does not lose its meaning when preceded by the
adjective peklevannyj 'wholemeal', although xleb is the only
noun one can expect after this attribute. In a sequence of two
English obstruents, if the first is voiceless, the second too must
be voiceless: [kukt] cooked. In this instance, however, the appar
ent analogy between the grammatical and phonemic sequences
is misleading. Redundancy does deprive the phonemic feature
of its distinctive value, but it cannot rob meaningful units of
their proper sense.
Naive attempts to deal with variations without attacking the
problem of invariants are condemned to failure. Such ventures
273
change the case system from a hierarchic structure to a sum

mative aggregate and hide the implicational universals which
actually form the pivot of the declensional pattern. An inter
lingual difference in contextual variants does not affect the equiv
alence of invariant oppositions. Though the genitive of negation
exists in Polish and Gothic but not in Czech or Ancient Greek,
the genitive does act as a quantifier in all four of these languages.
At present "there is an ineradicable conviction," as H. M.
Hoenigswald noted in his thoughtful paper, "that universals may
form some sort of system in their own right." The high number
of grammatical universals based on "semantic criteria" eloquently
proves the failure of the traditional belief cited by Weinreich,
that "the semantic mapping of the universe by a language is, in
principle, arbitrary."
The most profitable part of Weinreich's paper "On Semantic
Universals" is his effort to answer the question, "What general
izations can be made about any vocabulary as a structured set,
imperfect as the structuring may be1" The thoughts on language
by Weinreich's six-year-old daughter (transmitted to us by her
father between conference meetings) offer a particularly valuable
and realistic supplement to his argumentation. "The standard
works on semantics," Weinreich states, "are on the whole pre
occupied with the one semiotic process of naming." His daughter,
surprised to learn that there are thousands and thousands of
words in a language, surmises that most of them must'be "names"
(she means nouns), and, on the other hand, grants that this high
number of words is not so overpowering, since they go by pairs
(of antonyms), as up and down, man and woman. Water, the little
Shifra reasons, must be countered with dry, and to buy with to
make oneself(since she is accustomed to buying but not to selling,
there is no word alternation buy-sell in her thought). The astute
child has observed two important properties of vocabulary: its
structured arrangement, and the different status of diverse word
classes, especially the more open, expandable character of the
noun class.
The study of lexical patterning would be easier and more pro
ductive if it began not, as usual, with nouns but with more closely
274
Roman Jakobson
circumscribed word classes. Then the bonds between semantic

subclasses and their different syntactic treatment would prove
particularly revealing. Thus the research started by Professor
Gerta Worth (U. C. L. A.) within the frame of our Harvard
teamwork (Description and Analysis of Contemporary Standard
Russian) shows that the division of all Russian primary (un
prefixed) verbs into those which must or mayor cannot be com
bined with a given case or with an infinitive results in a set of
verbal classes, substantiated both formally and semanticaly.
A similar twofold delineation of nominal classes is more laborious
but still feasible. For instance, in Slavic and many other lan
guages, the class of nouns designating an extent of time is syn
tactically grounded by the fact that only they can be used in the
accusative with intransitive verbs (Russian bolel nedelju 'was ill a
week') and as a second accusative with transitive verbs (Russian
gody pisal knigu 'for years was writing a book'). An intralingual
classification of words which would at last tie together the
problems of lexicology and grammar is an essential prerequisite
to the cross-language investigation of lexical uniformities.
We have observed that the common joy at the universal
outlook of this conference threatened to change into a feeling of
frustration when the final debates over the prospective organiza
tion and further advancement of research proved inconclusive.
Since it is clear that typology and universals cannot be removed
from the agenda, and since without continuous collective effort
this research cannot be adequately promoted, I shall propose at
least one concrete measure.
We most urgently need a systematic, world-wide mapping of
linguistic structural properties: distinctive features, inherent
and prosodic-their types of concurrence and concatenation;
grammatical concepts and the principles of their expression.
The primary and less difficult task would be to prepare a phonemic
atlas of the world. Preliminary discussions aiming toward such
an atlas had been undertaken at an international meeting of
phonologists in Copenhagen, August 29, 1936, and were extended
in 1939-1940 by the remarkable community of Oslo linguists,
but were suspended at the German invasion. Today our Iin-
275
guistic section of the Center for Communication Sciences at

M. I. T. is desirous of inaugurating work on this atlas, but to
realize this project would require the wide cooperation of the
Social Science Research Council and of its Committee on Lin
guistics and Psychology. Linguists of different centers in this
country and abroad are to b~ enlisted in the work of our
~am.
,
The number of languages and'dialects whose phonemic make
up is already accessible to linguists is fairly high, but-let us
admit-at the beginning there will be controversial questions,
and some blanks will remain on our maps. Nevertheless, the
existence of unexplored areas can never be used as an argument
against mapping. The isophones 9btained, even if they should be
only approximate, will be immensely useful to linguistics and
anthropology. Matched with one another, these isoglosses will,
no doubt, reveal new implicational rules and present the phonemic
typology of languages in its geographical aspect. The phonemic
affinities of contiguous languages due to the wide diffusion of
phonemic features will be exhaustively displayed by the atlas.
Work on phonemic and gramm~tical atlases of the world will be
only a part of that vast internatronal cooperation which is neces
sary to reach the grand aims advanced by our conference.
To conclude: We all seem to agree that linguistics is passing
from the bare study of varit~gated languages and language
families, through systematic TYPOLOGICAL research and gradual
INTEGRATION, to become a thoroughly universal science of lan
guage. For centuries this field hq.s been a no-man's land, and only
a few philosophical contributiQ,ns-from the medieval treatises
on grammatica speculaliva, thrQugh John Amos Comenius' Glol
t%gia 9 and the rationalist essays of the seventeenth and eight
eenth centuries, to H usserl's 4 :and Marty's 8 phenomenological
meditations, and finally to the ~odern works in symbolic logic
have ventured to lay the found~lions for a universal grammar.
When questioned by my examiner in Moscow University
about the possibility of a universal grammar, I answered by
quoting that professor's negative view of Husserl's reine Gram
matik. There followed a demand for my own attitude; to the
276
Roman Jakobson
questioner's vexation I replied by advancing the necessity for

linguists' research in this field.
If present-day linguists finally turn to these problems, equipped
as they are with a strict methodology of their own and a rich
factual knowledge, they should revise and correct the extant
theoretical constructions; but by no means are they justified
ignoring or in underrating the abundant philosophical hints of the
past and of the present with the dubious excuse that in this
literature one meets now and then with a priori statements and
with inattention to verifiable realities. Thus Weinreich's indis
criminate rejection of the allegedly "new scholasticism" in
Carnap's and Quine's recent writings is hardly warrantable.
Likewise, the philosophers' distinction between autocategor
ematic and syncategorematic signs remains vital for the con
struction of a universal grammar even if some of its traditional
interpretations have proved to be "totally untenable." A careful
empirical check of the various general principles introduced by
philosophical grammar may be an effective auxiliary in the
guistic investigation of universals and a welcome preventive
measure against uneconomical, superfluous rediscoveries and
against the dangerous fallacies with which the so-called creeping
empiricism is too often menaced.
This conference has eloquently testified that isolationism in
its various shapes vanishes from linguistics when the device of
technical separation has served its useful experimental end.
The particular and the universal emerge as two correlated
moments, and their synthesis reaffirms the irresolvable unity of
the outer and inner side of any verbal sign. Linguistics is becoming
aware of its interconnection with the adjacent sciences of lan
guage, thought, and communication, and it strives to define
both the particular characteristics of language and its intimate
affinities with other sign systems. The question of language
universals inevitably raises the wider problem of the over-all
semiotic constants. The inside view of language is now supple
mented by a comparison of the verbal pattern with other vehicles
of human communication. The intensive collaboration of linguists
with cultural anthropologists and psychologists in the Conference
277
on Language Universals indicates that the present-day linguist

is about to reject the apocryphal epilogue which the editors of
Saussure's Cours added in italics: "The true and unique object
of linguistics ;s language studied in and for itself." 15.2 Do we
today not conceive language as a whole "in and for itself" and
simultaneously as a constituent part of culture and society?
Thus linguistics becomes a two-fronted science persistently con
cerned with the interrelation of whole-and-part aspects. Finally,
the question acutely raised by H. M. Hoenigswald and vividly
discussed here-"Are There Universals of Linguistic Change?"
-has enabled us to expose the most rigid of the habitual segrega
tions, the fictitious chasm between the study of constancy and
changes. The quest for universals is organically linked with all
other manifestations of a unitary attitude toward language and
linguistics.
References
1. Br~ndal, V. (1943). Essais de linguistique generale. Copenhagen.
2. Godel, R. (1957). Les sources manuscrites du Cours de linguistique
generale de F. de Saussure. Geneve.
3. Helmholtz, H. (1900). Popular Lectures on Scientific Subjects.
New York.
4. Husserl, E. (1913). Logische Untersuchungen, II. 2nd ed., Hallea. S.
5. Issatschenko, A. (1937). "A propos des voyelles nasales," Bulletin
de la Societe de Linguistique de Paris, 113. 267ff.
6. Jakobson, R. (1962). Selected Writings, I. The Hague.
7. Kramsky, J. (1946-1948). "Fonologicke vyuzitf samohlaskovych
fonemat," Linguistica Slovaca, 4-5. 39fT.
8. Marty, A. (1908). Untersuchungen zur Grundlegung der allgemeinen
Grammatik und Sprachphilosophie. Halle a. S.
9. MiSkovska, V. F. (1959). "La Panglottie de J. A. Komensky,"
Philologica Pragensia, II. 4. 97ff.
10. Peirce, C. S. (1932). Collected Papers, II. Cambridge, Mass.
11. Porzezinski, W. (1913). Vvedenie v jazykovedenie. 3rd ed., Moscow.
12. Pos, H. J. (1939). "Perspectives du structuralisme," Travaux du
Cercle Linguistique de Prague, VIII. 711f.
13. Sapir, E. (1930). Totality (LSA, Language Monographs, 6).
278
Roman Jakobson
14. Sapir, E. (1949). Selected Writings. Berkeley and Los Angeles.

15. Saussure, F. de. (1959). Course in General Linguistics. New York.
16. Uspenskij, V. A. (1957). "K opredeleniju padeza po A. N. Kolmo
gorovu," Bjulleten' Ob"edinenija po problemam mainnogo pere
voda, 5. 11ff. Moscow.
17. Wallerand, O. (19J3). Les tuvres de Siger de Courlrai. Louvain.
CHAPTER 10
LANGUAGE UNIVERSALS IN ANTHROPOLOGICAL
PERSPECTIVE
J.
JOSEPH B. CASAGRANDE
1. Introduction
Ever since the early days at Columbia University of our com

mon progenitor, Franz Boas, anthropology and linguistics have
in the United States been regarded as kindred disciplines. Boas'
students were regularly exposed to the intricacies of American
Indian languages in one of the main courses he regularly offered,
the other being one in "Statistical Theory" (Steward, 1961,
p. 1042). Thenceforth linguistics has always figured importantly in
the graduate training of American anthropologists. Linguists are
fully accredited members of most of the larger departments of
anthropology, and many anthropologists have contributed sig
nificantly to both fields-one need only mention Sapir and
Kroeber-a tradition that is at least as alive today in the work of
many younger people as it was in the past. And Boas also stands
secure among the intellectual forebears of American linguistics.
Indeed, Emeneau (1943, p. 35) has characterized Boas as "the
guru, the ancestor in learning, of all those in this country who
work in descriptive linguistics." But beyond this historical and
cognatic link. there are good substantive, theoretical, and method
ological reasons for the close and continuing affinity between the
two fields. I say "two fields" because, while many might regard
linguistics as a subfield of anthropology, or as a special kind of
ethnography, the two have had, particularly abroad, quite
279
280
independent if sometimes merging traditions. Anthropological

linguistics is primarily an American phenomenon. 1
Despite the relative autonomy of language as a subsystem and
the fact that it can have a separate historical career, we are widely
agreed that language is a part of culture, perfectly meeting the
criteria for anyone's definition of culture. In fact, language and
the ideological components of culture for which language is
prerequisite can be thought of as culture in its "purest" form,
that is, a kind of behavior unique to man, discontinuous with other
species, and for which there are not readily identifiable behavioral
analogues traceable to biosocial imperatives shared with other
animals. If, then, language is of the essence of culture, it behooves
us as general anthropologists to attend to what the linguist has to
say, and to ask linguists what light their studies can throw on the
nature of man, and especially on man as a symbol maker and user.
Anthropologists may perhaps need to be reminded from time to
time that as students of culture they are bound to have an interest
in linguistics despite their oft-expressed dismay with the technical
demands of the subject. Both fields stand to lose if they follow
separate paths. These are, of course, all familiar views. I rehearse
them here only to suggest that anthropologists can be expected to
have a natural interest in language universals and their ramified
implications.
2. Universals in Language and Culture
Among many parallel problems, anthropology and linguistics
share that of a dual concern with the particular and the general
in culture and in language. We both have the task of uncovering
the common pattern, or the universal design, that underlies the
exuberant variety of the particular configurations that we call cul
tures and languages, as well as that of accounting for the latter
themselves. 2 That there exist general designs for languages and
for cultures is a belief frequently affirmed by anthropologists
and linguists alike if seldom demonstrated in detail. Linton's
statement (1952, p. 646) is typical: "Behind the seemingly endless
diversity of culture patterns there is a fundamental uniformity."
LANGUAGE UNIVERSALS AND ANTHROPOLOGY
281
Clearly, the existence of these uniformities is implied in the

methods, concepts, and categories that are applied cross-culturally
and cross-linguistically by both anthropologists and linguists. And
in common-sense terms the reality of a universal pattern is attested
to by the fact that men can and do learn each other's languages
and, although perhaps with greater difficulty, can come to appre
ciate each other's cultures, if only by imaginative participation.
It is also interesting to note that the relative concern with the
particular and the general has shown the same vicissitudes in
anthropology and linguistics. The "three stages" of ethno
centricism, descriptive empiricism, and now of comparative
structuralism, to give them somewhat arbitrary labels, that
Hoenigswald (p. 30) describes for linguistics are equally applicable
for anthropology. Boas in anthropology, as Bloomfield in lin
guistics, marks the watershed; and both were looming figures on
this intellectual continental divide. Indeed, if one were to substitute
the word "culture" for "language" in the passage from Bloomfield
quoted by Ullmann (p. 218), one can virtually hear Boas uttering
it in the classroom. Among his students, perhaps only Sapir's
attention would be wandering.
This is not the place to recount in detail the history of anthro
pological concern with universals-it goes back well into the last
century-but let me mention briefly a few more recent writings,
particularly as they touch on language. Wissler in Man and
Culture (1923) called attention to what he termed the "universal
culture pattern" and the "skeleton of culture," a set of categories
which, taken together, Wissler believed, cover the entire range of
culture content. In discussing speech, which constituted one of
these nine "complexes" 3 in "the culture scheme," he observed
that "not only are all languages set up on the same lines, like so
many watches in the jeweler's show-case, but ... they are based
upon that which is fundamental in the mind" (p. 84).
More recently, Murdock (1945) and Kluckhohn (1953) have
written on the problem of cultural universals. Noteworthy, too,
is the Outlineo! Cultural Materials, now in its fourth edition (1961),
developed by Murdock and his colleagues as a set of categories
under which the massive materials in the Human Relations Area
282
283
,
Files are exhaustively catalogued. Murdock and Kluckhohn agree

that cultural universals are essentially similarities in classification,
not content. Kluckhohn describes universals as "invariant points
of reference for description and comparison" and as "substitutive
uniformities or near-uniformities." Both give some attention to
linguistics. Thus Murdock lists as "common denominators of
culture" greetings, joking, kinShip nomenclature, personal names,
numerals, and language itself. Kluckhohn gives more attention to
linguistics, both by citing examples of linguistic universals (e.g.,
"possession or the genitive is expressed in all languages," p. 517),
and by referring to linguistics as a modei to be emulated in the
quest for cultural universals. He notes that "Linguistics alone has
discovered elemental units which are universal, objective, and
theoretically meaningful" (p. 507). And, he adds, "In cultural
anthropology we are still too close to the phase in linguistics when
non-European languages were being forcibly recast into the
categories of Latin grammar" (p. 508).
On reading these several anthropological treatments of uni
versals, one is immediately struck by the fact that they parallel to
a great degree the discussion at this Conference. Clearly, here is
a common anthropological problem embracing both language and
culture. The several types of universals discussed in the Memo
randum, while. not identified as such or similarly classified, are
adumbrated in Kluckhohn's review article. And it is evident that,
understood in the sense of the Memorandum, the quest for
universals is the chief inspiration of current cross-cultural
research, whether synchronic or diachronic, as, for example,
represented by Murdock (1949) and his students (Whiting and
Child, 1953; Whiting, 1954) and Steward (1949), respectively.
However, in the case of anthropology, universals is usually taken
to refer to true or "unrestricted" universals rather than "statistical
universals," and in reference to the latter one speaks in terms
such as cross-cultural regularities, covariation, correlations, ad
hesions, types (as in Omaha- or Crow-type kinship system), and the
like.
One additional similarity in approach may be noted. This is
the fact that explanations for the occurrence of universals in both
language and culture, as touched upon in Lamb's and Osgood's

comments at this Conference and as discussed by Murdock and
Kluckhohn in the works cited previously, are couched in terms of
underlying determinant factors and not in terms of the phenomena
of language or culture per se. Thus Kluckhohn writes, quoting a
memorandum of which hewas a co-author, "These [invariant points
ofreference] are to be found in the nature of social systems, in the
biological and psychological nature of the component individuals,
in the external conditions in which they live and act, in the nature
of action itself, in the necessity of its coordination in social
systems" (1953, p. 513). In a sense, then, and as Osgood has sug
gested in his summary statement, these underlying factors become
the universals, and cultural and linguistic universals are epiphe
nomena. Presumably, many of the same or analogous factors
operate to determine both cultural and linguistic universals,
although one might want to specify some of them a bit differently,
for example, in the case of linguistic universals, in terms of neuro
logical, physiological, and psychological capabilities and capac
ities, and of the nature, functions, and content (including cultural
content) of communication itself..
Cultural anthropologists will doubtless be most interested in
semantic universals since it is in this domain of language that the
influence of culture is most transparent. Yet universals at all levels
can teach us something about the nature of man, and we may also
assume that physical anthropologists, anatomists, and neurol
ogists (e.g., Spuhler, 1959; Penfield and Roberts, 1959; Du Bruhl,
1958) will also be interested in universals insofar as they throw
light on the structural requirements for a bfain and speech
apparatus capable of creating and using language of a level of
complexity that is consistent with minimal cultural needs, a topic
to which we shall return later. Moreover, what we do know, or can
reasonably infer, about the nature of man and his culture may help
to elucidate the universals we can discover, even at the phonolog
ical level, which at first blush may seem far removed from socio
cultural influences. Thus the universality of the nuclear family and
commonalities in teaching-learning processes during the early
years of human socialization may work in concert with the
I)
i:
284
phasing of the maturational processes and the relatively long

period of infant dependency to produce phonological universals
such as bilabials, nasals, and reduplication in primary kinship
terms and baby words (Casagrande, 1949; Murdock, 1959). It is
conceivable that these universals might even have been shared
with our early Hominoid precursors.
As we have previously noted, semantic universals are of prime
anthropological interest because of their cultural content and im
plications. The universal features in descriptive semantics as dis
cussed in Ullmann's contribution are a case in point. For example,
the proliferation and specificity of terms, the higher degree of cod
ability, and the greater number of synonyms in areas of special
cultural interest are matters of perduring anthropological concern.
Similarly, anthropologists will find Ullmann's comments regarding
the relations between semantics and civilization (e.g., the correla~ ,
tion of polysemy and progress, particularism and primitiveness)
suggestive if debatable. Other topics taken up by Ullmann, such
as the widespread parallelisms in metaphors of various types and
in metollymy, and the role of language taboos, are of obvious
anthropological interest. One is reminded here of what Bastian
years ago called Elemen targedanken. And as we shall discuss in
more detail later, the common structural principles underlying
these more superficial semantic phenomena are, I believe, of
fundamental' interest to anthropologists, especially to those con
cerned with new developments in the comparative study of
cognitive processes.
Given their predilection for history, many anthropologists will
doubtless show keen interest in diachronic universals. That this
is so is evident from the eager if uncritical reception accorded
glottochronology by archaeologists and ethnohistorians who
seized upon the proposition that basic vocabulary is replaced at
a constant rate as another time-clock comparable to C 14 dating
and the like. If it could be ~hown, with perhaps greater reliability,
that phonological changes occur at a constant rate, or within a
fairly limited range of rates, clearly such a finding would be a boon
to historically minded anthropologists. In fact, a priori one might
conjecture that the rate of phonological change might be more
285
regular than that of lexical substitution since it is less immediately

susceptible to the vagaries of cultural interference. Moreover,
phonological change is less discontinuous than lexical change,
increments of change are doubtless smaller (perhaps involving
only a single distinctive feature at a time), and tend to occur
throughout a phonological system rather than affecting discrete
components of it, as is more apt to be the case in lexical substitu
tion. Moreover, the direction of change is more predictable and
hence more "discoverable" as one moves backward in time; that
is, given a particular phonological feature, the probability is high
that it developed out of one of a limited number of possible ante
cedent features. <I Here, then, is an intriguing problem for some
aspiring phonochronologist.
Further, with regard to diachronic universals, it is evident that
processes of change in language parallel those in culture. Indeed,
some lexical changes are quite simply "mirror effects" of culture
changes as when terms for obsolete items in the cultural inventory
drop out of colloquial speech. Change due to replacement as well
as simple deletion and addition, as these are discussed in Hoenigs
wald's paper, operates in other areas of culture as well as in
language. Ii Methodologically, the problems raised in tracing the
provenience of specimens in an archaeological assemblage are
closely akin to those confronted in historical linguistics when
dealing with the sources of a lexicon at a particular point in time.
From a somewhat different point of view, one may also discern two
universal types of change in both language and culture, cumulative
or noncyclical, and cyclical change. Examples of the latter are not
plentiful, but in the realm of language one might cite Saporta's
(Osgood and Sebeok, 1954) study on the phonemes of Spanish
which suggests that the efficiency of use of distinctive features
oscillates around the 50% point over time. In the sphere of culture,
Kroeber and Richardson's classic study (I940) of periodic fluctua
tions in women's dress fashions is a prime example.
3. Language Universals and Cultural Origins
The origins of culture and the role of language in the emergence
of a cultural mode of adaptation are related questions of abiding
";
of,
t!
286
interest to anthropologists. With his customary perspicacity Sapir

addressed himself to these problems in several telling passages.
On the antiquity of language he writes:
The universality and the diversity of speech leads to a significant
inference. We are forced to believe that language is an im
mensely ancient heritage of the human race, whether or not
all forms of speech are the outgrowth of a single pristine
form. It is doubtful if any other cultural asset of man, be it
the art of drilling for fire or of chipping stone, may lay claim
to a greater age. I am inclined to believe that it antedated even
the lowliest developments of material culture, that these
developments, in fact, were not strictly possible until language,
the tool of significant expression, had itself taken shape
[1921, pp. 22-23J.
Now, if language is very ancient in human history, and if we
succeed in identifying a universal groundwork, hopefully in some
detail, we may ask whether this ancient tongue or tongues might
also be expected to conform to this universal design. In other
words, has this basic design come down to us through the cen
turies, or is it the product of convergent or parallel development?
Lenneberg(1960), in a very interesting paper, has taken the former
view. If the latter, then what was language at its earliest stage like?
These questions are perhaps unanswerable in any final way, yet
they will certainly continue to intrigue anthropologists for years
to come. Sapir, again, has suggested one answer. He writes, "The
fundamental groundwork of language, the development of a clear
cut phonetic system, the specific association of speech elements
with concepts, and the delicate provision for the formal expression
of all manner of relations-all this meets us rigidly perfected and
systematized in every language known to us" (1921, p. 22). And
in another place he has stated that" ... language is an essentially
perfect means of expression and communication among every
known people. Of all aspects of culture, it is a fair guess that
language was the first to receive a highly developed form and that
its essential perfection is a prerequisite to the development of
culture as a whole" (1949, p. 7). Other arguments tend to buttress
287
Sapir's position. Thus, there appears to be little evidence that the

basic design features of language depend on elaboration or com
plexity in other aspects of culture, or that any evolutionary trends
can be discerned in natural languages (Greenberg, 1957, pp. 56
65). Hoenigswald as well remarks in his contribution to this
volume that "these processes [of change by replacement] lead very
rarely, if ever, to alterations of fundamental importance for the
species" (p. 45).
If one subscribes to the general point of view just suggested,
it then follows that what is universal in language may indeed be
very old; and the more precisely we are able to specify what is
universal in language, the greater insight into the pristine form
of language and speech we may hope to achieve.
It is commonplace for anthropologists to assert their belief that
language is a prerequisite for culture, but few go further to specify
in any but the most general terms just what would be required of
language if it is to do the work of culture. We might, then, ask the
cultural anthropologist what his specifications would be for a
language adequate to serve minimal or, perhaps better, basic
cultural needs. Among anthropologists, Hallowell (1955) virtually
alone has offered us some penetrating observations in his brilliant
articles, "Personality Structure and the Evolution of Man" and
"The Self and Its Behavioral Environment." In the latter he
writes, "Ifit be assumed that the functioning of human [his italicsJ
societies depends in some way upon this psychological fact [man's
capacity for self-awareness], it is not difficult to understand why
all human cultures must provide the individual with basic
orientations that are among the necessary conditions for the
development, reinforcement, and effective functioning of self
awareness" (p. 89). Among these basic orientations are self-other
orientation, spatiotemporal orientation, and object orientation.
Hallowell calls attention to the generic function of language in
providing linguistic means of orienting the individual to the cul
turally constituted world he apprehends. Thus, he quotes Boas'
observation (1911) that "the three personal pronouns-I, thou,
and he--occur in all human languages," and that "the underlying
idea of these pronouns is the clear distinction between the self as
r1;
I:
:r ~"
:,
288
speaker, the person or object spoken to, and that spoken of" (1955,
p. 89). In similar fashion, Hallowell calls attention to language
universals other than pronominal systems that serve to orient the
individual in a self-other dimension. Among these are kinship
terms (which we can regard as one subset of a larger set of status
terms) and personal names. To these one might perhaps add the
notion of personal possession however expressed, and terms for
grosser body parts and noncontinuous psychophysiological proc
esses (Le., those capable of being disrupted) that can be brought
into conscious awareness, for example, sleeping, eating, dreaming,
copulating, and listening.
As the second basic orientation, Hallowell notes that if the self
is to be prepared for action, all cultures must provide some kind of
spatiotemporal frame of reference. "Just as personal names medi
ate self-identification and personal reference, in the same way
names for places and significant topographical features are a
universal linguistic means for discriminating and representing
stabilized points in space which enable the self to achieve spatial
orientatkm" (1955, p. 93). Similarly, although the units may of
course vary, temporal intervals must be discriminated. If man by
virtue of his culture is emancipated from the here and now, then he
must be prepared to deal at the symbolic level also with the past and
the future. Pertinent to both types of self-orientation mentioned
here is Weinreich's discussion (Section 2.2.2., p. 154) of the
universality of person, time, and place deixis.
A third universal function of culture is the orientation of the self
to a phenomenological world of objects that are "discriminated,
classified, and conceptualized with respect to attributes which are
culturally constituted and symbolically mediated through lan
guage" (1955, p. 91). And Hallowell adds, "It is this objectifying
function of speech that enables man to live and act in an artic
ulated world of objects that is psychologically incomparable with
that of any other creature." Now if anthropology has taught us
anything, it is that these multifarious culturally constituted worlds of
objects are not semantically equivalent in any simple way, and that
there is no one-to-one correspondence between designata in any
two of these worlds. But neither, as we have also learned, is there
289
complete noncorrespondence. In Section 4 of his paper Weinreich

. has given us some suggestions about how we might deal with the
problem of partial correspondence; the question here, from the
point of view of the anthropologist, is what kind, where, and at
what level might we expect there to be correspondences, and, in
deed, given the observable broad similarities in the nature of man,
his environment, his social life, and his culture, what correspond
ences must there be.
While Hallowell has not, of course, provided a total semantic
design for language, he does suggest some major components for
such a design in a manner that, I believe, goes far beyond mere
speCUlation.
Another related approach that takes us a bit further down the
same road is that of Wallace, wherein he asks what functional
specifications for a human brain might be written based on the
anthropologist's knowledge of the tasks which a cultural mode
of existence requires that brain to perform (1961a, p. 132). Basic
to his argument is the concept of "mazeway," which he defines as
the "meaningfully organized totality of learned cognitive rep
resentations of people, things, processes, and values held at a
given time by an individual" (1961a, p. 139). Within a society (a
culturally organized human group) people act in mutually pre
dictable ways on the basis of semantic equivalences among in
dividual mazeways. A culture, then, may in Wallace's terms be
described as a set of mazeway equivalence structures shared by the
component members of a society. It is important to note that these
mazeways are equivalent, not identical. In fact, Wallace argues that
mazeway identity would be inimical to the development of
cultural complexity. Whereas Wallace assumes that mazeway
equivalence is a necessary condition of cultural behavior, he
flatly asserts that shared motivations are not.
Wallace's discussion is relevant to a consideration of universals
in several respects. First, his general description of the structure
ofa mazeway (l961b, pp. 17-19) can be regarded as a tentative and
schematic statement of cultural universals at the cognitive level,
and following Hallowell, and of course many others, we see
language as of central importance in shaping this widely shared
290
291
~1Iii
II
J
.\,
i'
.I,
.:1
i
I;
"
cognitive structure. Second, while Wallace does not dwell at

length on the role of language both in constituting the mazeway or
as the device par excellence for achieving mazeway equivalence,
he clearly regards it as having a central role in these essential tasks.
Thus he suggests that the fundamental cognitive processes of
perceiving and learning the meaning of stimuli and of relating
these meanings in problem-solving, although differing in cultural
content, may follow constant laws irrespective of culture or even
of species. And he sees componential analysis, "a recent develop
ment of major importance to anthropology," as a powerful
operational tool in revealing the complex semantic structure of the
mazeway.
In his paper, "Cultural Causality and Law: A Trial Formula
tion of the Development of Early Civilization," Steward writes:
"If the more important institutions of culture can be isolated from
their unique setting so as to be typed, classified, and related to
recurring antecedent or functional correlates, it follows that it is
possible to consider the institutions in question as the basic or
constant ones, whereas the features that lend uniqueness are second
aryor variable ones" (his italics) (p. 6). A similar statement might
with at least equal appropriateness be made about language. The
phonology of a language and the phonetic shapes of words, being
but the instruments of meaning and not in themselves meaningful,
are free to vary, and this variety serves in a sense to mask what
may be basic similarities. Similarly, as so beautifully exemplified
in Greenberg's paper in this volume on .grammatical universals,
there are alternative solutions to common grammatical problems
that again mask underlying uniformities. Indeed, although it may
appear to be a contradiction in terms, one might with some justice
speak of "universal alternatives," that is, a limited set of alternative
solutions to a problem, one or more of which may be used in a
particular language.
It appears, then, that there are discoverable universal principles
governing human behavior that lie at a deeper subphenomenal
level. Some might disdainfully comment that this is to rediscover
psychology, and others protest that this is unwarranted reduction
ism. To the latter I would say that we still are left with the largely
historical task of accounting for the particular phenomena of

specific languages and cultures, but I would ask whence come the
explanatory principles in terms of which these accountings are
cast, and in the case of comparative studies, whence come the
categories and concepts that permit valid comparison. To the
former I would say only that this is precisely the direction in which
work in a number of fields of anthropology has been moving.
To use a biological metaphor, the genotypical approach is gaining
a certain ascendancy over the phenotypical approach in many
quarters in both analytic-descriptive and in comparative studies.
Examples of the genotypical approach may be seen in the ap
plication of componential analysis to various aspects of culture,
most notably kinship.6 A similar approach has been used by
Conklin (1955) with respect to color terms in Hanun60, by Frake
(1961) in the analysis of medical concepts, and presumably it
could be extended to other spheres of culture as well. Related
efforts are those of Kluckhohn to handle values in terms of binary
oppositions [an effort in which he was inspired by the work of
Roman Jakobson], of Frake (1962) with respect to settlement
patterns, and of Levi-Strauss in the analysis of myths. And al
though its source is mathematical, one might also cite in this
connection the recent attempt by Davenport (1960) to apply what
in game theory is called minimax utility to the observed behavior
of Jamaican fishermen.
In many of these studies the ties with linguistic methods and
linguistic data are strong, and in reviewing them one has the
sense that here is an approach that promises to be equally pro
ductive in dealing with problems of language universals as it has
proved to be in other spheres of culture. Frake (1962, p. 54) has
clearly described the essence of the approach:
A successful strategy for writing productive ethnographies
must tap the cognitive world of one's informants. It must
discover those features of objects and events which they regard
as significant for defining concepts, formulating propositions,
and making decisions. This conception of an ethnography
requires that the units by which the data of observation are
segmented, ordered, and interrelated be delimited and defined
!'i
292
according to contrasts inherent in the data themselves and

not according to a priori notions of pertinent descriptive
categories.
As we have already suggested at several points in this summary
statement, the further problem then becomes: To what extent
are these features whereby the world of experience is cognitively
constituted universal? This question leads quite naturally to the
final topic that I wish to discuss.
4. Language Universals and Linguistic Relativity

Among the many problems of common interest to students of
human behavior perhaps none stand to gain greater clarification
through systematic work on language universals than those de
riving from the linguistic relativity hypothesis. Although adum
brated in the work of several intellectual forebears, notably
Franz Boas (l911a) and Edward Sapir (1949, e.g., p. 162), the
theory of linguistic relativity has been developed most fully and
persuasively in a series of influential articles by the late Benjamin
Lee Whorf (Carroll, ed., 1956). One cannot hope to improve on
Whorf's own eloquent statement of his position:
Actually, thinking is most mysterious, and by far the greatest
light upon it we have is thrown by the study of language.
This study shows that the forms of a person's thoughts are
controlled by inexorable laws of pattern of which he is uncon
scious. These patterns are the unperceived intricate system
atizations of his own language-shown readily enough by
a candid comparison and contrast with other languages,
especially those of a different linguistic family. His thinking
itself is in a language-in English, in Sanskrit, in Chinese.
And every language is a vast pattern-system, different from
others, in which are culturally ordained the forms and cate
gories by which the personality not only communicates, but
also analyzes nature, notices or neglects types of relationships
and phenomena, channels his reasoning, and builds the house
of his consciousness [po 252].
293
The implications of language universals for such a theory,

patently, are far-reaching.
In dealing with the Whorfian theory, one must distinguish care
fully between two different but related approaches to the study
of the determining influences of language on cognitive processes
and other symbolically mediated behavior. One approach grows
essentially out of work in the psychological tradition, the other
out of an anthropological tradition; these are the fields we have
come to call respectively psycholinguistics and ethnolinguistics.
The former most characteristically deals with the generic function
of language in shaping cognitive processes, while the latter is
typically concerned with the comparative problem of how struc
tural differences among languages, in both their lexical and gram
matical aspects, systematically relate to differences in the cog
nitive processes or other behavior of their speakers. The two
approaches are complementary. In fact, the latter presupposes the
former, but 'they approach the problem at different levels, using
different kinds of linguistic data, and, at least in part, employing
different research strategies. 7 It is of paramount importance, I
feel, in working on these problems to be clear about which of
these approaches is at issue or in what combination they are
involved.
If the linguistic relativity hypothesis is to be put in proper
perspective, it becomes crucially important to specify wherein
languages are alike as well as wherein they differ. Logically, the
very notion of variation assumes knowledge of the base from
which phenomena vary. Indeed, it has been said that the ultimate
task of science is precisely to account for variation.
It may well turn out that what is universal in language functions
much more powerfully, and in a more fundamental way, to shape
men's thoughts than what is different. There is some recent exper
imental work that suggests that this may indeed be so, for example,
by Osgood and his associates. Some of this research has already
been reported (1960), and some, on the cross-cultural and cross
linguistic generality of the basic dimensions of connotative
meaning, is still largely unreported. Wallace is also suggestive on
this point:
'.
294
The Whorfian and other hypotheses of extreme cultural

relativism assert a radical dependence of the very form of
rationality upon the local structure of language. But it seems
more likely that the elemental notions which are the common
base of the various logical and semantic calculi-notions of
'not', of 'and', of 'and/or', of 'identically equal', of 'equivalent',
of 'order', and the like-are symbolic representations of
processes intrinsic to such evidently universal psychic functions
as discrimination, conditioning, and the generalization of
learning. Indeed, a radical linguistic relativism would probably
be, by its own axioms, not only incapable of proof but incapable
of being described [1961a, p. 142J.
295
ogy have had such close relations over the years. However, this
apparent paradox is diminished when one is reminded, as I have
been by Dell Hymes and Thomas Sebeok, of the close links
between linguistics and anthropology on the Continent. De Saus
sure, Durkheim, Mauss, Meillet, and, today, Levi-Strauss, who
self-consciously uses linguistics as a model, all share in this
intellectual tradition. Nevertheless, the influence of linguistics on
British social anthropology is largely derivative; and it is further
remarkable that, except for Malinowski, British social anthropol
ogists have shown so little interest in linguistics. A thoroughgoing
account of the mutual relations between anthropology and
linguistics both in this country and abroad, clearly, would be
most illuminating of the joint history of our two disciplines.
2, In an early statement Kroeber (1916, p. 93) eloquently called atten
tion to this dual task. Because of both its content and historical
interest the passage is worth quoting at length:
The problem for future research on linguistic relativity is to

reconcile the two positions represented above in the quota
tions from Whorf and Wallace.
.., the determination of what they [Algonkin and Indo-Euro

pean] have in common, involving as it does the recognition of
that in which they are different, is an essential purpose of the
study of both; for whether our interest lies in the problem
of the nature or that of the origin of human speech, a classifi
cation is involved. In its widest ultimate aspect philology is
concerned not with Algonkin as such nor with Indo-European
as such but with all languages. Only when speech in general,
its scope and its methods, are better understood will both
Algonkin and Indo-European, or for that matter any particular
group of languages, be more truly understandable. The real
aim of the study of any American tongue, as well as the aim
of any deeper research in Indo-European philology, must
therefore be the more precise and fundamental determination
of their relations to all other languages; and this necessitates
concepts and terms which are applicable in common. It is
impossible to characterize the'wolf in terms of his skeleton, the
elephant of his embryology, the whale of his habits, and then
to construct a classification which will help to reveal the inherent
nature, the development, or the origin of the animal kingdom.
5. Conclusion
If individual men or whole peoples dwelled alone in incommen
surate worlds constituted only by their uniq'ue experiences or by
those shared within the bounds of isolated communities, communi
cation among men or among peoples would be impossible,
Clearly, this is not so, but the matter is still naggingly indeter
minate. We are left, then, with what has long been the essential
question for anthropology as it has been for this symposium: How
do we account for the simultaneous sense we have of the unique
and the universal in our fellow men? What we have done in this
symposium, and future work on language universals, will bring
us a bit closer to the answer. And what knowledge we can gain of
language universals will surely bring the intriguing but recalcitrant
problem of linguistic relativity into sharper focus.
J am indebted to Dell Hymes for calling my attention to this
quotation.
3. These nine are: speech, material traits, art, mythology and scien
tific knowledge, religious practices, family and social system,
property, government, and war.
4. In conversation with my colleague Kenneth L. Hale, I find that he
had independently arrived at similar conclusions.
Notes
I. It is paradoxical that the anthropological approach most closely
approximating the methods of present-day structural linguistics,
that of French-British social anthropology, was developed abroad
rather than in the United States, where linguistics and anthropol-
"II
296
297
Grimes, J. E. and B. F. (1962). "Semantic Distinctions in Huichol

(Uto-Aztecan) Kinship," American Anthropologist 64. 104-114.
Hallowell, A. l. (I955). Culture and Experience. Philadelphia.
Kluckhohn, C. (1953). "Universal Categories of Culture," Anthro
pology Today, A. L. Kroeber (ed.). Chicago.
Kroeber, A. L. (1916). Arapaho Dialects, University of California
Publications in American Archaeology and Ethnology 12.3. 71
138. Berkeley, Calif.
- - - , and Jane Richardson (1940). Three Centuries of Women's
References
Dress Fashions .. A Quantitative Analysis, Anthropological Records
5.2.111-153. Berkeley, Calif.
Boas, F. (1911a). "Introduction," Handbook of American Indian Lan
Lenneberg, Eric H. (1960). "Language, Evolution, and Purposive
guages, Part I, Bureau of American Ethnology, Bul. 40. 1-83.
Behavior," in Culture in History: Essays in Honor of Paul Radin,
- - (1911b). The Mind of Primitive Man. New York.
S. Diamond (00.), 869-893. New York.
Carroll, J. B. (ed.) (1956). Language, Thought, and Reality: Selected
Linton, R. (1952). "Universal Ethical Principles: An Anthropological
Writings of Benjamin Lee Whorf. New York, and Cambridge,
View," Moral Principles of Action, Ruth Anshen (ed.). New York.
Mass.
Lounsbury, F. G. (1956). "A Semantic Analysis of the Pawnee Kinship
Casagrande, J. B. (1948). "Comanche Baby Language," International
Usage," Language 32. 158-194.
Journal of American Linguistics 14. 11-14.
Moore, Harvey c., Jr. (1954). "Cumulation and Cultural Processes,"
Conklin, H. C. (1955). "Hanun60 Color Categories," Southwestern
Journal of Anthropology II. 339-344.
Murdock, G. P. (1945). "The Common Denominators of Culture,"
Davenport, W. (1960). "Jamaican Fishing: A Game-Theory Analysis,"
The Science ofMan in the World Crisis, R. Linton (ed.). New York.
Papers in Caribbean Anthropology. Yale University Publications
- - (1949). Social Structure. New York.
in Anthropology, No. 59.
Du Bruhl, E. L. (1958). Evolution of the Speech Apparatus. Spring
- - - (1959). "Cross-Language Parallels in Parental Kin Terms,"
field, Ill.
Anthropological Linguistics I. 9. 1-5.
Emeneau, M. B. (1943). "Franz Boas as a Linguist," Franz Boas, 1858
Osgood, C. E. (1960). "The Cross~Cultural Generality of Visual-Verbal
1942, American Anthropological Association, Memoir 61.
Synesthetic Tendencies," Behavioral Science 5. 146-169.
Epling, P. J. (1961). "A Note on Njamal Kin-Term Usage," Man 66.
- - - , and T. A. Sebeok (1954). Psycholinguistics, A Survey of Theory
152-159.
and Research Problems. Baltimore.
Fishman, J. A. (1960). "A Systematization of the Whorfian Hypoth
Penfield, Wilder, and L. Roberts (1959). Speech and Brain-Mechanisms.
esis," Behavioral Science 5. 323-339.
Princeton, N. J.
Frake, C. O. (1961). "Diagnosis of Disease Among the Subanun of
Romney, A. K., and P. J. Epling (1958). "A Simplified Model of
Mindanao," American Anthropologist 53. 113-132.
Kariera Kinship," American Anthropologist 60. 59-74.
- - - (1962). "Cultural Ecology and Ethnography," American
Sapir, E. (1921). Language. New York.
Anthropologist 64. 53-59.

- - - (1949). "Language," Selected Writings of Edward Sapir,

D. Mandelbaum (ed.). Berkeley, Calif.
Meaning," Language 32. 195-216.
Spuhler, J. N. (1959). "Somatic Paths to Culture," The Evolution
Greenberg, J. H. (1957). Essays in Linguistics, Viking Fund Pub
of Man's Capacity for Culture, J. N. Spuhler (ed.). Detroit.
lications in Anthropology, No. 24. New York.
5. Moore's (1954) treatment of cultural accumulation, for instance,
raises many of the same questions covered in Hoenigswald's

paper.
6. See Goodenough (1956), Lounsbury (1956), Romney and Epling
(1958), Wallace and Atkins (1960), Epling (1961), and Grimes (1962).
7. For an excellent discussion of various levels of approach to the
linguistic relativity hypothesis see Fishman (1960).
298
Steward. J. H. (1949). "Cultural Causality and Law: A Trial Formula

tion of the Development of Early Civilization," American An
thropologist 51. 1--27.
CHAPTER 11
- - (1961). "Alfred Louis Kroeber, 1876-1960," American An

thropologist 63. 1038-1060.
LANGUAGE UNIVERSALS AND PSYCHOLINGUISTICS
Wallace, A. F. C. (I 96Ia). "The Psychic Unity of Human Groups,"

Studying Personality Cross-Culturally, B. Kaplan (ed.). Evanston,
111., and Elmsford, N. Y.
- - (I 96Ib). Culture and Personality. New York.
- - , and J. Atkins (1960). "The Meaning of Kinship Terms,"
Whiting, J. W. M. (1954). "The Cross-Cultural Method," Handbook
of Social Psychology, G. Lindzey (ed.). Cambridge, Mass.
- - , and L L. Child (1953). Child Training and Personality. New
Haven.
Wissler, C. (19.23). Man and Culture. New York.
CHARLES E. OSGOOD
1. On Linguistics as a Science
At this conference we have been witness to a bloodless revolu
tion. Quietly and without polemics we have seen linguistics taking
a giant step from being merely a method for describing language
to being a full-fledged science of language. Of course, as is true of
any revolution, the step is only "in progress," and the participants
do not see themselves as revolutionary; but in the eyes of a suffi
ciently remote observer, the change can be noted and its signifi
cance recorded.
Linguistics is shifting its concern from the uniquely differential
to the broadly general. All of the papers prepared for this con
ference, and the discussion that completed it, take this revolution
for granted. Twenty, even five, years ago, this could not have
happened-without bloody eyes and heads. Today it happened
quietly. Of course, what is happening in linguistics is also hap
pening in the other sciences of man. There are pervasive swings in
the purposes and viewpoints of the social sciences. While Bloom
field was forming the objectivity and operationalism of descriptive
linguistics, Watson and Weiss were setting the framework for an
objectively descriptive psychology. Ghosts, like mind and meaning,
were laid; methodology became king. Uniquenesses were sought,
and found, and empiricism reigned.
But pendulums have a way of reversing their direction. In the
very process of describing languages and behavior, linguists and
299
300
Charles E. Osgood
psychologists could not deny the evidences of generality that kept

presenting themselves. Beneath the surface of uniqueness, a
bedrock ofcommonness has begun to impress itself upon us again.
It is a commonness that is not arbitrary, and therefore it is inter
esting scientifically. What does being a science imply?
At the lowest level, a science is merely a new way of talking
about familiar phenomena. It is an argot which binds the elite
in a warm embrace and separates them from the hoi polloi. Work
ing within their own discipline and following the rules, the
members of the cult may reach a higher level-of descriptive rigor
and carefully defined constructs. Linguistics did this with a
sureness and aplomb that embarrassed the other social sciences.
Of course, it was assisted by the "chunkiness" of language, and
the aptness of its disciples in creating "chunks" where only amor
phous continua may exist. But there is a higher level of descrip
tion-quantification. Here, I think, the courtship with psychology
over the past decade or more has been having a significant impact.
It is clearly evident in Sol Saporta's paper, for example.
But rigorous, quantitative description of a defined set of phe
nomena still does not represent a full-blown science. There are
two more facets to the whole, and these linguistics undertook here
at the Conference. First, there are general empirical functions
relating construct to construct-the more of this, the more (or
less) of that. Empirical science piles function upon function.
Psychologists have spent fifty years or more striving to make
their empiricism secure. However, an impressive structure of
empirical functions invites interpretation-theory. Without
theory, the massive collection of functions remains that-an
incredible challenge to analysis and simplification. Theory is thus
a higher-level description; it is a set of principles which, econom
ically and elegantly, encompasses the whole set of functions.
Linguistics hasn't arrived, but it now seems well on the way.
What will be the character of the basic principles of linguistic
science? Of course I am biased, and I am well aware of the dangers
in reductionism, but as a psycholinguist I am equally aware of the
fact that language behavior is part of human behavior in general.
Just as psychologists are now again reaching out toward universals
UNIVERSALS AND PSYCHOLlNGUISTICS
301
that cross the barriers of both language and culture, so am I sure

that linguists will find their basic principles in the universals of
humanness. There will be no special theory of language behavior
in psychology, nor should there be in linguistics. Even the hier
archical structure of language is replicated in nonlanguage
behavior. The phonic fingers that grasp the doorknob are a unit
within the morphic act of opening the door, which in turn is a
unit within the larger complex of "going to the store"-a sememe?
By virtue of its rigorous description, linguistics has thrown down
the gauntlet to psychology; it is a challenge for psychology to try
to comprehend and systematize within its principles the neatly
ordered universals of language behavior.
It is precisely at the level of language universals that psycho
linguistics has meaning and that a true science is a-borning. I recall
with deep pleasure the summer seminar at Indiana in 1953, where
linguists and psychologists shared themselves freely both in being
educated and in educating. While the linguists were pressing the
psychologists toward greater rigor in their descriptions of behavior,
the psychologists were pressing the linguists toward greater univer
sality in their conceptions. The psychology of language behavior
can be useful only in terms of universals. Here are the nexuses
between our fields. Where languages commonly use bilabial
nasals for "mamma" terms, where they are commonly more sus
ceptible to change in antecedent rather than subsequent positions,
where they display a regular ebb and flow in the ratio of phonemes
to features-here psychology can find a purchase.
But of what kind are the useful universals of language? As one
who has been exposed to this conference, I must stress a distinction
whose omission will lead us toward dust-bowl empiricism rather
than toward a productive science. There are two kinds of uni
versals:
VI' Phenotypes: empirical generalizations that hold for all
languages. As Hockett pointed out, many of these are useful
in a definitional sense-they would help a Martian linguist
distinguish between human language and organismic com
munication more generally. But the very fact that all human
'il;
"
..
302
Charles E. Osgood
languages display property X, even if it is not definitional,

renders property X mainly interesting, not particularly useful
in science-making.
U2 Genotypes: theoretical generalizations, principles in a
theory oflanguage behavior, that hold for aI/languages. These
will be the guts of a science of linguistics. They are the
fundamental laws governing the production of semantic
regularities, the production of grammatical regularities, the
source of language change.
And here we come to one of the nubs of conflict in our confer

ence. There are some who have clung to the traditional notion that
only the truly universal universals are significant. Traditional? Yes.
I have been a walking pincushion for the barbs of linguists who
have gleefully blunted my probes for psycho linguistic universality
with counter-points of "exceptions." I say that, as a matter of psy
cholinguistic theory, there will be greater diversity among prefixes
than among suffixes in languages. My linguist friend says yes-but
there is language Z, with which he happens to be familiar, where
the reverse holds true. The point is that scientific laws are not
merely honored by their exceptions-they are literally manufac
tured out of their exceptions. Exceptions re~ect interactions among
the whole set of functional laws. Language Z does not display
regularity X because laws I, 2, and 3 operate in one way
under condition alpha and quite differently under condition beta.
In other words, it is the nonuniversal, statistical universals that
are the most interesting to a science of language behavior. Out
of the complex patterning of non universal (but lawful) phenomena
we will draw forth the underlying principles of a science of lan
guage.
However, the search for universal universals-and the failure
to find them (quantitatively described)-is essential to the building
of a science of language. As this conference has made abundantly
clear, there are literally thousands of potential universals-or
quasi universals. The important thing at this point is to expand
the search for universals of language behavior and to systematize
them in the context of psycholinguistic theory. When Bloomfield
UNIVERSALS AND PSYCHOLINGUlSTlCS
303
said "the only useful generalizations about language are inductive

generalizations," he was telling only one part of the story of
science; the other part is the hypothetico-deductive system of
principles in which inductive generalizations are organized into
predictive theory. Thoughtful analysis of the array of quasi
universals, and their interactions, will generate general principles
which can then be tested against the whole set of human communi
cation systems. As both Murdock and Lounsbury implied, the
sample of languages need not be exhaustive, but it must be repre
sentative if these tests are to be adequate.
How should the universals of human language be systematized?
Here I can do no more than suggest that their ordering be relevant
to what we now think are the underlying principles of language
behavior. Of course, these principles will be wrong-if not in
fundamental assumption, then certainly in detail-but this is the
stuff of which science is made. For the present, I think our search
for universals and our presentation of them should be guided by
(1) our general methodology and its constructs, and (2) our
present assumptions about their bases, for example, in what we
know about human perceptual organization, in what we know
about human concept formation, in what we know about human
skills, and in what we know about human meaning systems. So
we look for universalness in the phonemic and morphemic
system, for universalness in rules of transformation, for univer
salness of the semantic map of the world-and in doing so we
certainly miss many potential universals that the linguistic science
of another age would seek and record. But science-making is a
slow and dubious process.
There are, however, some fundamental divisions of the plot
that we can make a priori grounds. We should make a distinction
between semantic and structural universals (I persist with this
word, but I am referring to regularities). The former will probably
relate to the psychology of symbolic processes and the latter to
skill and integrational mechanisms. We should also distinguish
between synchronic and diachronic universals. Again, the psycho
logical principles governing language patterning or structure will
probably not be entirely the same as those governing language
304
Charles E. Osgood
change-which is primarily a learning process. And the funda

mental methodological distinctions between phonological, mor
phological, and syntactical levels of language must not be for
gotten. This gives us a 2 x 2 x 3 framework within which to
pattern our search for universal laws of language behavior
synchronic/diachronic; semantic/structural; and phonological!
morphological/syntactic-a neat space with twelve regions within
which to allocate our universal (or quasi-universal) phenomena.
Without trying in every case to specify the Conference papers
which gave rise to these notions, but reacting to their sum, let me
now suggest four fundamental psycholinguistic generalities that,
on the most naive and unresearched level, seem to offer starting
points for a science of language behavior.
I. At all levels (phonological, morphological, and syntactic),
language systems like other behavioral systems will follow a
principle of progressive differentiation in their development. This
underlying law will be reflected in a large number of empirical
universals of the following general type: if a language displays
characteristic c, then it will also display characteristics b and a;
and if it has b, it will also have a; but not vice versa. Many years
ago, Roman Jakobson pointed to such unidirectional depend
encies in the development of phonemic distinctions in child
language; at the Conference we have added Ferguson's analysis
of the development of nasals and Greenberg's analysis of the
unidirectional dependency of the dual upon the prior differentiation
of singular and plural, for example. It may be noted in passing
that the same principle of progressive differentiation has long
been a major operating notion in developmental psychology, as
in the development of motor skills.
2. At all levels of units in a language, the competing alternatives
will be organized hierarchically in terms offrequency ofoccurrence
and with a relatively low-entropy distribution approximating the
Zipf function. Greenberg's data on the relative frequencies of
sentence-types, on the relative frequencies of suffixing versus

prefixing versus infixing, and so forth, based on a sample of
thirty languages provide the beginnings for such an analysis at the
305
syntactical level. Saporta's present data on the relative frequencies

of alternative morpheme lengths in Spanish and his previous data
on the relative frequencies of consonantal clusters in English
provide suggestive beginnings at the phonological level. In some
research of my own on the cross-linguistic and cross-cultural
generality of meaning systems (to which I will return later), we
have found rather remarkable stability of the slope of the fre
quency-diversity functions for qualifiers elicited by standard as
sociational procedures in some nine language/culture communities.
All of these "universals" suggest that when the human organism
must choose repeatedly among a set of alternatives, it tends to
choose a few alternatives with very high frequencies and many
alternatives only rarely. And this pervasive law is not restricted
to humans, either; rats tested in a "checkerboard" maze where all
alternative routes are equally long and equally rewarded display
the same lopsided distributions of preference. Living organisms
seem to be antientropic in nature.
3. At all levels of language organization, whenever there are
competing means of achieving some criterion of communication
performance, these competing means will be related inversely as a
compensating system. In other words, if both characteristics x and
yare alternative means of achieving the same communicative

end, as x increases y will decrease, both in one language over time
and across many languages synchronically. We have seen sug
gestive (if not sufficient) evidence for several compensating
systems of this sort. In maintaining an optimum balance between
the discriminatory capacities of speakers and of hearers, the ratio
of number of phonemes to number of distinctive features (and
hence the maximum possible number of phonemes) seems to
oscillate about an efficiency value of 50%; if a code becomes too
tightly efficient, speakers will begin to add in redundant features
in order to be understood, but as this process inevitably overshoots
its mark, and the code becomes too redundant, speakers will
begin to drop out certain features and "get away with it"-and
the cycle begins again. Other compensating systems have been
mentioned here: Inflection seems to compensate with word-order
as an alternative means of keeping the syntactical house in order;
.~
306
Charles E. Osgood
size of phonemic inventory seems to bear a compensatory relation

to average length of morphemes. Relations of this kind can be
studied either synchronically (across representative samples of
the world's languages) or diachronically (across historical time
samples of a single language).
4. At all levels, the laws of language change (diachronic uni
versals) will be found to reside in the principles and conditions of
learning as they operate upon individual speakers and hearers. Given
the whole tenor of our conference, this may seem rather obvious,
but I think it needs stress. For one thing, it leads us to look more
closely into the diachronic aspects of the language behavior of
iOdividuals for our general principles of language change-the
child learning his first language, the adult learning a second lan
guage, the aphasic losing and then recovering language, the effects
of fatigue, alcohol, and various drugs upon language performance.
Although the structure of a language as it exists at a particular
moment in time provides the conditions for learning (points of
discrimination difficulty, loci of competing divergent hierarchies as
opposed to loci of facilitating convergent hierarchies, regions of
overabundance or underabundance of the lexicon, for example),
it is the performance of hundreds of thousands of individual
speakers and hearers under these common conditions that gener
ates diachronic change. The relation of the efficiency of the
phonemic code (already mentioned) to phenomena of language
learning in children should be particularly interesting in this
respect.
2. On Semantic Universality
All of the preceding was an attempt to write down the essence
of what I said informally, and probably better, on the last day of
the Conference. Now I would like to say something about research
on the generality of affective meaning systems that is presently in
progress at The Institute of Communications Research at Illinois.
This can be viewed as a footnote to Stephen Ullmann's paper on
"Semantic Universals."
Over the past decade we have made a dozen or more factor
analytic studies of the structure of meaningful judgments of
UNIVERSALS AND PSYCHOLINGUISTICS
3Q7
American English-speaking subjects, and we have kept finding the

same three dominant factors or dimensions: an Evaluative Factor
(represented by scales like good-bad, pleasant-unpleasant, and
positive-negative), a Potency Factor (represented by scales like
strong-weak, heavy-light, and hard-soft), and an Activity Factor
(represented by scales like fast-slow, active-passive, and excitablecalm).
The problem before us now is this: how common is this se
mantic framework across the range of people doing the judging
and across different kinds of concepts being judged? Is it limited
to Americans speaking the English language, or is it shared by
all humans regardless of their language or their culture? Is it the
same for all concepts, be they aesthetic or political, familiar or
unfamiliar, words or pictures? Let me anticipate our conclusion
from a large number of studies-we find that the evaluation
potency-activity system is remarkably stable across people but
quite unstable across the concepts being judged. We will want to
inquire into why this should be so. But first, some evidence.
When a group of people judge a set of concepts against a set
of scales, representing what we call "a semantic differential," a
cube of data is generated. Each cell in this cube represents how a
particular person judged a particular concept against a particular
scale, using a number from I to 7. For example, in one cell we
might have a number 7, this being one man's judgment of the
concept TORNADO against a slow-fast scale (indicating that he
thinks of TORNADO as extremely fast). In the next cell down in the
cube we find a number 4, this being his judgment of TORNADO in
terms of honest-dishonest (the number 4 showing that he feels
neither one way nor the other on this scale). Each person, as a
subject, is a slice of this cube from front to back; each concept
being judged, like TORNADO or MY MOTHER, is a slice of the cube
from left to right; and each semantic scale is a horizontal slice or
row from top to bottom. In analyzing these data we usually are
interested in the correlations between scales-that is, in deter
mining how the semantic dimensions cluster together-but we
can rule these correlations either across subjects or across con
cepts, and we can do it either for all subjects or concepts as a
r;'1
illi'
308
UNIVERSALS AND PSYCHOLlNGUISTlCS
Charles E. Osgood
309
i,I',\1
ii;!1
II
group or for individual subjects or concepts. In other words, there

are many ways we can slice our semantic cake, and each method
of slicing serves to answer a different kind of question.
The most critical test for generality of these semantic factors
clearly would be between people differing widely in both language
and culture. We have already made a number of cross-cultural
comparisons-involving Japanese, Koreans, Greeks, and Navajo,
Zuni, and Hopi Indians in the American Southwest-and the
similarities in factor structure have been striking. But, for the
most part, these studies have involved simply translating English
scales into the various languages, and we are open to the criticism
that we have forced people of other countries to operate within the
limits imposed by an American English factor system. However,
we do have one study done completely independently by the Mar
keting Center Company in Tokyo in which the same general fac
tors appeared, and we are now in the middle of a large-scale cross
cultural study, involving some twelve countries and as ideal testing
conditions as we can devise.
With the help of cooperating social scientists in each country
without whom this type of research could not be done-we are
collecting data in Japan, Hong Kong, India (Hindi in New Delhi
and Kannada in Mysore), Afghanistan, Iran, Lebanon, Yugo
slavia, Poland, Finland, Holland, Belgium, and France,along with
the United States as a comparison base. We start with a list of 100
familiar concepts that have been selected by linguists and anthro
pologists as being "culture-fair" and that have survived a stringent
back-translation test with bilinguals for all of the six language
families represented. This is the only point where translation is
involved and could influence the results. From this point on,
everything is done in the native language and with native mono
lingual subjects in each country.
The first step is to have 100 young high-school boys in each
country give the first qualifiers (adjectives in English) that occur
to them when each of the concepts is given as a stimulus-for
example, to the word TREE one boy might say tall, another green,
another big, and so forth. This basketful of 10, 000 qualifiers
(l00 subjects times 100 concepts) is shipped to the University of
Illinois, where, using IBM and ILLIAC high-speed computers, we

determine a rank order of these ways of qualifying, in terms of
total frequency of usage, diversity of usage across the 100 con
cepts, and independence of usage with respect to each other. We
already have these rank-frequency-and-diversity measures for
nine countries; they are not only highly similar in statistical
properties, but when the ranked qualifiers are translated into
English and then correlated with both English and each other, the
correlations are all significantly positive. In other words, the
dominant ways of qualifying experience, of describing aspects of
objects and events, tend to be very similar, regardless of what
language one uses or what culture one happens to have grown up in.
The second step in each country is to take the 50 highest-ranking
qualifiers, elicit their common opposites so as to make scales like
good-bad and big-little out of them, and then have a new group of
100 young men judge each such scale against every other of the
50 scales-to what extent is good either big or little, to what extent
is big either happy or sad, and so on? This new basketful of data
is shipped back to Illinois, where we do the correlations and factor
analyses that represent our first test of the structure of the semantic
space. For the six countries that have been carried to this stage, I
can report that the first two factors are definitely as expected
Evaluation and Potency; the third factor is more variable across
countries, but seems to have at least the "flavor" of Activity
(semantic properties like hot, fast, young, and noisy keep ap
pearing). We hope that the third step will clarify this situation.
Here we will have yet another group of similar subjects judge the
original 100 culture-fair concepts against the 50 semantic scales,
correlate each scale with every other as used in actually judging
concepts, and do another factor analysis. We are only beginning
this phase of the research.
What is the purpose of all this busywork in many lands and
many tongues? The first, purely scientific purpose is to dem
onstrate that human beings the world over, no matter what their
language or culture, do share a common meaning system, do
organize experience along similar symbolic dimensions. A second,
more practical purpose of this research is to develop and apply
310
Charles E. Osgood
instruments for measuring "subjective culture"-meanings, atti

tudes, values, and the like-instruments that can be shown to be
comparable across differences in both language and culture. The
demonstration of common semantic factors--if indeed they can
be demonstrated-makes it quite feasible to construct efficient
"semantic differentials" for measuring the meanings of critical
concepts cross-culturally, with reasonable confidence that the
yardstick is something better than a rubber-band. Ultimately, it
would be my hope that both the demonstration of a shared seman
tic framework and the application of semantic measuring instru
ments would contribute to better international communication and
understanding.
Now, let us flip the coin over and ask about the generality of
semantic factor structures across the concepts being judged. You
will recall that the cube of data generated when a group of subjects
judges a sample of concepts against a set of scales makes it
feasible to compute separate correlation matrices for each concept
"slice" and factorize such matrices. In what we refer to as our
"Thesaurus Study"-because the adjectives were sampled from
that source on a rational, representative basis- 20 different con
cepts, like FOREIGNER, KNIFE, MODERN ART,.DEBATE, and HOSPITAL,
were judged against 76 scales by 100 college subjects.
Now, imagine the 20 separate correlation matrices for the dif
ferent concepts lined up as a deck; if we go through the deck at
the point of intersection of a particular pair of scales (for example,
of sober-drunk vs. mature-youthful), we will isolate 20 r's, one for
each concept. If scale relations were reasonably constant over
concepts, then we would expect only minor variations within
such rows of correlations-but this proved not to be the case.
Correlations were found to vary as much as from +.60 to -.60
in the same row. A couple of examples will serve to suggest what
is happening: sober goes with youthfUl for the concept DAWN, but
sober goes with mature for the concept UNITED NATIONS; pleas
urable goes withfeminine for the concept MOTHER, but pleasurable
goes with masculine for the concept ADLAI STEVENSON. It would
appear that the nature of the concept being judged exercises a
restriction on scale meanings. What about the correspondence of
311
factors derived from such single-concept matrices? Here the

picture is better: something identifiable as an Evaluative factor
appeared for each concept, and it was usually the first in order of
magnitude; something identifiable as a Potency (or Dynamism)
factor appeared for all but two concepts; but other factors
varied in most inconsistent ways.
This instability of scale relations and factors across concepts
contrasts sharply with the stability we have found across people.
This shows up most clearly in studies where both types of gener
ality can be compared. In one experiment designed specifically
to get at this problem, college girls in both Japan and the United
States judged three different classes of concepts-patches of color,
simple line drawings, and abstract words like LOVE and PEACE
against a 35-scale form of translation-equivalent semantic dif
ferential. Separate scale-by-scale correlation matrices and factor
analyses were run for each of the six combinations of two subject
groups and three concept-classes. Now, if our hypothesis-that
semantic systems are more stable across people than across
concepts-holds, then factorial similarities should be higher when
Japanese and American girls judge the same materials (both
judge colors, both judge forms, and so forth) than when the same
group judges different materials. This was true in every case. Even
the salience of the three major factors shifted in the same ways for
both Japanese and Americans-Activity is the dominant factor in
judging colors for both groups, Potency tends to be the dominant
factor in judging line forms, and Evaluation is clearly the domi
nant factor in judging abstract words, for both groups.
So much for evidence. Now let me speculate a bit on the
"why" of these observations. Why do we find such wide generality
of the evaluation-potency-activity framework across people,
both within and between languages and cultures? And why, given
generality across people, do we find such lack of generality in
semantic structure across the classes of concepts they judge?
While I certainly don't know the answers to these questions,
I do have some hunches.
First, I must confess that, when we began this research over
ten years ago, I had expected the major dimensions of the seman
I
\\
1li
IK'
312
Charles E. Osgood
tic space to reflect the ways in which our sensory apparatus

divides up the world. This was in flat contradiction to my own
behavioristic theory of meaning, in which the semantic compo
nents should be responselike in character. The accumulating
facts have proved my expectation wrong and my theory at least
"righter"-the dominant factors of Evaluation, Potency, and
Activity that keep reappearing certainly do seem to have a
responselike character, seemingly reflecting the ways we can
react to meaningful events rather than the ways we receive them.
But these major factors also seem to have an affective as well
as a responselike character. The similarity of our major factors
to Wundt's tridimensional theory of feeling-pleasantness, strain,
and excitement-has been pointed out to me. And, as a matter
of fact, we have done a number oJ experiments on the meanings
of facial expressions-coming out with Pleasantness, Control,
and Activation as three factors which seem pretty much to
exhaust the semantic space offacial communication. The similarity
between these factors in emotional communication and those
found in our more general linguistic studies suggests that the
latter may also have their grounding in the affective reaction
system.
Let me speculate a bit further and suggest that the highly
generalized nature of the affective reaction system-the fact
that it is independent of any particular sensory modality and
yet participates with all of them-is at once the reason why
Evaluation, Potency, and Activity appear as dominant factors
and the psychological basis for metaphor and synesthesia. It
is because such diverse sensory experiences as a whUe circle
(rather than black), a straight line (rather than crooked), a
rising melody (rather than a falling one), a sweet taste (rather
than a sour one), a caressing touch (rather than an irritating
scratch)-it is because all these diverse experiences can share
a common affective meaning that one easily and lawfully trans
lates from one sensory modality into another in synesthesia and
metaphor. This is also the basis for the high interscale correlations
which mathematically determine the nature and orientation of
general factors. In other words, the "common market in meaning"
UNIVERSALS AND PSYCHOLINGUISTlCS
313
seems to be based firmly in the biological systems of emotional

and purposive behavior that all humans share.
The evidence for a shared framework of affective meanings
obviously has implications for the types of semantic universals
discussed by Ullmann in connection with metaphor and synes
thesia. I would go so far as to suggest that in this shared frame
work we have the very breeding ground of metaphor. One other
experiment of a somewhat different nature is so relevant to the
problem of universals in metaphor and synesthesia that I will
describe it briefly,
Anglo, Navajo, Mexican-Spanish, and Japanese subjects were
shown in serial order a set of thirteen cards; each card was
divided in half by a vertical line, and on each side of the line
was a simple line drawing. The paired drawings on each card
differed from each other in only one visual property, for example,
blunt vs. sharp, thick vs. thin, and large vs. small. These words
merely describe how the drawings differed; the cards themselves
contained no words. A subject would be given a term in his own
language, such as "happy," and then he would go through the
deck pointing to the visual stimulus of each pair that seemed
"to go best," or "be most appropriate to" this word. Thus if
I were doing it for "happy," I would point to the multicolored
rather than the black-and-white, the upward arrow rather than
the downward arrow, the white circle rather than the black, and
so forth.
The first observation worth noting is the high degree of agree
ment over subjects within each language-culture group: approx
imately 50 % of the 364 items (28 verbal concepts judged against
13 visual alternatives) showed intracultural agreements signifi
cant at the I % level. What about cross-cultural agreement in
visual-verbal synesthesia? The fairest test is to take just those
items where both groups being compared show significant
intra-cultural agreement and ask what percentage of them agree
in direction of choice, that is, show inter-cultural agreement.
Applying this test, we find that Navajo and Anglo groups agree
on the direction of 87% of such items, and all other group com
parisons yield agreements above 90%, Japanese with Americans
314
Charles E. Osgood
315
;\"
'"I',,'
\~i
being 99% and Americans with Mexican-Spanish being 100%,

for example. All of these proportions are significant at well
beyond the .001 level. In other words, we have evidence for
a high degree of universality in what visual alternatives are
perceived as synesthetically appropriate to translation-equivalent
word meanings.
Now, if for the word "happy" a subject points to the colored
one, the up one, the white one, and so on, and if for him the
word "sad" is functionally opposite in meaning, then for the word
"sad" he should point to the uncolored one, the down one, the
black one, and so on-producing, thereby, a negative correlation
between the responses to the two words. We found that verbal
concepts treated functionally as opposites in English are treated
in the same manner by Navajos and Mexican-Spanish-with
the one glaring exception offast-slow for the Navajo (my anthro
pologist friends tell me that the Navajo conceive of fast and slow
as both being aspects of motion as opposed to motionless). The
Japanese data correspond to the American oppositions in every
respect. We have, then, rather impressive evidence for cross
linguistic sharing of visual-verbal synesthetic tendencies.
Finally, what about the lack of generality of semantic factor
structure across concept-classes? All of the evidence we have
clearly indicates that there is interaction between concepts and
scales in the process of semantic judgment. What are the im
plications of this? For one thing, this mearis that from the stand
point of applied semantic measurement there can be no such
thing as "The Semantic Differential." So, for particular concept
classes, we will need to construct appropriate differentials, and
in the area of personality measurement we have already made a
start. From the standpoint of psycholinguistic theory, the fact
of concept/scale interaction invites fresh speculation about how
it operates-and therefore a host of new experiments, most of
which are hardly to the point of conception..
If I were to ask you the question "Is a ,BABY large or small?"
you would undoubtedly say "small." And if I were to ask you
"Is a railroad SPIKE large or small?" you' would probably say
"large." After all, within the class of human organisms, a BABY
is "a small one," and within the class of nails a
SPIKE
~;
is "a large
one." I think the semantic differential technique, in which a

single stimulus is judged successively against a series of different
scales, is one which tends to draw out these intraclass connota
tions of signs. In all other psychophysical methods with which
I am familiar, even the so-called "absolute jUdgment" method,
many different stimuli are judged successively against a single
scale-for example, in judging weights or in scaling the loudness
of tones. I think that this method tends to draw out the denotative
meanings of signs. Note that if I ask you to compare BABY and
SPIKE in terms of size, you immediately say that BABY is "larger,"
now disregarding the intraclass connotations of these objects.
What has all this to do with concept/scale interaction 1 I think
that the semantic differential is subject to what might be called
denotative contamination. The terms that define our scales have
variable denotative meanings as well as their generalized affective
connotation. The denotation of masculine-feminine is brought out
by the concept ADLAI STEVENSON, whereas its potency connotation
is elicited by the concept DYNAMO; a concept like LAVA taps the
denotation of hot-cold, whereas concepts like JAZZ and FESTIVAL
call forth the general connotation of hot. It is clear that if certain
scales are denotatively relevant to certain classes of concepts,
they will fall away from their usual affective factors and hence
change the totaL structure. We have just begun a series of experi
ments comparing the two basic judgmental methods-one concept
at a time against many scales versus many concepts against a
single scale at a time-and I am hopeful that these experiments
will help us disentangle denotative and connotative aspects of
meaning.
Another probable source of concept/scale interaction is what
we call cognitive interaction. This is the tendency for two simul
taneously evoked meanings to change each other in the direction
of compromise-presumably because the affective system can only
assume one "posture" at a time. In making semantic-differential
judgments, one first looks at, and then "keeps in mind," the
concept being judged, while he goes down the page placing
appropriate check-marks on each scale. If he is jUdging the
1i~
,1.11
I,
i:i,;
"1
316
Charles E. Osgood
concept MOTHER, for example, and it has an intensely positive

evaluation, this meaning should interact with those of the scale
terms and cause them to become momentarily more evaluative.
In mathematical terms, this means a general rotation of scales
toward the dominant evaluative factor. In other words, each
concept or concept-class will tend to produce rotation of scales
toward its own characteristic attribute in the semantic space.
We now have some experiments on the planning boards in
which we will try to predict the rotations of scales in the factor
space from knowing the measured meanings of the concepts and
the scale terms that are interacting.
3. On an Absent but Very Lively Ghost
Considering the topic of this Conference, surprisingly little
was said about Benjamin Lee Whorf and his works. After all,
it was he who gave the relativity-universality pendulum its
strongest push in one direction and got it into a position where
we could push it back the other way. I am sure that had he been
alive today, Whorf would have participated in this Conference
and enjoyed it. And I think he would have agreed with much of
what we all had to say.
But I think that early in the game he would have insisted on a
distinction between three types of universals. First, there are
linguistic universals (as well as uniquenesses). Just as the crude
criterion of mutual translatability testifies to some universality,
so does the equally crude criterion of mutual unintelligibility
testify to some uniqueness. The success of descriptive linguistics
depends on the fact that all languages display a hierarchical
structure of units within units, yet the selection of units at each
level (e.g., phonemes from the common phonetic stock) seems
to be quite arbitrary. Similarly, although the lexicons of lan
guages are largely arbitrary, in the sense of correlations between
noises and events, what is talked about and how it is talked
about do not seem to be arbitrary at all. The second type is
psychological universals (and uniquenesses). The principles of
learning seem to be quite universal, but certainly what must be
317
learned with each language and in each culture is quite different.

Similarly, the dimensionality of emotional feeling and expression
seems to be universal, yet how one feels about rain, mother, dog,
or God is highly variable.
However, Whorf was not interested in either linguistic or
psychological universalities or uniquenesses per se, but rather
a third alternative. The hypothesis to which we associate his
name is psycholinguistic in nature. It concerns itself with relations
between linguistic and psychological (cognitive) processes.
Drawing on his extensive comparative studies of SAE (Standard
Average European) and American Indian languages, he hypoth
esized that how a person perceives the world about him, how
he thinks, and even how he formulates his natural philosophy
thus, his Weltanschauung-must depend upon the language he
uses.
I believe that the data I have briefly summarized on the general
ity of affective meaning systems and on the sharing of certain
metaphorical and synesthetic tendencies are contra-Whorfian
in nature. They say, in effect, that despite gross differences in
both language and culture, the underlying structure of the affective
meaning space and the metaphorical translations it facilitates
remain pretty much the same. And this is a strictly psycho
linguistic area, relating linguistic phenomena to psychological
phenomena. But I think there is also sufficient positive evidence
to support Whorf's thesis as well-so it looks as though we have
a dilemma on our hands.
There are several difficulties with Whorf's own materials
as far as scientific evidence is concerned. In the first place, it is
largely anecdotal; lacking the customary controls of scientific
experimentation, his examples must stand as hypothesis-setting
rather than as hypothesis-testing demonstrations. Secondly, the
anecdotes usually hinge on literal translation from some other
language (usually an Indian language) into SAE (usually English),
with comments then being made upon the strangeness of the
world view apparently expressed. Reversing this procedure serves
to lay bare its pitfalls: Suppose a Hopi linguist were to literally
translate the English term breakfast as "the termination of a
\'
318
Charles E. Osgood
period of religious abstinence"; he might then conclude that

contemporary English speakers must think of each night's sleep
as a religious experience, because each morning's meal serves to
break a fast! Thirdly, although the hypothesis is clearly psycho
linguistic in nature, Whorf's analyses are usually restricted to
the linguistic component-he rarely makes any independent
observations on the cognitive processes of language users.
One example from his many writings must serve to illustrate
these difficulties. In his "Languages and Logic" (Technology
Review, M. LT., 1941), Whorf describes how Apache would
refer to the same physical event that EIlglish speakers would
call "a dripping spring." "Apache erects the statement on a verb
ga: 'be white (including clear, uncolored, and so on)'. With a
prefix no- the meaning of downward motion enters: 'whiteness
moves downward'. Then to, meaning both 'water' and 'spring', is
prefixed. The result corresponds to our 'dripping spring', but
synthetically it is: 'as water, or springs, whiteness moves down
ward.''' And then Whorf concludes, "How utterly unlike our
way of thinking!" But nothing about Apache thinking has actually
been observed. To claim that Apache speakers must perceive
a waterfall differently than we because their way of talking about
it is different, and then infer how they perceive solely from how
they talk, is completely tautological. Unfortunately, much of the
"evidence" for or against Whorf's hypothesis has remained at
this level.
On the other hand, there is at least some adequately designed
experimental evidence. I shall cite two of the clearest studies
available. The first deals with the color lexicon and comes from
a series of papers by Brown and Lenneberg, and by Lenneberg
himself. The color spectrum is an ideal aspect of the environ
ment to study psycholinguistically because it is continuous in
human experience, whereas color terminologies are categorical.
First of all, on the intracultural level, it was shown that sec
tions of the spectrum (i.e., particular color chips) vary in their
codabi/ity for English-speaking subjects; whereas a chip near
680 mu might be consistently called "red" with short latency,
a chip near 600 mu might be labeled slowly, variably, and with
319
complex circumlocutions. The psycholinguistic question was

this: Do differences in cod ability relate to some independently
measurable cognitive process? First recognition and then ease
of learning were studied, and in both cases it was shown that ease
of cognitive manipulation varied with codability of the color
stimuli. Then, cross-culturally, it was shown that Zuni speakers
display the same relation between codability and cognitive
facility, even though the relative cod abilities of various portions
of the spectrum did not correspond for Zuni and English speakers
-that is, their denotative assignments were arbitrarily different.
The second experimental study is reported by Casagrande
and deals with the influence of grammatical structure upon
cognition. It is obligatory in Navajo to signal the shape of objects
being dealt with by an appropriate affix on the verb. The relativity
hypothesis requires that Navajo speakers, as compared with
speakers of a language not making this distinction, should be
more responsive to the shape of objects in their cognitive activities.
Casagrande first compared Navajo-dominant children of varying
ages with English-dominant children of the same racial and
cultural background on an object-sorting test; as predicted, the
Navajo-dominant subjects were shown to rely more on shape
similarities and differences, and this dependence increased with
age. Although a group of Harlem schoolchildren, matched with
the previous subjects for age and sex, behaved like the English
dominant Navajos, a group of white middle-class children in
Boston behaved more like the Navajo-dominant Navajos
a result which points to the complexity of factors, cultural as
well as linguistic, that may be operating on cognitive processes..
More evidence could be adduced, but I think this is sufficient.
We thus have the apparent paradox of experimental evidence sup
porting both the hypothesis of psycholinguistic universality and
the hypothesis of psycholinguistic relativity. The situation
remains paradoxical, however, only as long as we fail to discern
any basis for distinguishing the phenomena fitting one hypothesis
from those supporting the other.
Let me refer, rather arbitrarily, to the aspect of "meaning"
tapped by the major factors of the semantic differential as the
320
Charles E. Osgood
connotative meaning of signs. There is another aspect of "meaning"

which deals with the elaborate sets of essentially arbitrary
correlations between linguistic and non linguistic events-for
example, between the noise "apple" in English and the visual
perception of APPLE object-and this r shall refer to as denotative
meaning. This is one of the meanings of "meaning" with which
linguists and philosophers have been primarily concerned. You
may question the propriety of my choice of terms here, but
the thing I am concerned with is that we agree that there is a
significant distinction within semantic phenomena-between the
affective reactions to signs and their coding functions.
Now, I assume that connotative aspects of meaning are
mediated by the relatively "primitive" affective nervous system.
Indeed, the fact that in a number of experimental studies on the
emotional meanings of facial expressions we have found essen
tially the same dominant factors (Pleasantness, Control, and
Activation) as in our more general linguistic studies supports
this identification. I further assume that the innumerable ar
bitrary correlations between linguistic and nonlinguistic events
that I refer to as denotative meaning are mediated by the sensory
and motor discrimination systems of the "new" brain-regions
where lesions may produce various aphasic syndromes. (In this
connection, it is at least suggestive that in a few tests of visual
verbal synesthesia with aphasic patients they showed little or no
impairment, even though they could not even verbalize the
differences between the visual stimuli on the cards.) Both of these
biological systems-the affective, energizing system and the
sensorimotor discrimination system-are integrated in ordinary
behavior, and there is no reason to suppose things should be
different in language behavior. As a matter of fact, I suspect that
interaction between these systems is the basis for the concepti
scale interactions we have found in semantic measurement.
My suggested resolution of the paradox goes like this: When
ever the psycho linguistic phenomena in question depend upon the
structure of the mediating systems (either affective or discrimin
atory), psycho linguistic universality will be found. This, of course,
is precisely because these mediating systems are panhuman
321
biologically. Whenever the psycho linguistic phenomena are inde

pendent of the structure of the mediating systems, even though
they are mediated by them, psycholinguistic relativity will be the
rule. And this, of course, is precisely because both mediating
systems are essentially tabula rasa, and "what leads to what"
is dependent on experience. A few illustrations may make what
appears just a little self-evident-if not downright tautological
more meaningful and useful.
First, as far as the affective mediating system is concerned,
our data show that it is the factorial structure-the basic dimen
sions along which feeling-tones are differentiated-that is im
mutable and overrides differences in both language and culture.
Phenomena which depend upon this shared structure display
universality. Thus, since good, sweet, bright, white, up, smooth,
and the like share positive affect, they will tend to appear as
metaphorical and synesthetic equivalents all over the world. On
the other hand, since the affective meanings of particular con
cepts, like MOTHER, COMMUNISM, SNAKE, and RICE will depend
upon the affective learning experiences of individuals and hence
upon their cultures, we can expect psycho linguistic relativity
(arbitrariness, uniqueness). The primary application ofthe semantic
differential is to measure such differences in the affective meanings
of concepts-within a constant judgmentalframework.
Second, as far as the discriminatory mediating systems are
concerned, we may anticipate universalities when their structural
properties are involved. Thus because the general laws of per
ceptual grouping and patterning apply to all humans, we may con
fidently anticipate universal tendencies to have more discrimina
tive labels for finger vs. hand and for hand vs. arm than for
upper-arm vs. lower-arm or for chest vs. abdomen (i.e., independent
movement is one criterion for perceptual organization). Similarly,
since we are all primates with relatively more discriminative
visual and auditory brains than for the other modalities, we
might have anticipated Ullmann's generalization-that visual
and auditory metaphors for touch, taste, smell, and thermal
experiences will be more frequent than the reverse direction of
translation. On the other hand, since the mapping of nonlin
322
Charles E. Osgood
guistic events into linguistic codes is essentially arbitrary

using the sensorimotor discrimination system but independent
of its structure-we would expect to find psycholinguistic rela
tivity. And, as the Brown and Lenneberg and Casagrande experi
ments have shown, we do.
Finally, by way of analogy, may I suggest that the different
arbitrary conventions of language codes are like the one twelfth
of icebergs above the surface of the water-highly visible but not
correspondingly significant. Beneath the. surface lie the common
potentials for developing languages, the shared systems of sym
bolic representation, the universal mechanisms for metaphor
and synesthesia-all formed in the interaction of human biology
and psychology with a fundamentally common environment.
It is in the eleven twelfths of the ling.uistic icebergs below the
surface that we must search for the general principles of a science
of language.
BIBLIOGRAPHY
Aginsky. B. W. and E. G. (1948). "The Importance of Language

Universals," Word 4. 168-172.
Bally, C. (1944). Linguistique generate et tinguistiquefranfaise. 2nd ed.,
Berne.
Bazell, C. E. (1949). "Syntactic Relations and Linguistic Typology,"
Cahiers Ferdinand de Saussure 8. 5-20.
___ (1954). "The Choice of Criteria in Structural Linguistics,"

Word 10. 126--135.
Br~ndal, V. (1928). Ordklasserne. Copenhagen.
- - - (1943). Essais de linguistique generate. Copenhagen.
___ (1948). French summary of the preceding: Les parties du
discours (Partes orationis). Etude sur les categories Iinguisfiques.
Copenhagen.
Brown, R., and A. Gilman (1960). "The Pronouns of Solidarity and
Power," in T. Sebeok (ed.), Style in Language, 253-276. New York
and Cambridge, Mass.
'- Biijililer, K. (1934). Sprachtheorie .. die Darstellungsfunktion der
Sprache. Jena.
Carroll, J. B. (1958). "The Assessment of Phonetic Cluster Fre
quencies," Language 34. 267-278.
Cherry, C., M. Halle. and R. Jakobson (1953). "Toward the Logical
Description of Languages in Their Phonemic Aspect," Language
29.34-46.
Conklin, H. C. (1955). "Hanun60 Color Categories," Southwestern
Journal of Anthropology 14. 11-14.
323
324
BIBLIOGRAPHY
De Groot, A. W. (1948). "Structural Linguistics and Phonetic Law,"

Lingua 1. 175-208.
- - - (1948). "Structural Linguistics and Word Classes," Lingua
1. 427-500.
Finck, F. N. (1910). Die Haupttypen des Sprachbaues. Leipzig.
Frci, H. (1940). Interrogati/ et ltuiefini. Un probleme de grammaire
comparee et de linguistique generale. Paris.
- - - (1944). "Systemes de Deictiques," Acta Linguistica 4. 111-129.
(1948). "De la Linguistique Comme Science de Loi," Lingua
1. 25-33.
Fries, C. S., and K. L. Pike (1949). "Coexistent Phonemic Systems,"
Language 25. 29-50.
Meaning," Language 32. 195-216.
Grammont, M. (1895). La Dissimilation Consonantique dans les
Langues Indo-europeennes et dans les Langues Romanes. Dijon.
Greenberg, J. H. (1954). "A Quantitative Approach to the Morpho
logical Typology of Language," in R. Spencer (ed.), Method
and Perspective in Anthropology, 192-220. Minneapolis. (Re
printed in International Journal of American Linguistics 26.
178-194. 1960.)
- - - (1957a). Essays in Linguistics, ch. 6. Chicago.
- - - (I957b). "Nature and Uses of Linguistic Typologies," Inter
national Journal of American Linguistics 23. 6877.
Guireaud, P. (1954). Les Caracteres Statistiques du Vocabulaire;
Essai de methodologie. Paris.
Harary, F., and H. Paper (1957). "Toward a General Calculus of
Phonemic Distribution," Language 33. 143-157.
Hartmann, P. (1956-1957). Untersuchungen zur Allgemeinen Gram
matik. 3 vols. Heidelberg.
Haudricourt, A. G. (1939). "Methode pour obtenir les Lois Con
cretes en Linguistique Generale," Bulletin de la Societe Lin
guistique de Paris 40. 70-74.
Haugen, E. (1950). "The Analysis of Linguistic Borrowing," Language
26. 210-231.
Hermann, E. (1942). "Probleme der Frage," Nachrichten Akademie
der Wissenschaften in GiJttingen, Philosophische-historische Klasse,
122-408.
BIBLIOGRAPHY
325
.,.. Hjelmslev, L. (1928). "Principes de Grammaire Generale," Kong/ige

Danske Videnskabernes Selskab. Hist.fil. Meddelelser 16. 1-363.
- - (1935, 1937). "La Categorie des Cas. Etude de grammaire
generale," Acta Jutlandica: Aarsskri/t, vol. 7, fase. I; vol. 9,
fasc.2.
Hockett, C. F. (1955). "Manual of Phonology," Indiana University
Publications in Anthropology and Linguistics. Memoir II. Balti
more.
- - (1958). A Course in Modern Linguistics (especially ch. 64).
New York.
Hoenigswald, H. M. (1960). Language Change and Language Recon
struction. Chicago.
Hoijer, H. (1948). "Linguistic and Cultural Change," Language 24.
335-345.
- - (ed.) (1954). Language in Culture. Chicago.
Householder, F. W., Jr. (1960). "First Thoughts on Syntactic Indices,"
International Journal of American Linguistics 26. 195-197.
Humboldt, W. (1836). Ueber die Verschiedenheit des Menschlichen
Sprachbaues und Ihren Einfluss auf die Geistige Entwicklung des
M enschengeschlechts. Berlin.
Hymes, D. H. (1961). "On Typology of Cognitive Styles in Lan
guage," Anthropological Linguistics 3. I. 22-54.
- - - (1960). "Lexicostatistics So Far," Current Anthropology
1. 3-44.
Jakobson, R. {l 94 l). Kindersprache, Aphasie und Allgemeine Laut
gesetze. Sprakvetenskapliga Siillskapets i Uppsala Forhandlingar.
- - - , C. G. Fant, and M. Halle (1952). Preliminaries to Speech
Analysis: The Distinctive Features and Their Correlates. 2nd
printing. Cambridge, Massachusetts.
- - - , and M. Halle (1956). Fundamentals of Language. 's Graven
hage.
- - (1958). "Typological Studies and Their Contribution to
Historical Comparative Linguistics," Proceedings of the Eighth
International Congress of Linguists, Oslo, 17-25.
- - (1961). "Why 'Mama' and 'Papa'?" in B. Kaplan and S. Wapner
(eds.), Perspectives in Psychological Theory. 124-134. New York.
Kluckhohn, C. (1953). "Universal Categories of Culture," in A. L.
Kroeber (ed.), Anthropology Today, 507-523. Chicago.
326
BIBLIOGRAPHY
Kroeber, A. L. (1960). "On Typological Indices; I. Ranking of Lan

guages," International Journal of American Linguistics 26.171
177.
Kuznetsov, P. S. (1956). Die Morphologische Klassi/ikation der
Sprachen. Halle. a. S.
Lenneberg, E. H., and J. M. Roberts. (1956). "The Language of
Experience, A Study in Methodology," Indiana University Pub
lications in Anthropology and Linguistics. Memoir 13. Baltimore.
Lounsbury, F. G. (1956). "A Semantic Analysis of the Pawnee Kinship
Usage," Language 32.158-194.
Maclay, H., and E. E. Ware (1961). "Cross-Cultural Uses of the
Semantic Differential," Behavioral Science 6. 185-190.
Malkiel, Y. (1951). "Lexical Polarization in Romance," Language
27.485-518.
Mandelbrot, B. (1954). "Structure Formelle des Textes et Com
munication; Deux Etudes." Word 10. 1-27.
Martinet, A. (1955). Economie des Changements Phonetiques; Traite
de phonologie diachronique. Berne.
- - - (1960). Elements de Linguistique Generale. Paris.
Marouzeau, J. (1948). "Quelques Vues sur l'Ordre des Mots en Latin,"
Lingua I. 155-168.
Mathesius, V. (1939). "Verstarkung und Emphase," in Melanges de
Linguistique Offerts a Charles Bally, 407--413. Geneva.
Menzerath. P. (1950). "Typology of Languages," Journal of the
Acoustical Society 22. 698-701.
- - , and W. Meyer-Eppler (1950). "Sprachtypologische Unter
suchungen; I. Allgemeine Einfilhrung und Theorie der Wort
bildung," Studia Linguistica (Lund). 54-93.
- - - (1954). Die Architektonik des Deutschen Wortschatzes. Bonn.
Milewski, T. (1957). uLe Probleme des Lois en Linguistique Generaie,"
Lingua Posnaniensis 6. 120-136.
Moore, B. R. (1961). "A Statistical Morpho-Syntactic Typology Study
of Colorado (Chibcha)," International Journal of American
Linguistics 27. 298-307.
Murdock, G. P. (1959). "Cross-Language Parallels in Parental Kin
Terms," Anthropological Linguistics L 9. 1-5.
Pedersen, H. (1949). "1st eine allgemeine Sprachwissenschaft auf
empirischer Grundlage moglich 1", Archlv Orientaln/17. 236-238.
1
I
BIBLIOGRAPHY
327
Pike, K. L. (1948). "Tone Languages," University of Michigan Pub

lications in Linguistics. No.4. Ann Arbor.
Pittman, R. S. (1948). "Nuclear Structures in Linguistics," Language
24. 286-292.
Posner, R. P. (1961). "Consonantal Dissimilation in the Romance
Languages," Publications of the Philological Society. No. 19.
London.
Prieto, L. J. (1954). "Traits Oppositionnels et Traits Contrastifs,"
Word 10.43-59.
Reichling, A. (1948). "What is General Linguistics?" Lingua l. 8-24.
Robins, R. H. (1961). "Syntactic Analysis," Archivum Linguisticum
13.78-89.
Romportl, M. (1955-1956). "Zum Problem der Fragemelodie,"
Lingua 5.87-108.
Royen, G. (1929). Die nominale Klassi/ikations-Systeme in den Sprachen
der Erde. Modling bei Wien.
Salzmann, Z. (1950). "A Method for Analyzing Numerical Systems,"
Word 6. 78-83.
-Sapir, E. (1921). Language, ch. 6. New York.
_ _ (1944). "Grading: A Study in Semantics," Philosophy of
Science 11. 93-116.
_ _ and M. Swadesh. (1932). "The Expression of the Ending
Point Relation in English, French, and German," Linguistic
Society of America, Language Monograph. No. 10.
Saporta, S. (1955). "Frequency of Consonant Clusters," Language
31.25-30.
___ (1957). "Methodological Considerations Regarding a Statistical
Approach to Typologies," International Journal of American
Linguistics 23. 107-113.
Shalmon, C. (1951). "Prediction and Entropy of Printed English,"
Bell System Technical Journal 30. 50-65.
Schmidt, W. (1926). Die Sprachfamilien und Sprachenkreise der Erde.
Heidelberg.
Spang-Hanssen, H. (1959). Probability and Structural Classification
in Language Description. Copenhagen.
Steinthal, G., and F. Misteli. (1881-1893). Abriss der Sprachwissen
schaft. 2 vols. Bedin.
Stern, G. (1932). "Meaning and Change of Meaning," G6teborg Hog
skolas Orsskrift 38. No. 1.
328
BIBLIOGRAPHY
Sturtevant, E. (1917). Linguistic Change: An Introduction to the

Historical Study of Language. Chicago.
Trager, G. L. (1941). "The Theory of Accentual Systems," in L. Spier,
A. f. Hallowell, and S. S. Newman (eds.), Language, Culture
and Personality; Essays in Memory of Edward Sapir, 131-145.
INDEX
Menasha.
----. (1961). "The Typology of Paralanguage," Anthropological
Linguistics 3. part l. 17-21.
..Trubetskoy, N. S. (1939). Grundzilge der Phonologie. Prague.
Ullmann, S. (1953). "Descriptive Semantics and Linguistic Typology,"
Word 9.225-240.
(1957). Principles of Semantics, 2nd ed. Glasgow.
Voegelin, C. F. (1955). "On Developing New Typologies and Revising
Anticipation (continued)
Activity factor, 307, 309, 311, 312
and regressive sound change, 48
Adjectives, 77, 85-89
Old Ones." Southwestern Journal of Anthropology 11. 255-260.

Antonyms, 159
and nouns, 95
(1956). "Linear Phonemes and Additive Components," Word

and onomatopoeic pattern, 226
Adverbs, 79-80, 87-88
Aphasia, 306, 320
12. 429-443.
Affective reaction, 312-317, 320-322
Apodosis, 83-84
shared framework of, 313, 321
----. and J. Yegerlehner (1956). "The Scope of Whole System

Apology (Plato), 127
Affixes, 92, 110, 111, 112
(,Distinctive Features') and Subsystem Typologies," Word 12.

Arndt, W., 153
derivational, 122
Assertives, 150-151
444-453.
inflectional, 122
Assimilation, 47-49
positional classes of, 134
- - - . R. A. Ramanujan, and F. M. Voegelin (1960). "Typology of

and articulation, 47
Affricates, 25, 57
Density Ranges; I: Introduction," International Journal of

regressive, 48
analysis of, 68
Attitudes, 150, 152-153
Agglutination, 124-126, 129, 133, 134
American Linguistics 26. 198-205.

Augment, 135-138
and prefix-suffix ratio, 132, 139
Weinreich, U. (1953). "Languages in Contact, Findings and Problems,"

Austerlitz, Robert, 144
and sound change, 128
Austin, William, 39
Publications of the Linguistic Circle of New York. No.1. New

Aginsky, B. W., 144
Auxiliaries
Aginsky, E. G., 144
York.
inflected, 84-85
Allen, W. S., 45
(1957). "On the Description of Phonic Interference," Word

uninflected, 85
Allomorphs, 39, 122
Axmanova, Ol'ga S., 144
13. 1-11.
and analogic change, 40-41
nonzero, 94, 112, 270
Whorf, B. L. (1956). Language, Thought, and Reality: Selected Writings

Basic-order typology, 76-80, 90-91,
Allophones, 39, 42, 56
of Benjamin Lee Whorf, J. B. Carroll (ed.). New York and Cam

96-113
Alternation, 40-41
and affixes, 92-93
bridge, Mass.
Ambifixing, 92
in interrogative sentence, 80-83
Amelioration, 247, 254
Wits, J. (1935). De Nominale Klassi/ikation in de Afrikaansche Neger

and pronouns, 91-92
Amosova, N. N., 179
Bastian, A., 284
talen. Nijmegen.
Analogic change, 40-41
Bazell, C. E., 146
leveling effect, 43
Wolff, H. (1959). "Subsystem Typologies and Area Linguistics,"

Bendix, Edward H., 144, 186, 193
Animal systems, 6-14
Anthropological Linguistics 1. 7. 1-88.
Bengali Phonetic Reader, 118
and biological relevance, 10
Black, M., 178
genetic transmision of, 11-13
Zipf, G. K. (1925). The Psycho-Biology of Language. Boston.
Block, Jules, 118
transmission and reception, 9
and vocal-auditory channel, 8
Anthropology and language
universals, 279-294
Anticipation, xxiv
and assimilation, 47-48
Bloomfield, Leonard, 1,2,21,67,146,
191, 192, 218, 230, 236, 242,
248, 250, 281, 299, 302-303
Boas, Franz, 279, 281,287,292
Bolinger, Dwight L., 144
329
1
INDEX
Breal, Michel, 217, 230, 232, 247
Brondal, Viggo, 266
Brown, R W., 175-176, 318, 322
Budagov, R. A., 169, 186
Buhler, K., 170
330
Continuity, semantic, 187, 188
Cowan, G. M., 18
Creolization, see Pidgins
"Cultural Causality and Law: A
Trial Formulation of the De
velopment of Early Civiliza
tion" (Steward), 290
Culture, 280-285
relativism of, 292-294
uniformities in patterns of, 280-281
universal functions of, 285-292
Carnap, Rudolf, 149-150, 191, 192,

276
Casagrande, Joseph B., 30, 45, 154,
284,319,322
Case, 95-96, 112-113
and proximity, 103
Cassirer, Ernst, 144, 167. 169
Davenport, W., 290
Cenemes, 12, 15,20
Declarative sentence, 110, 111
Center for Communication Sciences,
word order in, 77, 80, 269-270
M.LT., 275
Declension, 168
Change, linguistic, 30-51, 126-127
Deictic elements, 21, 154-158, 160
assimilatory, 47
and bee-dancing, 22
gradualness of, 39
in language, 145
regularity of, 38-39
Demonstratives, 86
and replacement. 35-38, 45-46, 287 Derivational categories, 92, 122
see also Sound change
and inflection ratio, no, 134
Chao, Y. R, ISS, 171, 177
Desideratives, 137
Charmey experiment, 33
Designata, 145, 288
Chatterji, S. K., 118
compatibilityof,167-168
Child, l. L., 282
and denotation, 191
Chomsky, Noam, 65, 132, 145, 147.
universality of, 184-186
172
Designation, contextual effects on,
Circumlocution, 192
177-184
Clusters, 57-58, 67, 305
Designators, 143, 145, 149-150
Codability, 318-319
Diaphones, 38, 43
Cognition, 319
Dictionary, 233-234,253
Collinson, W. E., 156
Die gotische Bibel, 119
Color, 318
DiphthongS, 67
Comenius, John Amos, 275
Discourse.. semiotic, structure of,
Command, 150
163-177
Components, 185-187
Dolgopol'skij, A. B., 182
special characteristics of, 188-190
Duality, 12
Compounding, 129-130, 133
Duals, 41-42
Computers, 309
Du Bruhl, E. L., 283, 296
Concord, 133, 168
Conditionals, 103, 111
Eighth International Congress of
Congress of Linguists at Cambridge
Linguists, 45
(1962), 107
Elevation, 31
Conklin, Harold C., 144, 182, 185, Ellipsis, 159, 176
188,255,291
Emeneau, M. B., 279
Consonants, 55
Encli tics, 120
primary nasal, 56-57
Episememes, 146
secondary nasal, 57-58
Equivalence, restricted, xx
Constructions
Essai de semantique (Breal), 217, 230
endocentric, 63-64, 102
Essays in Linguistics (Greenberg)
morphological, 104
132, 135
relations among, 98
Ethnolinguistics, xv, 293
subject-predicate, 74
Etymology, 247-24d
Contamination, denotive, 315
Euphemism, 245-247
331
Evaluation, 307, 309, 311, JIl2
Extension, 244, 254
Factor analysis, 306-310
Features, xvi, 65, 305
and phonology, xvii
psycholinguistic implications of, xvii
replacement pattern, 36
voice-qualifying, 162
Ferguson. C. A., 266, 304
Flvrier, J, G., 172
Fichte, Johann Gottlieb, 224
Final particle, see Affixes
Fodor, J. A . 193
Form classes, 23
Formator, 145, 149-150, 160, 173, 184
attitudinal, 151-154
and endearment, 153
grammatical meaning and, 169
and tense, 156
Fortunatov, F. F., 271
Frake, C. 0., 291-292
Freeland, L. S., 153, 155, 175, 177
Frege, Gottlob, 191
French, D., 147
Friedrich, P., 45
Funke, 0., 191
Gender, xxi, 93-94, 113
neutralization of, 103
and number, 95,96, 112
Genitives, 78, 98-100, 273
Genotypes, 302
Geography, linguistic, 234-235
Gillieron, J. L., 237
Gleason, H. A., Jr., 155
Glossocentricity, 30, 31
Glottalization, xvi-xvii
Glottochronology, 34, 184-185,
284-285
Goodenough, W., 182, 185,191
Grammar, xxi, 4-5, 145-147
finite-state, 65-66
generative, 65-{)7
patterning of, 22
rules of, 65
transformational, 50-51, 65-{)6
universals of, 21-24, 61-{)2, 73-104
Grammatical classes, 115
Grammatical meaning, 169
Grammaticalization, 44
Grammont, M., 31,46
Greenberg, Joseph H., 28, 41, 44, 48,
51, 116, 117, 120, 122, 123,
125, 131-132, 134. 135, 144,
INDEX
Greenberg, Joseph H. (continued)
155, 156, 264, 265, 266, 268,
269-270, 271, 287, 290, 304
Hallig, R, 253-254
Hallowell, A. I., 287-289
H andbuch der altbulgarischen
Sprache, 119
Handbuch der neugriechischen
Volksprache, 119
Harris, Z. S., 182
Haugen, Einar, 51, 107, 157, 180
Havers, Wilhelm, 46
Hebrew, modern, 33
Heidegger, Martin, 224
HeJie, Pierre, 264
Helmholtz, Hermann, 263
Hermann, E., 33
Historical Greek Grammar
(Schwyzer). 31
Hittite Chrestomathy (Sturtevant).
118
Hjelmslev, L., 12, 146, 191,252
Hockett, C. F., 8, 12, 16, 26, 51, 56,
100, 104, 142, 144. 153, 157,
161. 162, 169, 170, 171, 174,
255, 301
Hoenigswald, H. M., 273, 277, 281,
285, 287
Homonymic clashes, 237
Homonyms, 179, 235-237, 238, 254
ambiguity, 237
and converging sound development,
235
differentiation methods, 236
Householder, Fred, 28, 104, 144, 156,
161,255
Hrushovski, Benjamin, 144, 168
Husserl, E., 191, 275
Hymes, D. H., 184, 190, 255
Hyperforms, 49
Hypersemanticization, 147-148
and incompatible designata, 168
Iconic relations, 10-11, 15, 16, 142
and word order, 269-270
Idiolects, 32
Idioms, 146, 181-184
and determination. 183
Iliad, 118-119
Immediate constituent, 63
Imperative, 151
Indo-European languages, 44, 57
and allomorphic variety, 41
dual in, 41-42
~~
'I
t
,I
)1
!I
332
INDEX
Indo-European (continued)
and inflection, 123
and suffixes, 134
Infixing, xx, xxii, 62, 92, 123, 132,
133,222
Indo-European, 134-135
Inflection, 4, 44, 168, 305
and derivation, 93, 122
Injunctive, 135
Inscriptions d' Asoka, Les, 118
Institute of Communications
Research, Illinois, 306
Intension, 191-192,244
Interaction, cognitive, 315
Intercalation, 92
International Journal 0/ American
Linguistics, 116
Interrogative sentences, 80-85, 160
Intonation, 19, 110
lroquoian languages, 56, 59
Isoglosses, 275
Isolation index, 132
Isophones, 275
Issatschenko, A., 277
lvie, Milka, 144
Ivic, Pavel, 144, 154, 160
Jakobson, Roman, 5, 25, 45, 51, 104,
144, 151,290,304
Jenkins, J., 264
Jesperson, 0., 217
Joos, M., 191
Kameneckajte, N. L., 183
Karolak, S., 178
Katz, J. J., 193
Kemeny, J. G., 3
Kent, R. G., 47, 118
Kernel sentences, 167
Khamena logia, 119
Kluckhohn, c., 281, 282, 283, 290
Kolmogorov, A., 272
Krader, Lawrence, 144
Knlmsky, J., 268
Kroeber, A. L., 130, 279, 285
Kuipers, A. H., 24
Kurylowicz, J., 46, 49
Labial consonants, 56, 57
Ladefoged, Peter, 56
Lamb, Sydney M., 12,28,283
Language
antiquity of, 286--287
arbitrariness, 10, 15, 16
and culture, 280-285
defined, 15-18
Language (continued)
defining set for, 3, 6
differences in, 20
differentiation in development of,
304
displacement, 11, 15, 16
feedback, complete, 9, 15, 16
generalizations, 18-20
generic function of, 287-288, 293
grammatical type of, 222-223
interchangeability, 9, 15, 16
learn ability, 13, 19, 306
lexicological type of, 222
openness, 11, 15, 16
philosophers of, 144
polysynthetic, 219
and rapid fading, 9,15,17
reflexiveness, 13
semantic structure of, 142-193
semiotic stratification of, 148-163
specific versus general terms of, 228
subsystems, 12
"Language games," 143
"Languages and Logic" (Whorf), 318
Lapsing and sound change, 48
Leaf, Walter, 118
Leisi, E., 167
Lenneberg, Eric H., 286--288, 318, 322
Leskien, A., 119
Leumann, Manu, 39
Lexeme, 182, 188
Lexical constants, 249-250, 254
Lexical fields, 250-253, 254
Lexical patterning, 273-274
Lexicology, Soviet, 192
Lexicon, 285, 306
see also Dictionary
Lexicostatistics, 34
Linguistic relativity, 292-294
Linguistics, 316
and language universals, 263-277
as a science, 299-306
Linking, 163-167, 168-169, 176,193
and summation, 170
symmetry of, 170
Linton, R., 280
Literacy and linguistic change, 34
Loan-translation, 239
Locatives, Osco-Umbrian, 44
Lotz, John, 144
Lounsbury, F. G., 185, 191,303
McQuown, N. A., 191
Major function, 172
"Mama," 5, 7, 301
333
INDEX
Morris, C. W., 144, 145, 192
Man and Culture (Wissler), 281
Motivated words, 222
Marcus, H., 175
Murdock, G. P., 29, 281-284, 303
Markers, 21
Marketing Center Company, Tokyo,
Naming, 143
308
Martinet, A., 39, 46, 47, 48,175,176
Narrowing, 31
Marty, A., 191,275
Nasality, 265-266, 267
Maspero, H., 170,183
prosodic features of, 55
Nasals, 53-59, 135,266,304
Mazateco whistle-talk. 18
Mazeway, 289-290
Neoprefixes, 138
Meaning, extension and restriction of,

Nesting, 163-169,176, 193
244
asymmetries of, 171-172, 174-175
and major function, 173
Megasign. 182
"Neutral" mode, 150
Mel'cuk, I. A . 183
Metalinguistic operators, 162-163
Neutralization
Metaphor, 31,221,224.238-244, 321
of asserti veness, 150-151
anthropomorphic, 241-242
and nasal vowels, 59
"dead," 186--187
and PNC's, 57
general tendencies, 240-244
and proximity hierarchy, 102-103
and parallel development, 220, 238

voiced-voiceless, 55
240, 254, 284
Nexuses, 123
as quasi-universal, 49-50
Nida, E. A., 188, 189
synaesthetic, 242-244, 248, 254, 312,

Nominal apposition, 89-90
313
Noun phrase, 85-87
Method and Perspective in Anthro
Nouns, 4, 62, 74
pology: Papers in Honor of

Novum testamentum Gr04ce, 119
Wilson D. Wallis, 116
Nucleus boundary, 120
Metonymy, 31,284
Number, 94-95, 96-98
Mill, J. S., 191
Numerals, 86
Monosyllables, 236, 237
Morphemes, xxvi, 12, 20, 30, 92, 112,

Obstruents, 57
Old Persian (Kent), 118
134,219,305
derivational, 120, 122
Onomatopoeia, 221, 224, 225-226,
inflectional, 120, 121
245,254
intonational, 19
in poetry, 226--227
radical, 120
Osgood, Charles E., 44, 264, 283,
and replacement pattern, 36
285,293
segmental, 19
Outline 0/ Cultural Materials, 281
segmenting of, 120-121
Oxford Dictionary, 233, 249
Morphemic change, 38-39
Morphemics, 36--37
Parallel developments, 220
Patterning, 303
Morphological motivation, 222-224
Morphology, 22, 44, 92-96, 304
duality of, 12
grammatical, 11,20
complexity of, 125
plerematic and cenematic, 20
diachronic, 114-139
and phonology, 69
subsystems of, 12
Morphophonemes, 40
Paul, Hermann, 41
Peirce, C. S., 181, 191, 265, 269
and allophones, 43
Pejoration. 31, 247, 254
Morphophonemics
and sound changes, 128-129
Penfield, Wilder, 283
taxonomy, 22 .
Perebejnos, V. I., 179
Perseveration, xxiv
Morphs, 37, 39, 45-50
frequency distribution of, 69-70
"Personality Structure and the Evolu
meaningless, 146
tion of Man" (Hallowell), 287
Phenotypes, 301-302
zero, 121
~:
INDEX
Phonaesthemes, 146
Phoneme sequences, 67-70
Phonemes, xvi-xvii, 12, 20, 24-25, 30,
305
change in, 38, 41
displacements in, 46
hierarchy of, 266
and mergers, 44
nasal, 55-59
and replacement pattern, 36-37
split in, 43
Phonerrrics, 265-277
Phones, 37
Phonetic motivation, 222, 237
Phonological word. 65, 66
Phonology, xvi, xvii, 304
asymmetries of, 26
generative, 65-66
and glottochronology, 284-285
symmetry of, 25
and universals, xxi-xxii, xxiii, 24-28,
53-59,142-143,283-284
variance in, 290
Pidgins, xxiii, 33, 126, 133
Pleremes, 12, 15,20
PNC, see Consonants, primary nasal
Pocket Oxford Dictionary of Current
English, 236
Polysemy, 177-180, 184, 185,232-235,
254
and ambiguity, 234
and depletion, 180-181
in generic words, 232-233, 238
and idiomaticity, 182
incidence of, 188-189
Port Royal Grammar, 4, 5
Pos, H. J., 267
Postpositions, 76-79, 110
Potency factor, 307, 309, 311, 312
Pragmatic operators, 150
Prague School, 94
Precis de semantique fram.aise, 238
Predicate, discontinuity of, 104
Predicators, 23
Prefixing, xx, 92, 112
adverbial, 138-139
derivational,122
Indo-European, 134
ratio to suffixing, 131-132, 133,302
reduplication, 137-139
and regressive assimilation, 48
Prepositions, 76-79, 110
Preterit tenses, 135-136, 137
Primes, 65-66
334
Pro-forms, 157, 161
Pronouns, 91-92, 96,287-288
Proper names, 21-22, 161
semiotic nature of, 161
Proper noun, 89-90, 112
Propositional operations, 158-159
negation, 159
Protasis; 83-84
Proto-Indo-European, 127, 133, 135,
138-139
Protolanguage, 114-115, 127, 133,
135, 138-139
Proxirrrity hierarchy, 102-104
Psichari, Jean, 119
PsychQlinguistics, xv, xvii, 293
and attitudinal operators, 152-154,
155
and language universals, 299-322
and nature of universals, xviii
and Whorf, 317-318
Pure inflection, 133
Qualifiers, 308-309
Quantifiers, 86, 159-162
existential, 160
Quasi-universals, 302-303, 304
Question, 151-152
particle, 151
Quine, W. V., 161, 191,192,276
Ravid, Wita, 144,179
Redundancy,xxi,24,272,305
Reduplication, 137-139
Regn.!}l, H., 143
Reichenbach, Hans, 144, 149, 150,
162, 164, 168, 173, 176
Reichling, A., 169
Reisig, C. c., 217
Relations, grammatical, 96-104
Replacement pattern, 35-45, 50
and semantic change, 49
Richardson, Jane, 285
Riffaterre, Michael, 144
Rig-Veda. 117
Roberts, L., 283
Root, 92, 122
Sapir, Edward, 48, 147, 151, 153, 169,
170, 173. 229, 230. 253, 265,
270, 279, 286-287, 290
Saporta, Sol, 44, 267, 285, 300, 304
Saussure, F. de, 36, 50, 191, 217, 220,
221, 222, 223, 264, 277
SavCenko, A. N., 169
335
INDEX
Comrrrittee on Linguistics and
Schaff, A., 143
Psychology, xvi, 275
Schleicher, August, 169
Sofronow, M. W., 183
Schuchardt, H. E. M., 225
SlIlrensen, H. S., 161
Schwyzer, Eduard, 31
Sound change, 31. 39
Sebeok, T. A., 285
assirrrilatory, 46. 47
Secondary nasal consonants, see Con
Indo-European. 134
sonants, secondary nasal
and irregular paradigms, 40
Seiler, H., 159
and lapsing, 48
"Self' and Its Behavioral Environ

and morphophonemics, 128
ment, The" (Hallowell), 287
nonassirrrilatory processes of, 49
Semantic depletion, 180-181
regularity of, 38
Semantic differential, 307, 310-311
Sound symbolism, xxi-xxii, xxiii. 145,
Semantic factors
225-227
generality of, 310-316
Sound system
similarities in, 308-310
common features of, 53
structure of, 306-316, 321
symmetry of, xvi-xvii. xviii
Semanticity, 10
Sound units, see Phonemes
Semantics, xxii, xxiii, 142-193
changes in, 31, 37, 39, 46, 49,

Southwest Project in Comparative
PsYcholinguistics, xxvi
284-285
Spang-Thomsen, B., 172
and civilization, 284
Specialization. 10
combinatorial, 163-184, 192
componential structure of, 185-187

Speeches to the German Nation, 234
Spelling, 236-237
continuity, 187-188
Sperber, Hans. 240-241
descriptive, 221-238
Spirants, 25
and grammar, 145-147
Spuhler, J. N., 283
historical, 238-248
Stankiewicz, E., 153
laws, 217-219
Stem, G., 50
and motivation, 222,237
Steward, J. H., 279, 282, 290
structural, 248, 303
Stops, 26-27, 266
synchronic, 50 .
Strawson, P. F. 159
typology of, 237-238
Streitberg, Wilhelm, 119
universals in, 217-255
Stress, 67
Sememes, 146
phonerrric, 68
Semiotics, 142-193
three main types, 54
mechanisms of, 142-144
Sturtevant, E. H . 41, U8
organization, marking of, 168-172
Suffixing, xx, xxii, 92, 112
and stratification of language, 148

in Indo-European, 134
163
ratio to prefixing, 131-132, 133
x and structure of discoun;e, 163-167
and regressive assimilation, 48
.:sendel, E. I., 169
Superlatives, 162
Shannon, C., 24
Swadesh, Morris, 33, 34
Shwayder, D. S., 156
Syllabic structure, xxv-xxvi, 66
Signs, 145
Syllabics. nasal, 55,59
combinations of, 163-167
Syllable, 267
deictic, 145, 154-158, 160
Synaesthesia, see Metaphor, synaes
function status of, 171-172
thetic
indexical, 150, 265
Synonyms, 230-232
metalinguistic, 145, 162-163
immediate, 186
motivatedness of. 184-185
"law of distribution," 230, 231
object-language, 145
"law of synonyrrric attraction," 231
'" syntactic, 150
and parallel change, 231-232
Smelev, D. N., 182
radiation of, 231
Social Science Research Council,
INDEX
Syntactic operators, 168
Syntax, 22, 80-92
Synthesis, 124-128, 133, 134
Taboos, 245-247, 250, 254, 284
Tagliavini, C., 239
Terminologization. 186
degree of, 187. 189
Thesaurus (Roget). 253
Thumb. Albert. 119
Tongue height, 27
Topic and comment structure. 23
Toropcey. I. S., 183
Trager. George L., 12
Traite de phonetique (Grammont), 31
Transformation, 65-66, 99, 146,303
Trier, Jost, 250, 253
Trnka, B. 236
Trubetzkoy. N. S. 57
Typology. linguistic, 4-5, 61-70
basic-order, 76-80
Indo-European, 114-139
phonological, 63
semantic. 220-221,237-238
Ullmann. Stephen. 143, 144, 153, 178,
181, 185, 281, 284, 306, 313,
321
Uniformities, xv
in patterns of culture, 280-281
Universals, xv-xxvii
and anthropology, 279-294
classes of, xxi-xxii
defined, xix, 219
definitional, xix, 3, 6
diachronic, xxii-xxiv, 31-32, 40-51,
.114-139, 221, 237, 284-285,
303, 306
of frequency distribution, xxi, 66

67, 304-305
grammatical, xxii. 20-24, 73-104
implicational, xix-xx
interrelations of, xxiv-xxvi
inguistic, 316
md phoneme distribution, 61-70
phonological. xxii, 24-28, 53-59,
271,283-284, 304
and psycholinguistics, 299-322
psychological, 316-317
semantic, xxii, 142-148, 217-255,
283,284,303,306-316
and sound change, 26, 30-51
sound symbolic. xxii
statistical, x, 220, 302
336
Universals (continued)
synchronic, XlIii-xxiii, 30, 50, 218,
221,237,303
syntactic, 80-92, 304
unrestricted, xix, 220
in vocabulary structure, 248-255
Urban, W. M., 232
337
Word order (continued)
and dominant order, 77, 96
and genitive, 78
within noun phrase, 85-87
and number agreement, 94
of verbal forms, 83-87
Works and Days (Hesiod), 127
Worth, Gerta, 274
INDEX
Writing, 14-15
Wundt, Wilhelm, 312
Ziff, P., 143
Zimmer, Karl E., 144, 193
Zipf, G. K., 233, 304
Zvegincev, V. A., 143, 169, 178, 180,
191
Verbs, 4, 93-96
impersonal, 176
Vi co, G. B., 241
Vocabulary, 23, 49-50, 143
and culture, 227-230, 245-247
and glottochronoiogy, 34-35, 284
and lexical fields, 253
structure and content of, 184-190,
218,221,248-254,273
Vocatives, 121-122, 176
Voegelin, C. F., 172
Voegelin, F. R., 172
Voice qualification, 153
Voicing, 48
and primary nasal consonants, 56
Vowel color, 25
and stress, 54
Vowel systems, 27, 54
Vowels, nasal, 55,58-59
Wallace, A. F. C., 188,289,293-294
Wartburg, W. von, 253-254
Watson, J. B., 299
Wegener, P., 149
Weinreich, UrieI, 144, 152, 187, 188,
192, 271, 273, 276, 288, 289
Weiss, A. P., 299
Wells, Rulon S., 51, 144, 191, 192
Werner, Heinz, 244
Whiting, J. W. M., 282
Whorf, Benjamin L., 102, 151, 166,
167, 168, 173, 190, 229, 253,
290,294,316-318
Widening, 31,37
Wissler, C., 281
Wittgenstein, L., 143, 191,233
Word, 219
frequency and polysemy, 233-234
neutral,247
semantic autonomy of, 238
structure and homonyms, 236
transparent and opaque, 221-227
Word boundaries, 120
Word order, 73-92, 168, 305
of adjectives, 87-89
in declarative sentences, 77
.1

Joseph H. Greenberg-Universals of Language - Revised 2nd Edition-The MIT Press (1966)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Joseph H. Greenberg-Universals of Language - Revised 2nd Edition-The MIT Press (1966)

Uploaded by

Copyright:

Available Formats

,

EDITED BY JOSEPH H. GREENBERG

REPORT OF A CONFERENCE HELD AT

THE M.I.T. PRESS

. The Massachusetts Institute of Technology

All Rights Reserved. This book may not be

Second printing, first paperbaCk edition, March 1966

The Conference on Language Universals was held at Gould

suIt. Moreover, no viable alternative had presented itself. As

contain either verbatim or edited reports of the highly stimulat

ing and productive discussions which took place. However, the

recordings of these discussions and subsequent written comments

and criticisms were circulated and have been incorporated in

In addition to individual bibliographies, a general bibliography

of language universals has been compiled as the final section of

this work. This bibliography is obviously not exhaustive. It

through empirical investigation, and, in a few cases, analytical

discussions which seem suggestive from this point of view. I

to Uriel Weinreich for advice and assistance in its preparation.

Memorandum Concerning Language Universals

The Problem of Universals in Language

Are There Universals of Linguistic Change?

Assumptions about Nasals: A Sample Study in

Phoneme Distribution and Language Universals

Some Universals of Grammar with Particular

6. A Search for Universals in Indo-European Diachronic

7. On the Semantic Structure of Language

9. Implications of Language Universals for Linguistics

10. Language Universals in Anthropological Perspective

11. Language Universals and Psycholinguistics

Dr. Einar Haugen

Dr. Harold C. Conklin

"'Dr. Charles F. Hockett

Dr. FrankUn S. Cooper

Dr. Henry M. Hoeoigswald

Dr. Warren C. Cowgill

Dr. Fred W. Householder,lr.

Dr. William I. Gedney

.Dr. Dell H. Hymes

"'Professor Charles F. Hockett of Cornell University could not attend the

Dr. Wallace E. Lambert

Dr. Herbert H. Paper

Dr. StepheD UIlmaDD

Dr. Uriel Welnrelcb

Dr. Floyd G. Louosbury

Dr. Rulon S. Wells

Since a number of the papers in the present volume, including

eralizations." However, it seemed also to be generally agreed

of universals. This also means that the proposed research on

New York, New York

MEMORANDUM CONCERNING LANGUAGE

presented to the Conference on Language Universals,

Gould House, Dobbs Ferry, N. Y., April 13-15, 1961

Underlying the endless and fascinating idiosyncrasies of the

found, the linguist at this point formed the hypothesis that a

uted roughly around the 50% point. A study of one language

The examples just cited illustrate that the term "universal" is

4. Logical Structure of Universals

These are characteristics possessed by all languages which are

These always involve the relationship between two characteris

Such implications are fairly numerous, particularly in the phono

a significantly greater probability of possessing some other charac

Finally we have instances where a certain measurement, for

A second basis of classification which obviously crosscuts the