PART 1

INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS

Vol. 21, No. 4, October 1955

INDIANA UNIVERSITY PUBLICATIONS

in
ANTHROPOLOGY AND LINGUISTICS
MEMOIR 11
of the

INTERNATIONAL JOURNAL OF
AMERICAN LINGUISTICS

A MANUAL OF PHONOLOGY
By
CHARLES F. HOCKETT

Published at the Waverly Press, Inc.

Mt. Royal and Guilford Aves.,
Baltimore 2, Md.
by Indiana University
under the auspices of
Linguistic Society of America
American Anthropological Association
with the cooperation of
Joint Committee on American Native Languages

572.082
I39m
no. 11
c. 2
 Digitized by the Internet Archive
in 2011 with funding from
LYRASIS Members and Sloan Foundation

http://www.archive.org/details/manualofphonologOOinhock
A MANUAL OF PHONOLOGY
MEMOIR 11

By
CHARLES F. HOGKETT

WAVERLY PRESS, INC.

Baltimore

J
955

UNIVERSITY OF FLORW*
IVmmfviCJ
\'/a
 Indiana University Publications
in Anthropology and Linguistics
Editors: C. F. Voegelin
Glenn A. Black
George Herzog
Paul Radin
Thomas A. Sebeok
Memoir 11, i-vii + 1-246
Issued October, 1955
Paper $3.50

Memoir of
International Journal of
American Linguistics

International Journal of American Linguistics is published quarterly at the

Waverly Mt. Royal and Guilford Avenues, Baltimore i, Md., by
Press, Inc.,
Indiana University. Subscriptions ($4.00 a year) will be received by the Waverly
Press, Inc., Mt. Royal and Guilford Avenues, Baltimore 2, Md., or by the De-
partment of Anthropology, Indiana University, Bloomington, Ind.
Entered as second class matter May 11, 1945 at the post office at Baltimore, Maryland under the Act of March 3, 1879.
 ;

PREFACE

This manual was begun during the Spring Semester of 1952-3, while I was on
Sabbatical Leave from Cornell University, and also the recipient of a grant from
The Rockefeller Foundation, administered through the University. I wish to
express my gratitude to both of those institutions, whose assistance made it

possible for me to spend my time in research and writing rather than in directly
remunerative activities.
When first planned, what appears here was to be but the initial section of a
treatise designed to cover all phases of linguistics. The size of the present volume
will convince the reader, I am sure, that my original notion was hardly feasible.
I hope that in due time other portions of the treatise as originally conceived may
appear, but the expression of this hope must not be construed as a guarantee.
My debt to my colleagues, both at Cornell and elsewhere, is of course enor-
mous. The bibliographical references will indicate some small portion of this
debt. It would be impossible to single out, for mention here, all the individuals
who have been of more direct assistance; were I to attempt such a thing, I would
be bound to commit unforgivable errors of omission. However, there are several
acknowledgments which must be overtly made.
First: to Gordon H. Fairbanks, who has read the entire manuscript in (almost)
final form. I have not always accepted his suggestions, and the book is doubtless
the worse for that fact; but it is certainly the better for those which I have
accepted, and no blame must attach to him for errors which remain.
Second: to Roman Jakobson, who has been, over the years, a source of inspira-
tion which I value highly. I about my
am particularly anxious to be emphatic
debt to him because, in several places in the body of the book, I have been
forced to express sharp disagreement with, or criticism of, some of his most
fondly held views. Unfortunately, the spirit of scientific investigation leaves no
place for a softening of criticism for friendship's sake, nor for any worship of the
Idols of the Marketplace.
Third to Martin Joos, who read an earlier draft of 5 with extreme care, and
:

was (by request) remorseless in his criticism. He has also scrutinized 5 as it now
appears, and allows me to state that he is in complete agreement with the points
of view and the interpretations of fact which the reader will find in that section
wherever there is disagreement between what is said there and Joos's earlier
statements in his Acoustic Phonetics (1948), the more recent statements represent
his current view, as well as my own. This sort of help verges on collaboration,
and I should have been happy to acknowledge the fact by having Joos's name
appear on the title page as co-author of 5; but he has preferred merely this
prefatorial credit.
Fourth: to Carl F. Voegelin, editor of the series in which this work appears,
to Thomas A. Sebeok, the business manager, and to Mrs. Elsie F. Dosch, their

very competent editorial assistant for accepting the work, for raising most of

iii
the necessary funds for publication, and for their great care in guaranteeing that
the typography and format should be of maximum help to the reader.
Fifth: to Indiana University for defraying the cost of the text of this book,
and to the Hull Memorial Publication Fund of Cornell University for defraying
the cost of illustrations.
This volume is dedicated to the Grand Old Man of American linguistics,
George Melville Boiling, in sincere appreciation of his clear guidance when the
writer was first discovering what Bloomfield has called the "strangeness, beauty,
and import of human speech."
Charles F. Hockett
Cornell University
26 January 1955

IV
 TABLE OF CONTENTS

0. Purpose and Plan 1

1. The Organs of Speech and their Functions; Phonetic Notation 23

2. A Typology of Phonologic Systems 42

21. Pause and Macrosegment 43
22. Syllables and Juncture 51
23. Accentual Systems 65
24. Syllable Peak Systems 72
25. Syllable Margin Systems 91
26. Ultimate Phonologic Constituents 126
27. Balance and Symmetry 138

3. The Principles of Phonologic Analysis . 143

31. Gathering 144
32. Collation 149

4. Structure, Pattern, and Abstraction 175

5. Acoustic Phonetics 180

51. Sound and its Physical Analysis 181
52. The Problem of Acoustic Phonetics 190
53. Tentative Results 193
54. Summary 208

6. Functional Load and Analytic Norms 211

Notes 222

Glossarial Index and Bibliography 226

 0. Purpose and Plan. As its title implies, this manual is intended as a survey
of the methods so far developed, and of the results so far attained, in the field of
phonology. The title fails to indicate certain restrictions on the intended cover-
age. First, we shall not be concerned with the phonologic aspect of linguistic
change: the coverage is synchronic. Second, we shall not be concerned with varia-
tions of phonologic pattern from individual to individual, or from group to
group, within a speech community, but with coherent linguistic systems viewed
as undifferentiated into dialects or idiolects: the coverage is not merely syn-
chronic, but descriptive. What we take as a "coherent linguistic system" will in
one case be an idiolect; in another case, the common core of a number of idiolects;
in yet another, the overall pattern of a number. The logic of common core and
overall patternis dealt with briefly in 03 below.

Despite the presence of some discursive sections, this book is intended to be

consulted, not to be read certainly not to be read as a unit. The discussion is

therefore not broken up into chapters, in the ordinary pedagogic way. Instead,
the entire text is sectioned by brief descriptive headings preceded by section
numbers in decimal notation: 23 is the third subdivision of 2, 232 the second
subdivision of 23, and so on. A cross-reference ending in zero, such as 10,
which precedes 11, though the actual section numbers
refers to that part of 1
never end in zero. The index is also a glossary of technical and semi-technical

terms. Notes are few, mainly by way of scholarly credit, and are placed at the
end (cross-referenced by section number) rather than being run as footnotes.
01. Special Features. Our approach to phonologic pattern is marked by two
special features which must be described here.

The first is that we attempt to develop a typology a taxonomic frame of
reference in terms of which different phonologic systems can be classified and
compared. This is by no means a new aim in phonologic study. Trubetzkoy Jakob- ,

son, and others of the so-called "Prague group" did a great deal of typologic
classification: Trubetzkoy 's Grundzuge is, among other things, a suggested
typologic framework. Some American linguists have felt not only that Tru-
betzkoy failed, but that failure in such an attempt is inevitable the goal is a
false one. I do not agree with either part of this view. Trubetzkoy's frame of
reference was not sufficiently complex, but it was a worthwhile first approxima-
tion, some parts of which cannot yet be improved on. More recently, and it would
seem quite independently, Voegelin has proposed the value of a general (even if
arbitrary) typology for archiving purposes. The typology developed in 2 of
the present manual not supposed to be arbitrary, and it is considerably more
is

complex than either Trubetzkoy's or Voegelin's; but I am sure that it falls far
short of what we must eventually develop.
We are in an incomparably better position for typologic work now than we were
in the thirties. We have reasonably reliable and reasonably homogeneous re-
ports on a much larger number of languages than we did twenty years ago, or
even ten unfortunately, some of the reports which were unquestioningly used in
;

earlier years must now be regarded as not wholly trustworthy. A number of

1 A MANUAL OF PHONOLOGY

factors have contributed to this widening of our empiric base, but one above all

others deserves mention here: the far-flung field research of missionaries trained
by the Summer Institute of Linguistics. Their brief factual reports are perhaps
rarely exciting, but show on the average a high reliability; furthermore, since
much the same methodologic approach is involved in most of them, one quickly
learns to interpret them. Also worthy of special mention is the vital archival role
of the International Journal of American Linguistics since 1944.
The other special feature is closely related to the first we have sought a con-
:

sistent and complete working-out of an immediate constituent approach on the

phonologic level. This, also, is not a new aim. Bloomfield's terminology of "simple
primary phonemes," "compound primary phonemes," and "secondary pho-
nemes" is a hint in the direction of a hierarchic approach. The insistence of Hjelm-
slev on a kind of parallelism ("isomorphism," though not in the usual mathemati-
cal sense) between the level of expression and the level of content opens the way
for an extension to phonology (expression) of the immediate-constituent approach
which has long held a prominent place in grammatic study (study of content);

Kurylowicz has recently developed this theme. Last but by no means least
few years ago Pike and Pike prepared and published an IC analysis of Mazateco
phonology. Their article seems almost to have been worked out with tongue in
cheek, as a sort of offhand experiment; but in my opinion it is one of the really
germinal contributions to linguistics in the last few decades. When more fully
elaborated, the IC approach in phonology proves to be of key importance in the
development of a phonologic taxonomy; we shall see in 2 how this is.
Although it hardly constitutes a "special feature" of this manual, it would be
well to state explicitly that our view will be empiric: I accept Bloomfield's asser-
tion that "the only useful generalizations about language are inductive generaliza-
tions," and hence affiliate with the "god's-truth" school of thought rather than
with the "hocus-pocus" school. A mere acceptance of this approach does not in
itself guarantee success, and the manual is certainly full of mistakes of every sort.

One result of an empiric approach that will be regretted by some is that it in-
evitably involves a recognition of areas of indeterminacy. Empiricism forces the
overt formulation of criteria for making classificatory decisions, and always, it

would seem, the criteria fail in some instances. The hocus-pocus approach does
not encounter this difficulty in quite the same way, since it allows greater free-
dom in making ad hoc decisions arbitrarily.
Arbitrariness can manifest itself at various different levels of abstraction.
Trubetzkoy's typical manipulation of the data on a single language is not ar-
bitrary, since he operated with a fairly well formulated set of principles, which
specified in advance, for example, the conditions under which a segment would
be regarded as a single phoneme and those under which it would be interpreted

as a cluster. Yet these general principles can in turn be regarded as arbitrary,

at least in that various other equally self-consistent operating principles can be
formulated. This particular kind of arbitrariness is comparable to our free choice
between feet and meters in measuring lengths, except for one detail: conversion
from feet to meters, or vice versa, is easy and mechanical. The non-arbitrary
 A MANUAL OF PHONOLOGY 6

phenomenon which one measures in terms of this or that arbitrarily

of length,
selected unit, an entirely relative matter between different objects the lengths
is

of which can be measured. The danger which we encounter in letting ourselves

become too firmly attached to any one arbitrary unit (feet Trubetzkoy's prin-
;

ciples; Bloch's postulates) is that we will fail to realize the extent to which our
generalizations are dependent on the frame of reference. Some of the Praguian
generalizations turn out to be of this sort: they are not invariant under selection
of different analytic criteria. Certainly we should like for our generalizations to
be as free of this as possible; I hope that the approach developed in this manual
will represent a step in the right direction.

02. Phonology and Language Design. There is currently no general agreement

on the precise relationship of the phonologic pattern of a language to the rest of
the linguistic system. Certain investigators, among them Pike, believe that in
order to understand a phonologic system it is necessary not only to know certain
things about the accompanying grammatic system, but to make active use of the
latter knowledge as criteria for phonologic decisions. Others disagree. But the
issue is not clear: in actual fact, we are probably in covert disagreement as to
what our overt disagreements are about. Rather than dwell in detail on this and
kindred differences of opinion, I shall simply describe my own current view.
021. A Theory of Speech Communication. As a first step in describing this view,
we present in this section a mechanical and mathematical model of a human being
regarded as a talking animal. The model developed here goes beyond any previous
ones, particularly in that meaning is taken into account in a potentially precise
way.
It willbe well to indicate at the outset the limitation of our aims. Humans
share a great many categories of behavior with other animals: learning, goal-
seeking activity, some measurable degree of "intelligence," trial-and-error
patterns in solving problems, probably (though it is hard to demonstrate for
sure) implicit trial-and-error followed by explicit successful action, and the like.
It is no part of our purpose to try to account for features of behavior shared by
speaking and non-speaking species; this is the business of biology or of biophysics.
Furthermore, if there are distinctively human modes of behavior not dependent
for their development and continued existence on human language, these also
fall outside the scope of our interest. For our purposes, then, a human being is an

animal, partly like and partly unlike other animals, but unique among all animals
in the possession of speech; our mathematico-mechanical model is intended to
account only for this single uniqueness.
It may also be well to say something here, in advance of our detailed discussion,
as to the nature of a mathematico-mechanical model. We shall be presenting a
picture which looks vaguely like the block-diagrams ("control-flow charts")
used by electrical engineers, and shall be assigning various names to the units
portrayed in the diagrams; but we shall also be speaking as though these units

were to be found somewhere inside a human being say in his central nervous
system. Now, in the present state of knowledge of neurophysiology, there is no
guarantee that the units which we posit do exist inside a human skin nor, for
4 A MANUAL OF PHONOLOGY

that matter, is it currently possible to prove that they do not. This implies that
our mode of discussion is not physiologic. Rather, it isif" mode,
a type of "as
which can be explicated in the following two ways: humans, as users of
(1)
language, operate as if they contained apparatus functionally comparable to
that we are to describe; (2) an engineer, given not just the rough specifications
presented below but also a vast amount of detailed statistical information of the
kind we could work out if we had to, could build something from hardware which
would speak, and understand speech, as humans do. Indeed, given such a device,
and the solution to the problem of neuron-to-wire linkage, one could produce a
talking dog. Let me add at once that I am entirely serious.
Whether or not a mechanical model for a phenomenon constitutes an explana-
tion of the phenomenon depends, not on the model, but on the temperament of
the investigator. An explanation is something which satisfies one until one has
looked deeper; then one asks for an explanation of the explanation. Certainly
even those of us with the greatest fondness for mechanical models must avoid
the error of argument from analogy. Mechanical devices have been constructed
which will "learn" to run a maze, overtly in much the same way that a rat will.
We know the structure of the mechanical "rat," and can thus describe quite
precisely the correlation between its structure and its behavior. But to infer from
this, and from the overt similarity of behavior of the mechanical rat and the live
one, that the structure of the latter is in some sense comparable to that of the
former, is to argue from analogy. The experimentum cruris is wanting dissection :

of the live rat, demonstrating that its structure is indeed as the analogy suggests
or revealing that, after
all, it is not. The analogy tells us, in the last analysis,

nothing more than what should be looked for when physiologic techniques have
evolved far enough. The model proves only that an entity need not necessarily
be endowed with a non-material mind or soul in order to learn mazes as rats
learn them.
0211. The Control-Flow Chart. of the materials we shall need for our
Most
exposition are displayed in Figure which shows, in highly schematic fashion,
1,

two speakers of a single language, one of whom is in the process of speaking to

the other. In a time-honored tradition, we have labelled the two Jill and Jack.
When Jill transmits a message in linguistic form to Jack, the sequence of events
and the route followed by the message are, according to our theory, as follows:
(1) Jill's Grammatic Headquarters (= "G.H.Q.") emits a discrete flow of
morphemes. This flow constitutes the input to the transducer, within Jill, labelled
"Phoneme Source."
(2) Jill's Phoneme Source, operating in conformity with a certain code,
transduces the discrete flow of morphemes which is its input into a discrete flow of

phonemes which is its output. The code involved is the morphophonemics of the
language spoken by Jill and Jack.
(3) The output from Jill's Phoneme Source constitutes the input to Jill's
Speech Transmitter. This transducer is in part directly observable from the

outside (Jill's mouth, nose, throat, and so forth the so-called "speech tract"),
and we have therefore drawn the box which represents it in such a way that it
 A MANUAL OF PHONOLOGY

JILL JACK
Fig. 1

shares parts of boundary with the larger box which represents all of Jill.
its

But the region which performs the transduction about to be described must
of Jill
also include portions just as inaccessible to direct observation as are her Pho-
neme Source or G.H.Q. The Speech Transmitter converts the discrete flow of

phonemes which comes to it into a continuous speech signal a continuous train
of sound waves. The code by which the Speech Transmitter performs this trans-
duction is the phonetic system of the language.
(4) The speech signal reaches Jack's Speech Receiver, wherein it is retrans-
duced into a discrete flow of phonemes; the latter constitutes the output of a
Speech Receiver. The code by which the Speech Receiver operates is thus neces-
sarily the same as that by which a Speech Transmitter functions, except that
the Speech Receiver applies the code hind-end to.
(5) The output of Jack's Speech Receiver constitutes the input to his Mor-
phemicizer. This transducer converts the discrete flow of phonemes back into a
discrete flow of morphemes, which is its output. In performing this transduction,
the Morphemicizer operates according to the morphophonemics of the language
(also the code by which the Phoneme Source operates), but, once again, ap-
plies it in reverse.

(6) The morphemes is fed back into Jack's G.H.Q.

resulting discrete flow of
In the above we have used "morpheme" and "phoneme" advisedly,
the terms
though without implying that the unit signals emitted by a G.H.Q. or a Mor-
phemicizer (morphemes) or by a Phoneme Source or Speech Receiver (phonemes)
can be identified in any simple one-to-one fashion with the similarly labelled
units which a linguistic analyst uncovers and describes in the course of his work
with a language. Within our theory, the aim of the linguist, in working on a
single idiolect, would be to determine just the units of internal flow to which we
have assigned these two terms; but the accuracy of his analytic conclusions can
only be tested indirectly, so that various types of discrepancy no doubt remain.
b A MANUAL OF PHONOLOGY

Possibly it would be better if we used, in the present context, the terms "gram-

matic units" and "phonologic units," since this at least would avoid the identi-
fication of the units of internal flow with the sorts of morphemes and phonemes
established by any one brand of grammatic or phonologic analysis. But, for the
sake of conciseness, we will use the one-word terms.
As so far discussed, our representation could be interpreted as an enlargement
and modification of Bloomfield's much simpler diagram:
S -> r s -> R.
Here "S" represents a stimulus on Jill; the arrow stands for obscure inner ac-
tivity of some sort within her, and "r" for her overt speech response to the stimu-
lus. The row of dots represents sound waves passing through the air from Jill's

mouth to Jack's ears (quite like our dotted arrow, in Figure 1). "s" represents
the speech stimulus on Jack; the second arrow represents obscure inner activity
in him; and "R" represents his non-speech response. According to Bloomfield,
that represented by "S," by "R," and by the arrows stands outside the strict
sphere of interest of linguistics, which is concerned only with the segment
"r s".
Our more complex diagram contains no analog for Bloomfield's "S" and "R,"
for a reason which has already been stated humans are not essentially different
:

from other animals in possessing sense-organs and muscles. We might, however,

say that Bloomfield's "S" and "R" are provided for by the indicated "back
doors" to Jill's and Jack's G.H.Q.'s: it is through these back doors (as we shall
see later) that the impulses pass which account for the partial correlation be-
tween what people say or hear said, and what they do. Our diagram pretends to
push back the boundary between what concerns a linguist and what does not,
by stealing some things from what Bloomfield represents with his arrows and
drawing boxes to represent them: everything covered by the five boxes we draw
within Jill and within Jack might be regarded as correlating with Bloomfield's
"r" and "s."
0212. Feedback and Monitoring. However, our representation provides for
certain features not covered by Bloomfield either in his diagram or in his discus-
sion feedback
: and monitoring. The Figure shows a second dotted arrow outside
of Jill and Jack, passing from Jill's Speech Transmitter back to her own Speech
Receiver. This is included for the obvious reason that Jill hears herself as she
speaks. Also, a solid arrow is drawn from Jill's Transmitter, close to her outer
boundary, directly to her Receiver Jill also feels the motions of her articulatory
:

organs as she speaks. There is in addition a solid arrow leading from her Phoneme
Source directly to her Morphemicizer, but this feedback route is more purely
hypothetic.
There is evidence to show that the first two sorts of feedback, respectively
auditory and importance in the act of speaking: a speaker
kinesthetic, are of great
constantly observes his output, and monitors it by these types of feedback just
as the monitor in a radio station listens to what is being sent out over the air
and makes minor adjustments in it as necessary. Joos describes a pair of "ex-
periments" which can test, separately, the importance of the two types of feed-
back named above. To test auditory feedback, one goes slightly deaf. A partially
 A MANUAL OF PHONOLOGY 7

deaf person often speaks rather mumblingly; or, if he manages to speak clearly
(without a hearing aid), he makes more than the usual use of kinesthetic feed-
back for monitoring purposes. To test kinesthetic feedback, one imbibes a cer-
tain amount of ethyl alcohol. Anyone who has done so reports that in order to
make one's speech sound right, it is necessary to articulate more slowly than
usual and to monitor one's speech by listening to it more carefully that usual. A
sufficiently drunken person does not do this, and the resulting slurred effect is
familiar to us all.

We may conclude, from such evidence, that under normal conditions one con-
stantly monitors one's speech via both types of feedback. Jill's Speech Receiver
isnot quiescent as she speaks, but is functioning just as are her Speech Trans-
mitter and her deeper units. Similarly, we may suspect that Jack's Speech Trans-
mitter is not completely quiescent just because at the moment he is broadcasting
nothing. As he listens to Jill, his Speech Receiver is able to decode the signal
partly because the incoming signal is constantly compared with the articulatory

motions which Jack himself would have to make in order to produce an acous-
tically comparable signal. There is only less direct evidence for this assumption,
but we can at least point out that in learning a foreign language one has con-
siderable difficulty hearing correctly until one can also pronounce correctly.
In our theory of the functioning of G.H.Q., developed below, we shall incor-
porate feedback as one of the key mechanisms.
0213. G.H.Q. as a Source. The functioning of G.H.Q. is complex, since it serves
both as a source of a discrete flow of morphemes, the destination of such flows,
and also, via its "back door," as the unit within which is brought about some
degree of correlation between what people say and hear and what they do. We
shall describe one part of this functioning at a time, beginning here with the
temporary assumption that G.H.Q. does nothing at all except emit morphemes.
A unit regarded as purely a source of a discrete flow of signals can be mathe-
matically characterized in a complete fashion on the basis of the statistics of the
signal-flows which it emits the technique for doing this was developed by Claude
;

Shannon. We imagine that G.H.Q. can be in any number of a very large number
of different states. At a given moment it is necessarily in one or another of these
states. Associated with each state is an array of probabilities for the emission of
the various morphemes of the language: a certain relative probability that the
morpheme and will next be emitted, a certain relative probability that the
morpheme tackle will next be emitted, and so on. When some morpheme is
actually emitted, G.H.Q. shifts to a new state. Which state the new one is de-
pends, in a determinate way (not just probabilistically) on both the preceding
state and on what morpheme has actually been emitted. There are vastly more
states than morphemes, since a given morpheme may bring about not only a
transition from state A to state C, but also that from B to C, or from D to E,
and so on noting, as already asserted, that a combination of preceding
specific
state (say A) and actually emitted morpheme (say tackle) results always in the
same next state (say C).
Given complete information on G.H.Q. states and transition-probabilities, a
complete mathematical specification of G.H.Q. as a source can be formulated

, ^
8 A MANUAL OF PHONOLOGY

in the shape of a square matrix, with n rows and n columns, where n is the total
number of states the entry pi3; in the ith row and the jih column, is the prob-
:

ability that when G.H.Q. is in the ith state it will next pass to thejth. The sum
of all the probabilities in a row is necessarily unity. To this mathematical speci-
fication we need only add an indication of what morpheme emitted in connec-
is

tion with each state-transition.

A tiny example is not hard to construct, though it does not show everything
that needs to be shown, and will thus have to be supplemented. Figure 2 gives
the matrix for an almost infinitesimally small G.H.Q. the "morphemes" emitted ;

by G.H.Q. are English words and phrases, plus one sentence-final intonation
this
(denoted by /./ in the Figure). Arbitrary symbols would do just as well mathe-
matically, but it is more satisfying to make the results look like English. Once
set into motion, this miniature G.H.Q. will emit, over and over again, one and
another of the following eight English sentences:
(1) Jones takes the ball. (5) Smith takes the ball.

(2) Jones takes him out. (6) Smith takes him out.

(3) Jones passes the ball. (7) Smith passes the ball.

(4) Jones passes the ball to Smith. (8) Smith passes the ball to Jones.

So Si s2 s3 St s, S6 s, 08 s S io

SQ .00 .50 .50 .00 .00 .00 .00 .00 .00 .00 .00
Jones Smith
Si .00 .00 .00 .60 .40 .00 .00 .00 .00 .00 .00
takes passes
& .00 .00 .00 .80 .00 .20 .00 .00 .00 .00 .00
takes passes
s 3 .00 .00 .00 .00 .00 .00 .90 .10 .00 .00 .00
the him
ball out
Si .00 .00 .00 .00 .00 .00 .00 .00 1.00 .00 .00
the
ball
sb .00 .00 .00 .00 .00 .00 .00 .00 .00 1.00 .00
the
ball
St 1.00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00

Si 1.00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00

Ss .40 .00 .00 .00 .00 .00 .00 .00 .00 .00 .60
to
Smith
s3 .60 .00 .00 .00 .00 .00 .00 .00 .00 .00 .40
// to
Jones
Sio 1.00 00 .00 00 00 .00 00 .00 .00 .00 .00

Fig. 2


A MANUAL OF PHONOLOGY 9

With the indicated transition probabilities, the relative frequency of each sen-
tence in the long run (and thus the probability that, starting from state So,
a given sentence will next ensue) is as follows: (1) .27, (2) .03, (3) .08, (4) .12,

(5) .36, (6) .04, (7) .06, (8) .04. Naturally these eight figures add up to unity,
since it is certain (or at least has probability unity), whenever the G.H.Q. is

in state So, that one or another of the eight sentences will be emitted before the
unit returns again to state S .

Another important measurement is that of the fluency or entropy of each of

the states: that is, the uncertainty, for each state, as to what the next state will
be. To measure this we make use of the function I(p), the informational com-
ponent of the probability p, where p is any of the transition probabilities. The
function I(p) is denned as
I(p) = p log 2 p f or < p < 1

= for p = or p = 1,
and is expressed in binits. The fluency of a G.H.Q. state is the sum of I(p) for
all the transition probabilities from that state to any other state; that is, in
terms of Figure 2, the sum of I{p) for all the probabilities in a single row of the
matrix. Computation shows the following values:
for S : 1.0000
&: .9709
&: .7219
S3 : .4690
D4j Sb, 06, S7, S10: 0.0000
#8 &:
, .9709.
It is tobe noted that the state of greatest indeterminacy is that, so to speak,
"between sentences." True enough, we so selected the probability figures as to
make this hold good. The decision to do this was based, however, on realistic
considerations: when we listen to someone speak, we can often guess pretty
accurately how he will finish a sentence when he has gotten far enough into it,
but are far more uncertain in such guessing if we try it at the beginning of a
sentence. Of course this statement is an oversimplification but so is our minia-
ture G.H.Q.
The average entropy per state of the model G.H.Q. of Figure 2 is .4640 binits.
If the G.H.Q. emits "morphemes" at the rate of n per second, then the entropy
is .4640n binits per second, or shannons; if n is, say, 6 (which is about the rate at

which a person produces morphemes in speaking English), the entropy is 2.7840

shannons. The actual entropy of English (and maybe of any real language) has
only been estimated by indirect methods, but comes out several times larger
say 5 or 6 shannons in normal rapid speech.
Our sample can be made more realistic if we pretend that the non-zero proba-
bilities in Figure 2 are all slightly smaller than there listed, and that some of the

zero figures are slightly greater than zero. We can suppose, for example, that the
G.H.Q. might pass, rarely, from S to #3, emitting (we shall say) the "morpheme"
takes, and then going on (with highest probability) to produce such a quasi-
sentence as takes him out /./; or, again, that the transition from $ 4 to So might

^
10 A MANUAL OF PHONOLOGY

occasionally occur, to result in a fragment like Jones passes (without the sentence-
closing /./, but nevertheless returning to state S ). With or without such modi-
fications, it is important to note that, although any morpheme-sequence which
can occur can also recur, it is also possible for a particular morpheme-sequence
to be emitted for the first time. The first occurrence of a morpheme-sequence is
suspiciously similar to what in everyday parlance we call "originality" or "in-
novation."
By making an enormous count of relative frequencies of occurrence of all the
morphemes and many morpheme-sequences in actual English, followed by an
enormous amount of computation, and by writing very small entries on an
enormous sheet of paper, the entire grammatical structure of the language
could be portrayed in just the style of Figure 2. No one is going to do this, but it

is important for us to consider (1) the assumptions which would be involved if

we did do it; and (2) the further assumptions which are involved in the sort of
grammatic analysis we do indulge in.
(1) For this, our assumptions are necessarily as follows:
(a) The number of states does not change with passage of time. This is cer-
tainly contrary to fact, since a child learns a language, and even an old person
may learn new morphemes. One is forced to assume in the first instance that a
linguistic system is static.
(b) The transition probabilities do not change with passage of time. Mathe-
matically, this means that the system is a stationary Markoff process. This, also,
is a contrary-to-fact assumption, just within the framework of our theory, as will

be shown below.
(c) Any state may recur, and has a greater-than-zero probability of recurring in

a finite amount of time. Mathematically, this means that the system is ergodic
in the weak or Wienerian sense. Linguistically, it is the assumption always made,
at least tacitly, that the actual occurrence of a particular string of morphemes
does not change the system in such a way as to reduce to zero the probability
that the same string will occur again. This assumption renders analysis descrip-
tive and synchronic; we know that it is false to fact, and when we abolish the
assumption we are working historically.
(2) In our actual grammatical analyses, we take a further and (from the
practical point of view) crucial step we replace zero probability, and all extremely
:

low probabilities, by impossible, and all higher probabilities by possible. (Per-

haps we keep some vestige of the continuous grading of probabilities when we
remark, of certain arrangements of morphemes, that they are "grammatically
possible but not very likely.") This greatly reduces the complexity of our de-
scription. In Figure 2 it is necessary to distinguish between *S4 and S6 for just one
reason: we wish to eliminate entirely the possibility that the G.H.Q. will emit
the utterances Smith passes the ball to Smith and Jones passes the ball to Jones,
because these utterances, within our small pseudo-system, are semantically im-
probable.But in ordinary description we should, in effect, identify Si and &,
and exclude the likelihood of the two utterances just cited by appealing to the
common sense of the reader of our description.


A MANUAL OF PHONOLOGY 11

In effect, then, ordinary grammatic analysis handles morphemes in batches,

classing together those the occurrences of which have approximately the same
effect in switchingG.H.Q. from one state to another. By the same token, G.H.Q.
states cease to be that, and become something like form-classes.

0214. G.H.Q. as a Destination. There is another key difference between con-

ventional grammatic analysis and the view of grammatic pattern required within
our theory: the former focusses attention largely on morphemes; our theory
focusses attention on the "inter-morpheme" states of G.H.Q., and morphemes
take very much of a back seat. But to show in just what way this works, we must
enlarge our description of G.H.Q. and show how it functions as a destination
for morpheme-flows.
When G.H.Q. emits a morpheme, a feedback report comes to the same G.H.Q.
telling what morpheme has been emitted. Figure 1 shows three feedback routes
for this, but it is perfectly conceivable that there is another one leading out of
G.H.Q. and back thereto without passing through any other unit. We assume
that it is the feedback report of what morpheme has been emitted which serves to
"trigger" G.H.Q. into its next state.
By assuming this, we can account for the behavior of a G.H.Q. as a destina-
tion. A morpheme arriving at Jack's G.H.Q. has exactly the same triggering effect
thereon whether it comes, ultimately, from Jill's G.H.Q. or, via feedback, from
Jack'sown G.H.Q.
Thus when Jill speaks to Jack, the effect of her speech is to lead Jack's G.H.Q.
into a series of statesand state-transitions paralleling more or less closely those
through which her own G.H.Q. passes as she speaks. Jack "understands" Jill

that is, communication is effected if this parallelism is close enough. The mor-
phemes which leave Jill's G.H.Q. and which, after a double coding and decoding,
finally arrive at Jack's, are merely triggers by virtue of which the migration
through successive states of Jill's G.H.Q. can "induce", as it were, a similar
migration of Jack's. This we take to be the essential nature of speech communication.
We can think of an individual's G.H.Q. as continually clacking from one state
to another, throughout the individual's life from the development of speech in
early childhood until death. A vast amount of this state-migration leads only to
an emitted morpheme flow which is turned back in, via feedback, without "break-
ing through" to the surface so that someone else may hear: we account, thus,
for "thinking in words." From time to time the morpheme flow breaks through
and is perceived by others; from time to time, a morpheme flow emitted by some-
one else's G.H.Q. arrives and takes control (complete or partial) of the state-
migration of the first individual's G.H.Q.

0215. The Semantic Link. There remains one essential factor to consider:
goings-on through the "back door" of G.H.Q.
It is fashionable to say that a human being, as he speaks, "encodes experience
into linguistic form." This more cute than true. It is apt to suggest that a
is

human being, as a speaker, works though at a more complex level essentially
like a speedometer, a sphygmomanometer, or a Geiger counter. A speedometer,
for example, has a single "sense organ" through which it constantly "experiences"
 ,

12 A MANUAL OF PHONOLOGY

a rate of motion, and this "experience" is continuously encoded into the position

of a pointer on a dial. The position of the pointer is a message, and the message
is
about something it has meaning.
Now it is of course true that some about something in this
of our speech is
same narrow sensewe report, let us say, on what
happening to us at the
is

moment of reporting. But this is by no means the only way in which speaking

takes place. Suppose one sees a cat. One may immediately say There's a cat like
the speedometer giving its report. But one may not. One may say nothing at
all about the cat (even to oneself, "thinking in words"). And it is possible to

talk about cats when there are none around. A sports announcer can give a run-

ning account of the football game he is actually watching like the speedometer;
but an author can write a story about things that have never happened and
probably never will.
This points up also the second major defect in Bloomfield's simple stimulus-
response diagram no matter how complex the obscure internal activity subsumed
:

by his arrows may be, the inclusion of his "S" and "R" still seems to imply that
what one says is necessarily immediately and deterministically related to the
individual's field of stimulation of the moment. The link between G.H.Q. and
the other, non-speech, goings-on in and around a person must for accuracy be
viewed as probabilistic.
The probabilistic linkage can be provided for at the back door to G.H.Q.
without venturing into the welter of complexity in an individual outside his
speech units (a complexity which is beyond our concern because it is much like
that in any non-speaking animal). We need only say that any internal or external

stimulus of an appropriate kind say the sight of a cat works through G.H.Q. 's
backdoor to increase all those transition-probabilities for transitions which are
accompanied by the emission of the morpheme cat (and those associated with
certain other morphemes, like pussy, meow, animal, feline, etc.). Such an increase
by no means guarantees that any of the transitions in question will actually occur:
G.H.Q. may at the moment be wandering through a series of states where the
probability of emission of cat is so low that the slight increase is not apt to have
any result. But at least it implies that, other things being equal, one is more apt
to say cat when a cat is in evidence than otherwise.
We must also provide for impulses passing through the back door of G.H.Q.
in the opposite direction. If someone says Look at that cat, the incoming mor-
phemes direct our G.H.Q. through a certain series of states, and those states have
a probabilistic (not deterministic) effect on what we do in a non-speech "prac-
tical") way: we may
but do not necessarily look for a cat. Quite similarly, if
the state-to-state migration of our G.H.Q. leads to the emission of the morpheme
cat, that morpheme, fed back into G.H.Q., may lead us to look around for a cat,

even if we have not uttered the word aloud. It is along these lines that pre-
sumably the entire matter of the impact of linguistic patterns on behavior
(as discussed, for example, by Whorf) can be provided for.

0216. Summary. Suppose that we went through the laborious process of

gathering the statistics and making the computations whereby the grammatical
 A MANUAL OF PHONOLOGY 13

pattern of English could be described with an n2 matrix. It would then be possible

for engineers to build a G.H.Q. of hardware
though a truncated G.H.Q.,
without any back-door connections. (The problem of designing the ancillary

speech units Phoneme Source, Speech Transmitter, Speech Receiver, and

Morphemicizer might prove technically much more refractory; but for our
purposes these could be replaced by some other sort of input and output channels
say keyboard.) This hardware G.H.Q. would speak and "understand" English.
It would emit understandable utterances, and from time to time (depending on
the statistics of its construction) would emit sequences of morphemes which had
never occurred before, either in its own life-history or in the English-speaking
world. That is, it would coin utterances which, linguistically, is a necessary
condition for creative or innovating activity. The condition is not sufficient: it
is also necessary for the new utterance to be of some use to someone, and one may
suspect that many mechanical G.H.Q. would be classed as
of the coinages of the
nonsense. Of course, so far as the mechanical G.H.Q. was concerned, neither
repeats nor coinages would be about anything, since we have not equipped the
device with a back door: our machine would be the archetype of the would-be
purely a priori philosopher.
If, next, we were able to build our mechanical G.H.Q. into a larger device which

had receptors and effectors something along the line of those found in all animals,
linking these with each other and, through a back door, into the G.H.Q., we
would have an even closer analog to a human would
being. This enlarged device
run a greater chance of saying things that had not been said before and that were
worth hearing.
But it would still have one crucial limitation as compared with humans: it
would have, as long as it operated, only the number of G.H.Q. states which were
originally built into it. This can't be the case with humans. As infants we have no

G.H.Q. states at all perhaps no G.H.Q. When we learn to speak we acquire at
first a very simple G.H.Q., which then becomes increasingly complex, and need

not cease to be restructured from time to time until the death of the organism.
Our model does not provide for any such "growth" under the impact of experience
(change of transition-probabilities is perhaps also "growth," but would seem to
be of a different sort). But mechanisms of a much less complex order have been
designed to "learn" new responses; the problem is certainly intricate, but not
necessarily insurmountable.
The model also does not provide for certain obvious facts about speech, be-
cause they are not exclusive to speech or to humans. What determines whether a
flow ofmorphemes is coded into overt speech or remains inside? One could depict
a switch somewhere in the diagram of Figure 1 the problem is not where the
;

switch is, but what turns it off and on. But we can also "at will" think of walking
or actually walk, think of slapping someone or actually slap him, and so on: the
problem of the mechanism for such decisions is real, but not necessarily linguistic.
Related to this is implicit trial-and-error: it is introspectively obvious that we
often "plan" what we are going to say aloud before saying it. In creating a poem,
one works with a retained pattern of stress and pitch distribution, and performs
14 A MANUAL OF PHONOLOGY

trial-and-error until a morpheme-sequence comes out which fits; in proving a

theorem, one works with an understood pattern to which a statement must con-
form in order to count as a proof, and works by trial-and-error until a morpheme-
sequence produced which does conform. Where are these "patterns" retained?
is

Where do we store the acceptable morpheme-sequence when it has been pro-

duced? Again, if we memorize a verse or a bon mot, where is the memorized
morpheme-sequence stored? But these questions can be paralleled by questions
which are not about language at all. Where (and how) does a dancer "store" the
sequence of innervations which, when "read off" from storage, will put his body
through the desired sequence of motions? Where (and how) does a Hon "store"
his knowledge of how to stalk a gazelle? It is wrong to ask questions of this kind,
however important they may be, purely in a speech context.
It is evident that a key feature of our model of a human being as a speaking
animal is that it evokes the mathematically manipulable notion of probability
in place of the mentalist's free will, mind, soul, or ectoplasm. However, it does
this only insofar as man's linguistic behavior is concerned. It should be recognized
that even if the experimentum crucis had been performed, showing that the
mechanism of human language is as here described, one would still not have
proved the non-existence of the mentalist's non-physical entities. In the light of
the extent to which language seems to be responsible for many of the differences
between humans and non-humans, it would only become somewhat more likely
that souls, if they exist, are shared by men with other animals, rather than being
the exclusive prerogative of homo sapiens.

022. The Design of a Language. The discussion of the preceding section (021)
supplies the frame of reference in which we can describe the design of a language
and the position of phonologic pattern within it.

A language is a complex system of habits, involving five interrelated sub-

systems. Three of the subsystems are central, the other two peripheral.
The three central subsystems are:
(1) the grammatic system: a stock of morphemes, and the arrangements in
which they occur relative to each other;
(2) the phonologic system a stock of phonemes (or phonologic units) and the
:

arrangements in which they occur relative to each other;

(3) the morphophonemic system, which ties together the grammatic and
phonologic systems.
Some investigators prefer to regard the proper sphere of linguistics as including
only the three central subsystems which we have just listed. I agree, to the point
of labelling them "central," but beyond that it seems largely a matter of taste
in defining terms; we shall not so restrict the coverage of the word "linguistics."
The two peripheral subsystems are:
(4) the semantic system: this associates various morphemes or sequences of
morphemes with certain things or situations, or kinds of things or situations, in
the world around us;
(5) the phonetic system: this is the code which governs the slurring of a dis-
crete flow of phonemes into sound waves and the recovery of the former from
the latter.
 A MANUAL OF PHONOLOGY 15

The semantic system resides in the probabilistic ties through G.H.Q.'s back
door.
The grammatic system is the economy of the signals emitted by G.H.Q. The

special assumptions which we make, for practical reasons, in ordinary grammatic

analysis have already been mentioned (0213).
The morphophonemic system is the code according to which the Phoneme
Source operates, converting its input (a flow of morphemes) into its output (a
flow of phonologic units); the same code governs the inverse operation of the
Morphemicizer.
The phonologic system is by the Phoneme
the economy of the signals emitted
Source. This has available only a certain stock of signals, and can emit them
only in certain arrangements to the exclusion of others. Whatever may be the
input to the Phoneme Source, it can emit signals only within the economy of
the phonologic system by which it operates.
The phonetic system is the code according to which the Speech Transmitter
converts its input (a flow of phonologic units) into sound waves; the same code
governs the inverse operation of the Speech Receiver.
We must discuss some of the differences between this conceptualization of
language design and certain more traditional views.
There is a traditional view which sees phonologic and grammatic units as
whole design of language involves
differing primarily as to size-level, so that the
but a single hierarchy: a morpheme consists of one or more phonemes; a word
consists of one or more morphemes; a phrase of one or more words; and so on.
The present view is radically different. Morphemes are not composed of phonemes
at all. Morphemes are indivisible units. A given morpheme is represented by a
certain more or less compact arrangement of phonologic material, or, indeed,
sometimes by one such arrangement and sometimes by another. If we call any
such representation a morph, then it becomes correct to say that a morph has a

phonologic structure that it consists of an arrangement of phonemes.
An analogy may help. In telegraphy, two short voltage pulses represent the
letter I. The letter does not consist of two short voltage pulses
a letter is an
indivisible unit in a writing system. But the representation of the letter in teleg-

graphy has a structure in terms of smaller recurrent units the representation
consists of two dots. Similarly, the English morpheme {boy} does not consist of
phonologic material. It is represented phonologically by a morph /boy/, which
indeed consists of an arrangement of smaller phonologic units.
Under this view, then, morphophonemics becomes not the way in which
different morphs, in different environments, "belong to" one and the same mor-
pheme, in the sense that elements belong to a class, but rather the complicated
code governing the phonologic representation of morphemes and morpheme
sequences.
In quite the same way, of course, we cannot say that a phoneme or other
phonologic unit is physically present in the speech signal: what is present
is a representation of the phoneme.
therein, susceptible of direct observation,
Neither morphemes nor phonologic units, then, are susceptible to direct ob-
servation. Both must be deduced from what is directly observable: the behavior
16 A MANUAL OF PHONOLOGY

of humans, be it speech or otherwise. Within our frame of reference, we are free

to use any sort of evidence which can be obtained. Normally we use actual
samples of speech, plus observations of accompanying behavior or descriptions
by an informant of the "meanings" of utterances, to build our portrayal of both
the grammatic and the phonologic systems, and use whatever we have learned
about either of the latter, however we can, to help us find out more about the
other. In this heuristic sense, there can be no objection to the use of grammatic
information for the elaboration of our understanding of the phonologic system,
any more than there is to the reverse. The logical validity of the procedure de-

pends only on how we conceptualize the results. If we have found words, as

grammatic units, and automatically decide that there is a phonologic mark of

some sort an open juncture, let us say between each two successive words in a
phrase, then we are misusing our evidence. But if on the basis of grammatic
words we merely decide that we must look for a more or less regular marker of
word boundary on the phonologic level, we are proceeding soundly. The frame of
reference which has been elaborated above should help to prevent misuse of
evidence.
The technique of descriptive study of a language becomes, in this conceptuali-
zation, very closely akin to cryptanalysis. The cryptanalyst has one or more
messages in some secret system, plus, in the most favorable (and rarest) case,
information as to the meanings of the messages: by comparing the two, he works
out the code. The linguist observes speech and other behavior, and by comparing
the two works out the complicated three-step code which correlates them.
Our frame meshes well with Hjelmslev's terminology of expression
of reference
and content understand that terminology: Hjelmslev's actual in-
(at least as I
tentions may be something else). Morphemes are units of content: analysis of
the content "plane" is grammatic analysis. Phonemes and other phonologic
units are units of expression. Content should not be confused with "meaning,"
which has to do with the way in which experience is encoded into arrangements
of morphemes. Nor should expression be confused with "sound," the latter
term referring rather to the physically measurable sound waves which constitute
the speech signal. It is also either a mistake, or a difference in terminology, to
think of morphemes as elements on the expression "plane": if this is merely a
difference in terminology, then it involves a use of the term "morpheme" for
our "morph." Hjelmslev also distinguishes between "form" and "substance":
the latter term has to do with what we have called the phonetic and semantic
subsystems of a language, while the former has to do with phonology and gram-
mar. Finally, Hjelmslev distinguishes between "pattern" and "usage." With
somewhat less certainty, I find it convenient to identify the latter with be-
havior, the former with habits (4). Only behavior can be observed: habits must
be inferred. The methodology of linguistic analysis is the methodology of making
such inferences; for phonology, it will concern us in 3.
The term "tactics," which I was using in this frame of reference until recently,
will not appear in this manual. Its coverage was that now assigned to the term
"grammar," while the latter term was used to subsume both tactics and morpho-


A MANUAL OF PHONOLOGY 17

phonemics. But the word "tactics" has unfortunate overtones, and there seems
no real need for a generic term to cover just "tactics" and morphophonemics;
consequently, the terminologic adjustment can be made.
023. The Boundaries of Language; By-Systems. The preceding two subsections
(021-2) leave one problem untouched. Granting that linguistic messages
become overtly observable in the form of a speech signal, we must nevertheless
recognize that not all observable sound-producing behavior of a human is to be
classed as speech. Some instances are so obvious as to preclude discussion:
coughing, laughing, hiccuping and the like. But then just where is the boundary
between sound-producing behavior which counts as speech and that which does
not?
Perhaps no hard-and-fast line can be drawn, but there are two criteria which
seem useful: discreteness of contrast and duality of pattern.
By discreteness of contrast we mean that no phonologically relevant contrasts
are of the continuous-scale type. In a language where voicing and voicelessness
are phonologically functional, a given bit of speech is either voiced or voiceless
not, structurally, half-voiced or three-quarters voiced. The opposite situation is

found, for example, in the dynamic scale for Western music: there no theoretic
is

limit to the fine shades of contrast of volume which may be used by an expert
composer or performer. The embedding medium of linguistic messages also shows
a continuous scale of dynamics, organized to some extent in any given culture:
one may speak softly, or more loudly, or more loudly still, or anywhere in be-
tween, with no theoretic limit to the fineness of gradation. But we distinguish
between this (applying to relatively long stretches of speech) and the sharper
contrasts of prominence between different smaller pieces within an utterance
which, in some languages, function as an accentual system (23). In general,
then, we find continuous-scale contrasts in the vicinity of what we are sure is
if

language, we exclude them from language (though not from culture).

By duality of pattern we refer to the fact that linguistic signals have both a
phonologic and a grammatic (morphemic) structure. Few other communicative

systems of the sort which are close enough to language that we might not
know where to draw the boundary between them and language proper have
such a duality of pattern. If music is a communicative system (and this is not
clearly the case), then we find ourselves unable to sort out any analog for pho-

nologic units and for grammatic units music has the sort of unitary structure,
with units of all sorts differing only as to size-level, which, as we pointed out in
022, has been assumed by some to be the case for language, but which we have
rejected.
A combination of these two criteria serves, at least in a practical way, to con-
struct a working boundary to what we shall call "linguistic behavior"; this does
not necessarily imply that the term "language" must be restricted in this way
for efficient use. Yet even these combined criteria leave certain marginal cases
untouched. These marginal cases can be referred to collectively as linguistic
by-systems: they derive from a language, and are closely related to it, yet are
not perhaps to be classed as language in the primary sense.
18 A MANUAL OF PHONOLOGY

One example is Mazateco whistle-talk. Mazateco has a complicated tone-

accent system; in whistle-talk, one whistles the melodic contours of an utterance,
retaining all the relevant features of stress, pitch, and rhythm, but leaving out
(necessarily) vowel color contrasts and all the consonantism. This whistle-talk
is used for quite complicated messages; it is more powerful than one might at
first imagine. African drum signals are reputedly derived from the tonal contours
of utterances in certain African languages in a more or less similar way, but one
suspects that there has been greater conventionalization.
Another example, widespread if not universal, is whispering. A whispered
utterance mocks the phonologic structure of the same utterance spoken in the
normal way, but almost always omits certain contrasts which are functional in
normal speech. The details depend on the language. But by classing whispering
as a by-system, one can describe the whispering habits associated with any one
language, and then not complicate one's phonologic analysis of the language as
one might have to if whispered utterances were thrown into the same analytic
hopper as ordinary speech.
A final example, also widespread, is singing words. English, as sung, is intona-
tionless, because the melodic contours of the music claim the machinery used in
ordinary speech for intonation. Certain styles of Chinese singing are toneless
in just the same way.
That such by-systems are in general easy for a speaker of a language to learn,
and that messages transmitted in them rather than in ordinary speech are usually
intelligible, are facts tied up with the redundancy of the speech signal, a point
on which we shall touch in 5 and 6. For the bulk of our discussion we regard
by-systems as not properly a part of the object of our analysis.

03. Common Core and Overall Pattern. Our diagrammatic portrayal of

Jill and

Jack (Figure assumes that they can understand each other: this in turn im-
1)
plies that there is no great discrepancy between Jill's idiolect (including its
semantics, grammar, morphophonemics, phonology, and phonetics) and Jack's.
In actual experience, we find some pairs of idiolects so close that we are hard put
to it to discriminate between them at all, whereas other pairs are radically
different. Close similarity implies mutual intelligibility, but a fair degree of
difference need not imply mutual unintelligibility. People manage to understand
each other even though they signal by different codes.
A detailed discussion of the implications of this fact lies outside the scope of
this manual, but brief mention is necessary. People whose idiolects are virtually
identical often understand each other despite the presence of a good deal of

external noise that is, sound of various sorts from various sources which strikes
the ears of a hearer along with the speech signal from a speaker. Such external
(or channel) noise sometimes renders communication difficult or impossible,
but often it does not. When it does not, it is because the speech signal actually
contains (as it leaves the speaker) a great deal more evidence as to what message
the speaker is transmitting than the minimum which the hearer must receive in
order to reconstruct the message accurately. Channel noise distorts and destroys
some of that evidence, but does not seriously impair communication so long as a
sufficient percentage remains undistorted.
 A MANUAL OF PHONOLOGY 19

Divergence between the codes of two people who communicate with each other
The reason why
via speech can be regarded as another sort of noise: code noise.
people can understand each other despite code noise is exactly the same as the
reason why channel noise, up to a point, does not destroy communication. The
speech signal which leaves one person contains, in terms of his own total code,
a great deal more evidence as to what message he is transmitting than the mini-
mum which he himself would have to receive to understand the message. Some of
this evidence is irrelevant for a particular hearer, but if a sufficient percentage of
it falls within the identical portions of the two codes, the hearer will still under-
stand.
The effect of the two sorts of noise on communication is entirely the same.
If, for a certain two idiolects, there is virtually no code noise, then communica-
tion is possible despite a relatively great amount of channel noise. If, for another
pair of idiolects, there is on the average a larger amount of code noise, then com-
munication is possible only in the presence of relatively less channel noise. All of
us have had experiences which show can follow almost any sort of English
this. I

even in a noisy cafeteria. At one time, I was able to follow Mexican Spanish quite
well providing I could hear it well enough, but was completely lost under condi-
tions approximating the noisy cafeteria.
Effective communication in the face of divergence of idiolects is due to two
factors. One is that the set of idiolects involved share certain features: the whole
set of shared features we shall call the common core of the set of idiolects. Barring
channel noise, speech in any one of the idiolects understandable to speakers
is

of all the others so long as it remains within the common core, while any mo-
mentary resort to the features peculiar to the speaker's idiolect and not shared

Fig.3 Fig. 4

Fig. 3. Two Idiolects with a Common Coke

Each circle represents an idiolect; the vertically hatched area represents their common
core.

Fig. 4. Three Mutually Intelligible Idiolects with No Common Core

Each of the elongated lozenges represents an idiolect. Each two share a common core,
represented by an area with vertical hatching, but the three taken together share nothing.
(There is, of course, no "scale" to the diagram: all that counts is regions, boundaries, and
intersections. Speaking mathematically, this and the other similar diagrams are topologic,
not metric.)
20 A MANUAL OF PHONOLOGY

by the others constitutes code noise. Even in the most homogeneous speech
community, there are some individual idiosyncrasies, and it happens from time
to time that someone says something that others do not understand.
In Figure 3 we show, in a very simple way, two idiolects which have a common
core. Theoretically, a set of three idiolects might be mutually intelligible to some
degree, without implying that the three, as a whole set, had any common core
at all. We show in Figure 4 how this is possible idiolects A and B have a com-
:

mon core; idiolects B and C also do; and idiolects A and C; but A, B, and C,
taken together, have none. If speech is produced in idiolect A, the speakers of
B and C may both understand, but on the basis of different portions of the whole

Fig. 5. Two Idiolects with a Common Core, Showing the Distinction between
Productive and Receptive Control
The left-hand circles represent A's idiolect the smaller circle for his sphere of productive
:

The right-hand circles similarly repre-

control, the larger for his sphere of receptive control.
sent B's idiolect. The hatched area is their common core; i.e., their shared spheres of pro-
ductive control. If something A says falls in region X, it is still perfectly understandable
to B, though B would not have said it; if something A says falls in region Y, it constitutes
code noise for B.

signal. In practice, however, this sort of situation hardly seems to be discover-

able. More typical isthe situation in which hundreds, thousands, or even mil-
lions of different idiolects share a discernible common core, which can be ob-
served approximately (via statistical sampling) and described subject to some
degree of indeterminacy.
But the common core of two idiolects is not the only factor responsible for
mutual intelligibility despite divergence. A given speaker may constrain his
own speech, largely or wholly, to the bounds of his own code, and yet be trained
to understand speech which falls well outside. In Figure 5 we elaborate the
representation of Figure 3 to show this. For each speaker, the inner circle marks
the bounds of his own (productive) idiolect, while the outer circle marks the
bounds of what he is trained to understand. Speech from A may thus fall outside
the productive idiolect of B without automatically constituting code noise: it
constitutes code noise only if it falls outside the larger circle for B.
Furthermore, we know that the boundaries of a given individual's productive
and receptive control are highly labile. A private uses the word sir oftener than a
 A MANUAL OF PHONOLOGY 21

commissioned officer, but the latter knows the word just as well as the former:
ifa private is commissioned, his productive system is slightly changed. Language-
learning never ceases. What stands outside an individual's sphere of receptive
control today may be within it tomorrow; what stands outside his productive
idiolect today may be within it tomorrow. In these terms, it makes sense to
speak of an overall 'pattern for any set of idiolects which are in direct or indirect
contact with each other and which contain a common core. The overall pattern
includes everything that is in the repertory of any one idiolect, productively or
receptively. It includes, typically if not by definition, more than any one idiolect,
while any one idiolect includes, typically if not by definition, more than the
common core. In Figure 6 we show three idiolects, their common core, and their
overall pattern, to make more graphic the statements just given.
two people to communicate without their idiolects sharing
It is possible for
any common core at all. Imagine a Frenchman who understands, but cannot
speak, German, and a German, who likewise has receptive but not productive
control of French. The situation is depicted in Figure 7 the circles representing
:

the (productive) idiolects do not intersect, but the larger circles representing the
boundaries of receptive control do. In such a case (or any rough approximation
to it) we do not speak of an overall pattern. It makes practical sense to talk of

an overall pattern only when there is also a clear common core.

Linguistic description can be based on observation of a single idiolect, or on

^^~^>',, '" l
^-" <!
<^

Fig. 6. Three Idiolects with Common Core and Overall Pattern

Here we use one idiolect, dashed-line circles for the second, and
solid-line circles for
dotted-line circles for the third. vertically hatched area is the common core of the
The
three; everything within the outermost boundaries (marked with parallel ticks) belongs
to the overall pattern.
22 A MANUAL OF PHONOLOGY

that of two or more idiolects; in the latter case under conditions which we would
describe conventionally by saying that the idiolects all belong to the same lan-
guage, or even to the same dialect. If description is based on observations of two
or more idiolects, then we are free to follow either a common core approach or an
overall pattern approach
both make sense, and neither is necessarily any more
"valid" than the other. Often enough, an investigator does not consciously and
intentionally chose between these two lines of interpretation; in the case of a
small speech community with an obviously high degree of homogeneity, this
usually does not matter. In the present manual we shall not be concerned with
the ultimate ramifications of the idiolect, common core, and overall pattern
frame of reference. We shall be concerned only with the analysis and classifica-
tion of phonological systems as systems, whether they be idiolectal, or common

Fig. 7. Mutual Intelligibility with No Common Core

The idiolects of A and B share no common core, and there is therefore no overall pattern;
yet each can understand the other at least some of the time.

core, or overall pattern. In the few cases where we know which sort of system is
being reported, we shall state the fact, but we regard it as of secondary impor-
tance.
This type of abstraction of a system, from the actual network of speakers and
speaker-relationships in which it is in one or another way manifested, is what

produces, as I see it, a purely descriptive (or "structural") approach, as over

against one which is purely synchronic but more than purely descriptive. In
other words, descriptive study is a phase of synchronic study, but the latter is

more inclusive than the former.

04. Outline. The sector of language design which is of concern to us in the

present manual can be delimited in the diagram of Figure 1 : everything between
Phoneme Source and Morphemicizer and Jack's Phoneme Source and
Jill's

Morphemicizer, exclusive of the named units. However, our discussion cannot

devote equal portions to all the features included within these limits.
As a preliminary task, we must describe in detail the structure of the observ-
able parts of the Speech Transmitter that is, the organs of speech (1). The
technical jargon of phonologic description is largely articulatory, so that this
 .;

A MANUAL OF PHONOLOGY 23

survey is essential. It is worthy of note that the terminology ordinarily used in

phonologic description is not completely articulatory. Some terms make reference,
at least in the first instance, to acoustic features of the sort that can be measured
quite accurately enough for linguistic purposes by the trained human ear: for
example, "high" and "low" with reference to pitch. It is also quite probable
that some often-used terms are understood among specialists only because their
meanings have been demonstrated by actual example. These are what Pike
calls "imitation-label" terms: one illustrates a variety of speech-sound, trains
one's audience to identify and imitate it, and then gives it a label. We shall see in
due course (4) that the somewhat hodge-podge nature of our initial descriptive

terminology does not or at least need not impair the accuracy of our work
for most of the purposes which we seek. Indeed, in 1851 Bohtlingk published an
excellent description of the phonology of Yakut, quite as reliable for linguistic
purposes as many recent reports, in which the "imitation-label" technique is the
only one used: he describes Yakut sounds in terms of the sounds of French,
English, German, and Russian which are presumably familiar to his readers
through direct experience. It has only been by a slow process that a more gen-
eral, and largely articulatory, frame of reference has become available, and in

large measure this frame of reference, however useful, is meaningful for any
individual only to the extent to which he has received face-to-face instruction in
general phonetics. Thus, the "imitation-label" procedure is by no means a last
resort in case of desperation: rather, in a sense it is the fundamental procedure,
to which any others are ultimately reducible.
In 2 we shall turn to our main task the elaboration of an empirically-based
:

typology of phonologic systems and subsystems, involving a thoroughgoing

immediate-constituent approach. In this we will spend little time on meth-
odologic quibbles, though some minor points of a methodologic sort will have
to be touched on. Then 3 will be devoted completely to the problem of meth-
odology. In 4 we discuss the problem of structure, pattern, and abstraction;
in 5, acoustic phonetics, and in 6, functional load and analytic norms.

The Organs of Speech and their Functions; Phonetic Notation. An utterance

1.

inany language involves a series of motions in certain portions of the body:

"The lungs, together with the diaphragm and the other muscles that control
them, the larynx, the vocal cords, the arytenoid cartilages, the pharyngeal
musculature, the faucal pillars, the velum, the uvula, the buccal walls, the lower
jaw, the tongue, the teeth, and the lips." Collectively we call these, by a sort of
metonymy, the organs of speech, or refer to the whole region as the speech tract.
The upper portion of the speech tract is shown diagrammatically in Figure 8
omitted are the lungs and the bronchial tubes below the larynx. The figure shows
a vertical cross-section along the plane of symmetry of the body (sagittal section)
This is often the only figure supplied in discussions of phonetics and will be
here, because of the cost of printing plates. Heffner gives a bit more, but the
best way anatomy of the speech tract
to acquire a real acquaintance with the
(short of dissecting a cadaver) is to consult good anatomy textbooks. Extremely
useful illustrations are to be found in Gray (Figures 360, 3, 5; 375; 393-4; 850-3,
 Fig. 8. Schematic Sagittal Section of Speech Tract (upper portion only)
Abbreviations: a.r. = t = tip; bl = blade; c = center; d = dorsum; r =
alveolar ridge;
root; ep = =
vocal cords (parenthesized because they are not visible; the
epiglottis; v.c.
figure shows only the entrance to the larynx).

24
 A MANUAL OF PHONOLOGY 25

856; 944-55; 989-93; 1006-7, 9; 1017-18, 20-23) and in Sobotta (Figures 348-
51, 373^; 381; 387-91; 457-60; 471-83; all in volume 2).
Nevertheless, for most purposes a sagittal section serves quite well. There is an
interesting reason for this. Almost known articulatory motions manifest the
all

same rough bilateral

sort of symmetry which characterizes the human body.
There no case on record in which a raising of the left side of the tongue, let us
is

say, takes place independently of and in contrast to a raising of the right side of
the tongue. The nearest thing to an exception is that in some languages the
front and one side of the tongue may be pressed against the roof of the mouth
while the other side is held far enough away that air can pass through. However,
in such cases it seems not to matter on which side one leaves the opening: some
speakers leave the passage on the left side, some on the right, and doubtless some
vary.
It is not hard to see why articulatory motions should show bilateral symmetry.
We must remembernot only here but throughout our discussion of the speech
tract that in their function as "organs of speech" these portions of the body
necessarily produce sound. A difference motion which cannot be
of articulatory
heard by others cannot be linguistically relevant. It is, of course, surprising in
some cases to discover what very slight differences can indeed be relevant, but
this does not alter the principle. Now the very fact of the approximate bilateral
symmetry of the speech tract implies that articulatory motions which differ only
as to "left-sided" versus "right-sided" will be practically impossible to dis-
criminate. One might equally well expect the tone of a flute to change as one
turns the main tube into various alignments with the mouthpiece. It does not,
which leaves the player free to adjust the alignment so as to fit the mouthpiece
optimally to his embouchure and, at the same time, to render the keys optimally
manipulable by his fingers.
Apart from bilateral symmetry, the speech tract as a sound-producing mech-
anism is a highly irregular and peculiarly-shaped region, as compared, at least,
with some apparatus that one might construct of buzzers and metal or glass
resonance cavities. This would lead us to expect that any phonologic system
would be a highly irregular matter. Yet every known system manifests some
variety of partial symmetry or balance; we shall see in the sequel how this comes
about.
We shall now work through the speech tract, from "south" (the lungs) to
"north" (the lips), and position of which each
specifying the varieties of motion
movable part is capable, insofar as these motions and positions are known to be
relevant in some language. As we proceed, we shall supply a rough notation for
phonetic transcription, placing the symbols, as is customary, between square
brackets. At the outset, certain terms, particularly "vowel" and "consonant,"
will be used in very loose senses, and should be so accepted, even though later on
some of them will be assigned more precise technical meanings.
11. The Lungs. In their relaxed state, when one is not speaking, the lungs are of
course not quiescent: they are drawing air inwards through the nose or mouth
or both, or releasing (or actively pushing) air out by the same routes. In almost
26 A MANUAL OF PHONOLOGY

all speech, the lungs are actively pushing, more or less forcibly, whether at a
given moment the air can pass out unimpeded or, at the opposite extreme, is

completely cut off by some closure in throat, or in mouth and nose. Since this is

the normal situation, no special mention is ordinarily made of it (in this manual
or elsewhere).
The which the lungs push, on the other hand, is often of
relative force with
importance. This governed by a rather complex musculature, including the
is

diaphragm and the exterior and interior intercostals. No one knows nearly as
much as should be known about the articulatory functions of these muscles the :

attention of most linguists has been focussed on the region from the glottis
northwards, and all too often the lungs have been dismissed with the casual re-
mark that they can do nothing but push and pull. One theory, by no means
proved, holds that (at least in English) successive pulses of the diaphragm corre-
late with larger rhythmic units, while the more rapid successive pulses of the
intercostals correlate with the rhythmic units we ordinarily call syllables. Be
this so or not, casual dismissal of the sub-glottal region is out of order; we set
aside more detailed discussion here only because the facts are still unknown.
In a few recorded instances, speech takes place during the intake of air. The
various different functions of the larynx (12) are more difficult to control dur-
ing inhalation than during exhalation (at least for speakers of Western lan-
guages, and probably for everyone) ; in particular, voicing sounds a bit rough or
"hoarse," and pitch differences are reduced and modified. Occasional speech
during intake occurs even in English, under various abnormal circumstances.
Less sporadic is the habit of some speakers of English of producing an assenting
"grunt," something like yeah, with inflowing air. Even this, however, is in some

sense marginal. But in Maidu there are reported to be two consonants, other-
wise much like b and d, which occur only before a vowel, and which, together
with the vowel, are pronounced with inflowing air. This occurs in the middle of a
word or sentence the rest of which is pronounced with outflowing air, and can
in no sense be regarded as marginal.
The business of breathing has to be carried on whether one is speaking or not.
This sets a rough outer limit to the length of the segment of speech which can
occur without pause, both to get more oxygen to the system and to get more air
into the lungs with which to speak. The term "breathgroup" has sometimes
been used, either for actual stretches of speech between successive inhalations,
or, more usually, for the kinds of stretches of speech which it is not normal to
interrupt for inhalation, and which are often separated by some slight pause
even if no breath is taken. The correlation between actual interruptions for
inhalation and any kind of phonologically relevant longer stretch (of a number
of successive syllables) is at best rather rough, and one can suspect that other
criteria for such segmentation are of more fundamental importance. This matter
will come up again in 210.

The Larynx. The organs of importance in the larynx or glottis are the vocal
12.
cords,which are capable of a rather wide variety of functions. In their quiescent,
non-speaking, state, the vocal cords are relaxed and relatively far apart, leaving a
 A MANUAL OF PHONOLOGY 27

passage so wide that air can pass through, in either direction, quite noiselessly.
(A sore throat, or other pathologic or unusual condition, can of course render
ordinary breathing "noisy" by impeding the passage of air just at this point;
linguistically, the least "noisy" passage of air in such a case counts as functionally
"noiseless.") Articulations further north in the speech tract which are accom-
panied by this relaxed position of the vocal cords (and usually by exhalation
from the lungs) are called voiceless. At the beginning
an English word such as of
heap, hand, hose, there is a brief moment lips and
of such voicelessness, with the
tongue in approximately the position for the vowel which is to come next. Such
a brief phase can be symbolized by [h] or, when there is any special reason for it,
;

by a symbol for a vocoid (roughly, a "vowel": see 151) accompanied by some

diacritic to represent the voicelessness. Thus the beginning of the English word
he might be represented as [hi . . .1 or, using capitalization as the diacritic for
voicelessness, as [Ii . . .]. The latter notation is awkward, but does have one
great merit: it circumvents the common tendency to think of [h] as a "spirant"
(1522). The articulatory phenomenon which we are here dealing with is not
a spirant; the best term for it is simply "aitch."
At the opposite extreme, the vocal cords can be drawn tightly together so that
no air can pass by. This is variously called a glottal catch, a glottal stop, an aleph
or a hamzah: symbol [?]. We shall use the first of the four terms listed. There is a
perfectly valid definition of "stop" by which [?] counts as one, but it is more
convenient to restrict the term in a way which excludes [?] (1521). In English,
we often begin an emphatic exclamation such as ouch! with a glottal catch
(though phonologically we usually say that it begins with a vowel). In many
languages glottal catches occur much more frequently than in English, and in a
greater variety of environments in utterances.
The vocal cords can be stretched taut, with little space between them, but
held flexibly enough that the passing airstream forces itself through, and, in so
doing, sets the cords into vibration, rapidly opening and closing in turn. This
vibration of the vocal cords is voice, and articulations further north which are

accompanied by voice are called voiced. In an English word like heap, the brief
initial voiceless phase is followed by a much longer voiced phase, which ends
only with the consonantal closure of the lips at the end of the word (the p).
We do not use a separate letter-sized symbol for voicing; instead, pairs of sym-
bols are in general available for articulations further north in the speech tract, one
of them for voiceless, and one of them for voiced. Thus [p] and [b] represent
the same articulatory motions and positions in mouth and nose, but the former
implies that the vocal cords are quiescent, the latter that they are in vibration.
In some cases there is only one symbol readily available, perhaps implying
voicelessness, perhaps implying voice, and a diacritic of some sort can be added
to indicate the other.
By varying the tension on the vocal cords during voicing, and the force of the
passing stream of air, one can vary the pitch and the volume of the voicing. Dif-
ferent individuals have pitch-ranges of from one to three octaves. There is no
simple measure for the range of volume variation, which is any case is not as
28 A MANUAL OF PHONOLOGY

great at the extremes of pitch as in the middle of the pitch range. A misunder-
standing often found among beginners in phonetics is the assumption that the
vocal tract produces no sound save when the vocal cords are in vibration. This is
of course quite false; there are momentary phases in utterances in many lan-
guages during which nothing at all is audible, but these are not coterminous with
the phases between successive periods of voicing. What voicing does is to supply
a musical note, the glottal tone, extremely rich in harmonics, which can then be
modified as to overtone structure and as to distribution of energy at various
frequencies by the conformation of the speech tract north of the glottis. When
this musical note is not involved, there are still, most of the time, cavity-friction

and friction at various local points, rich enough to carry most of the differences
which are carried during voicing.
Various notations are used for the indication of pitch and volume differences;
there is no need to present them until they are required in our typologic survey

(2).
The vocal cords can be drawn close together, and held taut, so that the pass-
ing air stream is set into local turbulence, and this turbulence can be accompanied
by slight voicing (murmur, voiced glottal spirant) or not (voiceless glottal spirant,
whisper). Words like heap, hand, hose are pronounced by some people with an
initial voiceless glottal spirant rather than simply with voiceless vowel.
The term "whisper" is also used to cover another mechanism: the vocal cords
are closed together, but the arytenoid cartilages behind them are spread apart
so as to afford a triangular passageway for air. The acoustic effect is much the
same as that of voiceless glottal spirant. In what is ordinarily called whispering,
the distinction between theinitial h and the following vowel of a word like heap

isapparently maintained in either of two exactly opposite ways: one may begin
with pure voicelessness, and tense the vocal cords so as to produce the vowel
with glottal spirantization, or one may do the reverse. The former is probably
more typical, since it is during the normally voiced stretches that whispering
sometimes breaks into slight (murmur-type) voicing. A medial h, as in ahoy
(spoken aloud, not whispered) is often pronounced as a glottal murmur: so-called
"voiced h."
In our survey of articulatory motions further north in the speech tract we shall
see some further ways in which the various glottal functions described above
can be combined with motions elsewhere.

13. The Pharynx. As in the glottis, a complete closure can be made in the
pharyngeal region, by drawing the root of the tongue back against the back
wall of the pharynx. Such a pharyngeal catch occurs in Arabic dialects, but usually
(if not always) only as one alternative pronunciation of a consonant which is more
typically pronounced in a different way: as a voiced pharyngeal spirant. A pharyn-
geal spirant involves, not complete closure in this region, but sufficient constric-
tion to set the passing airstream into local turbulence; it can be voiced or voice-
less. The usual symbol for a voiceless pharyngeal spirant is [h], and for a voiced

one, ["?].

A narrowing of the pharyngeal passage may not produce any local friction;
 A MANUAL OF PHONOLOGY 29

instead, it may simply modify the coloring of a sound in which one or another
oral articulation participates. In Arabic, many consonants, involving oral articu-
lations, come one with pharyngeal constriction (the so-called emphatics)
in pairs,
and one without. A diacritic, say a dot under the letter, suffices to indicate
pharyngealization. Nootka has both a voiceless pharyngeal spirant [h] and a
glottal catch pronounced with pharyngeal constriction, [!]; these pair off, struc-
turally, much as do the ordinary [h] and [?] of the same language.

14. The Velic. The term upper

"velic," as can be seen in Figure 8, refers to the
and rear surface which constitutes part of the bounding surface of
of the velum,
the nasal cavity, at the boundary between the latter and the throat. This is the
only structure in the nasal cavity which is capable of significant sound-producing
or sound-modifying motion, and its motions are strictly limited. At one extreme,
the velic can be closed against the back wall of the pharynx; this cuts off the
nasal cavity from the rest of the speech tract, so that any air passing in or out
must go through the mouth, and so that the nasal cavity functions as an un-
connected rather than as a connected resonance chamber. At the other extreme, the
velic can be pulled well away from the back wall, so that passage of air between
throat and nasal cavity is unimpeded. The size of the aperture can naturally
vary, and in some languages it is necessary to specify at least one degree of aperture
between the two extremes.
A cold in the nose can fill the rear portion of the nasal cavity with phlegm, im-
pairing the contrast between articulation with closed and with open velic this, as ;

everyone knows, makes ra's and n's sound much like 6's and d's. Colloquially we
are apt to say that the voice of a person with a bad cold sounds "nasal," but
this sense of the term "nasal" is almost the opposite of its sense as a technical
term. Technically an articulation with open velic is called nasal, and one with
closed velic is called nonnasal or oral. If there is an intermediate degree of aper-
ture, one speaks of slight versus strong nasalization, or the like.
For some articulations the contrast of velic position is irrelevant. A glottal
catch, for example, cannot be "nasalized," since while the passage of air is cut
off completely at the glottis the open or closed position of the velic is quite
inaudible.
In describing speech sounds, nonnasal articulation is usually assumed, the
presence of nasality being specifiedwhen necessary. In notation, many symbols
imply by definition the absence or the presence of nasality: thus [p t k b d g]
imply nonnasal articulation, while [m n n] imply presence of nasalization. Where
basic symbols imply nonnasal articulation and there are no special paired sym-
bols for the corresponding nasal articulation, a diacritic is used: thus oral [a]
but nasal [aj or [a].

15. The Oral Cavity. It is within the oral cavity that the greatest variety of
articulatory motion occurs. It is convenient to divide the whole range of ar-
ticulatory motions in the mouth roughly two classes: those which most
into
typically are involved in the production of vocoids, and those often involved in
the production of contoids. As indicated by the wording, the distinction between


30 A MANUAL OF PHONOLOGY

a vocoid and a contoid cannot be made completely in terms of what goes on in

the mouth. The two terms are acoustic or impressionistic rather than strictly
articulatory: a vocoid is a sound in which resonances of one sort or another seem

to be of primary importance, whereas contoids are sounds involving clearly audi-

ble turbulence in the airstream at one point or another in the vocal tract, or else
complete shutting-off of that airstream. Thus in the English word fat the vowel
in the middle is a vocoid, whereas the / and the t are contoids.
Two illustrations will show why the contrast cannot be made completely in
terms of the functioning of parts of the oral cavity. A b and an m (bat, mat)
are made with precisely the same motions of mouth parts, yet the first is a con-
toid, the second either a vocoid or else on the boundary between contoid and
vocoid. The difference is brought about, of course, by the different functioning
of the velic. More striking is the following example: hold the tongue in the proper
position for a prolonged buzzing z-z-z-z, and produce such a prolonged buzz.
This sound is a contoid. Now without moving any part of the speech tract north
of the lungs, a diminution of pressure from the lungs can eliminate the "buzz"
of this prolonged z-z-z-z, whereupon the result sounds like an obscure vowel
a vocoid.
Such examples demonstrate that the vocoid-contoid distinction must be taken
with a grain of salt, but do not destroy its usefulness as a rough basis for classify-
ing oral articulations.
151. Vocoid Articulations. Most vocoids in most languages can be described in
terms of three factors lip position, tongue-height, and tongue frontness or back-
:

ness (there is, unfortunately, no convenient short term for the designation of the
last of these three factors).
Lip position is specified along the scale rounded-unrounded or rounded- spread.
If we compare the articulation of the words he and who, we find the lips relaxed
and (relatively) spread for the former, but somewhat rounded for the latter;
for many speakers of English, the rounding for who increases during the pro-
nunciation of the word. The maximum difference in degree of lip-rounding in
English is not very great; in some languages it is much more prominent.
If we compare the vowels of bit, bet, and bat, we find that the lower jaw is held
progressively further away from the upper jaw, and that the front part of the
tongue, at least, is progressively further away from the region of the upper teeth
and alveolar ridge, moving downwards in the mouth. The different positions of
the lower jaw and of the tongue are related, and in a sense only the latter counts.
For what counts, in differentiating such sounds, is the shape of the oral cavity as
a resonance chamber; this shape can be modified by different degrees of aperture
between the front of the tongue and the roof of the mouth, and the wider or
narrower opening of the jaws is merely part of the mechanism by which this de-
gree of aperture is controlled. A similar kind of difference appears if we compare
sue, sew, and saw, though it is not so clear in this case that it is the front part of
the tongue that is held in positions to give varying degrees of aperture. But, in
both cases, the scale of contrasts involved is that of tongue-height, from high
(through varying intermediate heights, mid) to low.
 A MANUAL OF PHONOLOGY 31

The third scale of contrasts is harder to describe in accurate articulatory terms.

If we compare he and who, setting aside the difference of lip-position which has
already been noted, we find (at least for many speakers of English) that in he
the whole upper surface of the tongue, from dorsum to blade, is close to the roof
of the mouth, whereas in who only the dorsum is high, the remainder of the
tongue curving downwards and away from the roof of the mouth. In both he
and who we have high vocoids; where the front of the tongue, as well as the back,
is high we speak of a, front vocoid, but where the back is high and the front is not

we speak of a back vocoid. Intermediate between front and back are (varying
degrees of) central; it should be noticed that the adjective "mid" is arbitrarily
assigned to intermediate positions on the height scale, while "central" is, equally
arbitrarily, assigned to the front-back scale. A high central vowel occurs in the
speech of many Americans in the adverb just (as in / was just going not in the ;

the adjective with the same spelling, as in a just man). The back and central
portions of the tongue are fairly high for this vowel, but the tip is held some-
what further down.
This description serves for the front-back scale in the case of various high
vocoids, and perhaps for mid vocoids, but is not so certain for the low ones. It is
customary to speak of the vocoids of back, bock, saw, all low, as respectively front,
central, and back, and these assignments are perfectly usable, even if it happens
that in this case we are using an imitation-label procedure (04) rather than ac-
curate reference to articulation. In the writer's own pronunciation, the vocoid of
back has the tongue tip down, behind the lower teeth, and the central portion of
the tongue pushed somewhat upwards and to the front; in bock this upwards and
frontwards bunching is less, and in law there is a concavity instead of a convexity.
Other speakers may achieve acoustically equivalent results with other tongue
positions this, indeed, is possible, though less drastically so, for the other varie-
ties of vocoid which have been discussed so far. However this may be, the ter-
minology which has been presented is that ordinarily used, and for linguistic
purposes it serves quite well.
In addition to the three scales of contrast which have been described so far,
one often hears of a distinction between tense and lax. It is easy to demonstrate
this difference in English: hold the fingers on the bundle of muscles which is to be
found above and in front of the glottis, within the framework of the lower jaw,
and say bit, beat, bit, beat. For beat, one can feel a bunching and tension in the
bundle of muscles which is lacking, or at least much less pronounced, for bit.
The same difference can be detected with could and cooed; for some speakers it
can be detected with bet and bate, but for many or most it is harder to feel in this
case. There is some problem as to just what is accomplished, in the oral resonance
chamber and thus in the speech signal, by this tension. Presumably the tension
brings about a slightly different contour of the upper surface of the tongue, thus
changing the shape of the resonance chamber. In English, other factors also
differentiate the voiced vocoid parts of bit and beat, could and cooed, bet and bate,
and seems probable that the lax-tense distinction is secondary rather than
it

primary. The writer has found no reliable report of a language in which the dis-
 ;

32 A MANUAL OF PHONOLOGY

tinction plays a primary phonological role, though this has been suspected to be
the case for certain languages of West Africa.
There are not many purposes for which one needs on hand a great variety of
symbols for the notation of vocoids (differentiated in the ways so far described)
those engaged in some specific enterprise, such as dialect geography, where a
well-defined and elaborate set of symbols is needed, will devise one. Most of these
symbologies make identical or highly similar use of certain letters, and some of
this we can indicate here, [i e e se] normally represent front vocoids, without lip
rounding, from high to relatively low. [u o o] normally represent back vocoids,
with lip rounding, from high to low. [a] represents a low vocoid, sometimes fairly
far front (in contrast to [a] representing a central or back variety), but often
central. Small cap [1} and [u], or else the Greek letters [i] and [v], are often used
in addition to [i] and [u] if some sort of contrast has to be indicated within, re-
spectively, the high front unrounded and the high back rounded region usually :

in such a case the modified (or Greek) letters represent slightly lower or laxer
varieties than the ordinary letters. Mid and high central vocoids without round-
ing can be represented by front-vocoid symbols plus a diacritic, either a super-
posed dot or, in the case of "i", a bar: [i]. With rounding, the back-rounded
symbols can be used with similar diacritics. There are also available a couple of
special letters, [a] and [a], the former for a central or back mid unrounded vocoid,
the latter for a mid or high central unrounded vocoid. Two dots ("umlaut")
over a back-vocoid symbol represent a front rounded vocoid, e.g. [ii 6] two dots ;

over a front-vocoid symbol represent a back unrounded vocoid. For convenience

of reference, these symbols, and a few others, are presented in Figures 9 and 10,
the former for unrounded vocoids, the latter for rounded ones. Some of the sym-
bols appear in two or more of the boxes in these figures, because the purpose of
the figures is to show the ranges of value often currently assigned to the symbols,
rather than to supply a unique symbol for each position. As soon as an analyst
has discovered, even in part, how the vocoid system of a particular language
works structurally, he will normally simplify his notation, choosing, wherever
possible, "simpler" letter-shapes rather than more complex or alien ones, though
with some care to supply mnemonism in terms of traditional associations.

front central back

higher i i a ii ui

high:
lower i i i 1*9 i ui i x

higher e e a a e a y
mid:
lower e e e e a a e e a

higher se e A A

low:
lower ae a a a a a

Fig. 9. Unrounded Vocoids

 A MANUAL OF PHONOLOGY 33

front central back

higher ii u u
high:
lower ii u u u u u v

higher 6 < 6 o
mid:
lower 6 <t> O 03 6 5 o

low: o oe 5 a o a d

Fig. 10. Rounded Vocoids

1511. Modifications. There are not many further purely oral factors which can
participate in the differentiation of vocoids. One which can and sometimes does
is retroflexion the curling back of the tip of the tongue, towards the dome of the
:

roof of the mouth. Many speakers of American English (not in eastern New
England or certain other regions of the East and South) pronounce a word like
bird with a mid central or mid high slightly rounded retroflex vocoid. In Badaga,
all vocoids come in threes, differing fundamentally in the presence of no retro-

flexion, slight retroflexion, or strong retroflexion. An acoustic effect quite in-

distinguishible from the retroflex vocoid of bird is achieved by some speakers of
English not actually by a curling-back of the tongue tip, but by a peculiar con-
tour of the central part of the tongue, the tip being held down behind the lower
teeth. We can define the term "retroflexion" to subsume the result, whether
achieved in the one way or in the other.
Symbols which connote retroflexion include [gr], for a mid-central vocoid with
retroflexion, and [r] or an inverted [i]; the last is apt to be used in the rare case
that a language has also a trill (153) for which the [r] is being used. Where there
are more extensive contrasts between retroflexed and nonretroflexed, one makes
use of some diacritic not otherwise needed.
Various non-oral factors can distinguish vocoids produced with identical posi-
tions of the mouth parts. A vocoid is most frequently voiced, and the ordinary
vocoid symbols imply concomitant voicing unless there is special indication to
the contrary. But vocoids also occur voiceless, or whispered, or pronounced with
glottal murmur
not, of course, with glottal closure, although sometimes a vocoid
is produced with intermittent closure of the glottis too slow to be called a glottal

murmur. As we mentioned in 12, the h at the beginning of a word like he is for

most speakers pronounced without glottal (or other) obstruction of any kind:
that is, as a voiceless vocoid, not as a spirant. If we wish to use the symbol [h]
for this, writing say [hi .] (he), [ha
. . .] (hot), [ho .1 (hope), then "h" must
. . . .

be defined as representing a whole family of voiceless vocoids: before any given

voiced vocoid, it represents a voiceless vocoid of the same oral properties.
Vocoids can also be modified by pharyngeal constriction. As this involves pull-
ing back the root of the tongue, the position of the upper surface of the tongue is
likely to be somewhat different from what it is with the root of the tongue not
34 A MANUAL OF PHONOLOGY

retracted, and the difference in orally-produced coloring may be more prominent

than anything added directly by the pharyngeal constriction. In Arabic, for ex-
ample, a vocoid more or less of the [ae] type occurs after nonpharyngealized con-
sonants, and one more or less of the [a] type after pharyngealized consonants;
functionally, the two are the "same," the difference residing in the consonants,
yet in some instances it may be the orally-imposed vowel coloring which serves
as the main acoustic signal to the hearer, to identify the preceding consonant as
nonpharyngeal or as pharyngeal.
Nasalized vocoids are of course very common the mechanism has already been
;

described (14).
152. Typically Contoid Articulations. To
describe these, we distinguish between
different articulators, along the lower margin of the oral cavity, and between
different points of articulation, along the upper margin and we distinguish be-
;

tween closure, in which an articulator closes against a point of articulation so

tightly that no air can pass, and spirantization, in which there is a constriction
which sets the passing air-stream into local turbulence. If an oral closure is ac-
companied by velic closure, then we speak of a stop this is the only sense in which
:

this term will be used (cf. 12, 13).

The articulators which it is convenient to differentiate include dorsum, cen-
ter, blade of tongue (the frontmost upper surface), tip of tongue, and the lower

lip. Sometimes it is necessary to specify whether "blade" includes or excludes

tip; there is no convenient shorthand expression for the distinction.

The points of articulation are the velum (sometimes requiring subdivision into
front and back), the dome, the alveolar ridge, the backs of the upper teeth ap-
proximately at the edge of the gum, the edges of the upper teeth, and the upper
lip; occasionally the last two function together.
A combination of articulator and point of articulation constitutes a position
of articulation; these are labelled by a compound descriptive term, the first part
referring to the articulator, the second to the point of articulation: dor so-velar y
front and back dorso-velar, centro-domal, lamino-domal, lamino-alveolar, apico-
domal, apico-alveolar, apico-dental, apico-interdental, apico-labial, labio-dental y
and (labio-labial =) bilabial. The term "lamino-alveolar" may have to be sup-
plemented by the specification that the tip is, or is not, included in the articulator;
in either case, the reference to the alveolar ridge does nothing more than to specify
that it is included in the area touched or approached by the blade, since a laminal
articulation cannot involve merely a point or a thin transverse line along the
roof of the mouth as its "point" of articulation. When the lower lip meets both
the upper lip and the upper teeth, the situation requires lengthier description.
Other combinations of articulator and point of articulation, not covered in the
preceding paragraph, are impossible (apico-velar), improbable (labio-alveolar),
or simply so far unreported (lower teeth and upper lip). However, it should be
marked clearly that the listed positions of articulation are simply a small set of
points of reference. Some selection of them almost always suffices for the descrip-
tion of the relevant contrasts in any one language, but articulations in two dif-
ferent languages, both conveniently describable as, say, apico-alveolar, may
 A MANUAL OF PHONOLOGY 35

differ slightly in their most typical conformation and in the resulting acoustic
characteristics.

1521. Stops. Stops can be made at all the positions of articulation listed above,
and inone or another language all of these positions for stop-closure are known
to occur. In English we have dorso-velar [k] and [g], respectively voiceless and
voiced: these two are sometimes front (key, geese), and sometimes back (cool,
goose), but the difference is accompanied by difference of environment and is not
in itself distinctive. We also have typically apico-alveolar [t] and [d], and bi-
labial [p] and Lamino-alveolar stop closure occurs in English at the beginning
[b].

of check, jack, but stop closure in this position in English is never followed im-
mediately by a vocoid articulation or by silence, as are the others.
Various symbols are available for the indication of stops, voiceless and voiced
(or otherwise modified), in other positions of articulation. In the present mono-
graph a dot will be placed under "p" and "b" to indicate labio-dental stops,
under "t" and "d" to indicate apico-domal stops, and under "k" and "g" to
indicate back dorso-velar closure when there is a significant contrast with a more
front dorso-velar position; as an exception to this, "q" will be used instead of
"k" with a dot. A small crescent () above or below "k" and "g" will indicate
centro-domal; under "t" and "d" the same diacritic will indicate apico-inter-
dental when this contrasts with apico-dental or apico-alveolar. A superscript
"y" after "t" and "d" will indicate lamino-alveolar. Where other distinctions
have to be made, simple symbols will be chosen, supplemented by description.
Oral closures are subject to various accompaniments at velic and in pharynx
and glottis. If the velic is open, the resulting total articulation is not a stop, but
a nasal continuant (a vocoid): starting with [m n n] for bilabial, apico-dental or
apico-alveolar, and dorso-velar, the diacritics already described serve to dis-
tinguish other positions of articulation as necessary. Pharyngealization (13)
has already been mentioned, as have voicing contrasts. Another variety of articu-
lation occurs in glottalized stops. To produce a glottalized stop, the glottis and the
and some oral stop closure is made; the muscles of throat and
velic are closed
mouth then squeeze the air contained in this completely closed chamber, causing
pressure at all three closures. Typically the oral stop closure is released first,

the compressed air bursting past the releasing closure, and then the glottis is

opened. But various timings are possible, and various degrees of pressure before
the release. The notation for a glottalized stop is the symbol for a voiceless stop
with an apostrophe either over it or immediately to the right; we use the latter
(e.g., [p']), as easier to print.
Two other "styles of delivery," as it were, of stops can be described. In the
case of a voiced stop, it is possible approximately to reverse the muscular motion
involved in glottalization, so that even though some air is passing north through
the vibrating vocal cords, the chamber above them contains air which is being

rarified rather than compressed. A release of the oral closure is then followed by
a slight inward rush of air, giving a sort of popping sound. Such voiced stops are
called injectives; our diacritic will be that mentioned above for glottalized stops,
since the letter so marked will in this case always be that for a voiced stop.
36 A MANUAL OF PHONOLOGY

With a dorso-velar closure, a closure further forward in the mouth can be

made and the air in the chamber thus established can either be compressed or
rarefied; the latter variety of articulation, with the more frontward closure re-
leased first, is Such sounds are clicks. We use clicks in English
attested in speech.
in one or two interjections (tsk-tsk!) and in signals to horses, but they do not oc-
cur in ordinary speech. But in a number of languages of South Africa and a
couple in East Africa they do. A click may be voiced or voiceless and even
nasalized, since the status of the velic does not affect the possibility of the neces-
sary oral articulations.
Different from clicks are coarticulated stops: usually a simultaneous dorso-
velar and bilabial closure. In coarticulated stops there is no compression or

rarefaction of the air between the two stop closures, but simply the approxi-
mately simultaneous closing and opening at the two positions. These, also, are
apparently most widespread in Africa, but in this case perhaps mainly in West
Africa.

1522. Spirants. Spirants can also be produced at all the positions of articula-
tion which have been described above. For dorso-velar spirants the customary
symbols are [x] and [7], respectively voiceless and voiced; for bilabial spirants
and OS], for labiodental spirants [f] and [v]. The diacritics described earlier for
[if>\

more subtle indication of position of stops can be used with these symbols also,
except that the common "f" and "v" obviate any need for them in the labial
range.
But for the front part of the tongue further distinctions have to be made, not
applicable in the case of stops. In English both s (sick) and th (thick) are (nor-
mally) apico-alveolar, but the former is a rill spirant, the latter a slit spirant.
In a rill upper teeth or the
spirant, the front edge of the tongue closes against the
gum on both sides, leaving only a tiny opening through which the airstream can
pass. In a slit spirant, instead of this tiny opening there is a transverse slit. The
difference is comparable to that between water coming out of a hose and water
pouring through a horizontally wide but vertically narrow sluice-gate. For rill
spirants we have symbols [s] and [z], to which the positional diacritics presented
for stops can be added; for slit spirants the basic symbols are [01 and [6], to
which the same diacritics can be added.
It is also possible to make a lamino-alveolar rill spirant; many speakers of
English do this for their s, holding the tip of the tongue down behind the lower
teeth, and the acoustic effect can be quite indistinguishable from that of an
apico-alveolar [s]. Other lamino-alveolar spirants are not slit spirants but in-
volve close approximation of a whole area, from side to side and from front to
back, of the blade of the tongue, to a comparable area behind the upper teeth
and perhaps including the backs of the teeth. In contrast to rill and slit spirants,
these may be called surface spirants. Our English sh (she) and zh (vision) are
of this type; the symbols which will be used in this monograph are [s] and [&].
In British English, lamino-domal spirants usually occur where in the United
States we use a lamino-alveolar position; the symbols would be [] and [?], with
the underposed dot for the retracted point of articulation.
 A MANUAL OF PHONOLOGY 37

There is one more variety of articulation which occurs perhaps most typically
in spirants, though not in the languages most familiar to us: lateral articulation.
An apico-alveolar lateral spirant, for example, is in a sense the reverse of an
apico-alveolar rill spirant: in the latter the sides of the tongue close off the air,
the tip being so held as to leave a small passageway, while in the former the tip
is and the sides are held close enough to the sides
closed against the alveolar ridge
of the upper teeth to cause turbulence in the passing air, but not close enough to
cut off the air altogether. The symbol for a voiceless lateral spirant is [1], and for
a voiced lateral spirant [1] but the latter is also used for a nonspirantal lateral
;

(a variety of vocoid), so that one has to specify in words, or use a diacritic, to

distinguish the two.
Spirants are subject to most of the modifications involving action outside the
mouth which we have mentioned for stops. Nasal spirants seem to be very rare:
perhaps it is difficult, when the outflowing stream of air meets no obstruction at
the velic and can go through the nose, to obtain enough pressure behind an oral
spirantal closure to produce audible turbulence.Apart from this, pharyngealiza-
tion, voicing contrasts,and glottalization, are all quite feasible for spirants.
Glottalized spirants seem to be as rare as are nasal spirants, but one can find
no comparable articulatory rationalization.
Figure 11 displays the basic symbols for the stops, spirants, and nasals with
which we have dealt in this section (152).

centro-
apico-
apico- domal, back
bila- labio- apico- dental, apico- lami no- lamino- dorso-
inter- fronted dorso-
bial dental labial apico- domal alveolar domal velar
velar
dental dorso-
alveolar
velar

Stops vis P P t t t V V k k qk
vd b b d d d d* dy g 8

Nasals vd m mm n n n ny ji n ny 5 q 9

Spirants vis <t> f s s ss X x x

Rill :
']
vd (3 V z .J z I z y y 7
6

Slit:
j
6
M
S *J %

Lateral : \
I
M \

1 >j \
A> S\y

Fig. 11. Stops, Voiced Nasals, and Spirants

The symbols for voiced laterals are also used for laterals of the vocoid (non-spirantal)
type. There are no generally accepted symbols for apico-labials. In 2 we shall use the
apical symbols with the small curve placed beneath them (as, in this table, for apico-
interdentals), since there will be no ambiguity.


38 A MANUAL OP PHONOLOGY

153. The Vocoid-Contoid Borderline. We have already pointed out that the
boundary between vocoid and contoid not sharp (15). One can start with a
is

clearly contoid voiced spirant, and, by either relaxing the air pressure slightly or
by relaxing the tension with which the articulator is held against the point of
articulation, eliminate any local turbulence in the air and have left only a vocoid
of particular coloring. In many cases the precise degree of tension or pressure is

irrelevant in a language, and one finds a sound-type which occurs with varying
degrees of local turbulence but always with noticeable coloring of the vocoid
type. The Arabic voiced pharyngeal spirant is a sound-type of this kind: in fact,
only rarely does one hear the friction of the airstream passing through the con-
tracted pharynx.
For spirants a distinction is sometimes made between fricative and (relatively)
frictionless,and for some languages this is of importance. The former term im-
plies a greater amount, the latter term a lesser, of local friction. Labiodental
spirants typically have more friction than bilabial ones; apical rill spirants usually
have more than slit spirants a similar contrast can be made in the range of dorso-
;

velars.
Although voiceless laterals are usually spirantal, many more nonspirantal
that is, vocoid voiced laterals are reported in various languages than voiced
laterals of the contoid type. Our English I is vocoid rather than contoid. For
vocoid laterals several distinctions are worth noting. Dorso-velar laterals occur
closure in the center, but open passageway on both sides ; some speakers of Eng-
lish use this instead of an apical closure. Where the closure is apical, the point of
articulation can be dental or alveolar or domal ([1]), and in any case the position
in which the remainder of the tongue is held can be relevant. Some languages
contrast an apico-dental or apico-alveolar lateral of the vocoid type with an
apico-domal one. Other languages contrast an apico-dental or apico-laminal
lateral with a lamino-alveolar one ([A"] or [I] or [l y ]). Still other languages con-
trast two apico-dental or apico-alveolar laterals, one made with the middle por-
tion of the tongue held down away from the roof of the mouth, the other made
with the middle portion of the tongue held up close to the roof of the mouth,
giving a kind of "i"-color to the lateral. This last scale of difference is often re-
ferred to as dark versus clear. British English has a clearer lateral at the beginning
of a word like little and a darker one at the end the ; difference in most American
English is far less noticeable.
Vocoids produced with tongue retroflexion, like the American English [r], can
be rendered spirantal (and contoid-like) by either increasing the tension in the
tongue or the pressure of the airstream Mandarin has a type of [r] which differs
;

from the American English [r] largely in this way. Contoids made with apico-
domal or lamino-domal position of articulation involve tongue retroflexion, and
so the particular retrofiex coloring which we associate with the English [r] ac-
companies them even when they are clearly contoids rather than vocoids.
Finally, there is a variety of articulation, trilling, which it seems quite impos-
sible to classify as contoid or as vocoid save completely in terms of the environ-
ing articulations, in its occurrences in actual forms in specific languages. Most
 A MANUAL OF PHONOLOGY 39

common is the apical trill: the tip is held in a semirelaxed state, perhaps near the
teeth, perhaps near the alveolar ridge, and perhaps retroflexed, in such a way
that the outflowing airstream causes it to flutter rapidly back and forth, first
touching and then moving away from some point of articulation. Trills differ
in length; the shortest with only one flutter, is usually called a tap or a scrape
trill,

depending on the contour of motion. For most Americans the t of matter or

automobile is a voiced tap or scrape, not lasting long enough to qualify as what
we would ordinarily call a stop. Trills may be voiced or voiceless, or otherwise
modified by events outside of the mouth, but only rarely are such accompanying
events in contrast.
The "Bronx cheer" is often a bilabial trill; this must be distinguished from
made by pressing the lips tightly together and squeezing air
bilabial voicing,
through so that the lips vibrate rapidly and produce a sound of definite pitch.
Neither of these bilabial articulations, however, is reported as an ordinary event
in the process of speaking any language. Uvular trills are well attested, for ex-
ample in many varieties of French, German, and Danish. If the uvula is held just
too tensely, then instead of a trill one gets a spirant, but in many cases a language
has a single sound-unit which is sometimes one and sometimes the other, per-
haps depending on environment. No other types of trills are known to occur in a
language, though phonetic virtuosos can produce several.

16. Timing and Coordination. We have now completed our topographic survey
and yet some vitally important matters have not been men-
of the speech tract,
tioned. These concern not the more or less clearly isolable and individually de-
scribable functionings of the different parts of the speech tract, but the timing
and coordination with which various motions of different parts take place in the
process of speech. Speaking is a temporal process, a succession of articulatory
events in time, and any static survey of the speech tract misses many significant
contrasts which appear only by virtue of the arrangement of events in time.

161. Length. The simplest timing contrast is that of length. In many languages,
two utterances may be composed of exactly the same key articulatory motions,
yet be distinctively different because some articulatory position is held for a longer
time in one than in the other. The distinction may be made for a vocoid or for a
contoid: in Italian, fato has a relatively long [a] and a short [t], whereas fatto
has a shorter [a] and a longer [t]. The symbol is [] or [:] after the letter for the
longer variety: [fa*to] and [fa: to].
162. Transition and Release. Somewhat different is the relative timing of two
articulations or changes of articulatory function in two different parts of the
speech tract. In English, utterances like big, dig, get begin with voiced stops,
but the voicing begins quite weakly and increases during the holding of the stop
closure; in French, utterances like belle, digne, gare begin with stops which are
strongly voiced from the outset. In French, pas, tasse, cul begin with voiceless
stops, and voicing begins almost exactly as the stop closure is released; in Eng-
lish, pass, touch, catch begin also with voiceless stops, but the onset of voicing

for the vowels which follow is delayed for a perceptible length of time after the
40 A MANUAL OF PHONOLOGY

release of the stop closure, and a brief phase of voicelessness, sometimes involv-
ing some local turbulence of the air at the point of articulation of the stop, is

clearly audible. For this type of stop with audible voiceless release there is a
descriptive term: they are as-pirated stops; the French voiceless stops are un-
aspirated. The common symbol for aspiration is an inverted comma: [p'].

Exactly the same factor of relative timing makes the difference between mince
and mints for some speakers of English, particularly along the American east
coast (for many the two words are alike). For the [n] word, the tip of
of either
the tongue closes against the alveolar ridge, and the velic
opened the latter
is

often within the preceding vowel, if indeed it has not remained open from the
beginning of the word. Subsequently the velic is closed, voicing is stopped, and
the tongue is moved to the apico-alveolar rill position. If the velic is closed pre-
cisely as the other two changes take place, one has mince; if the velic is closed
as voicing stops, but tongue motion is delayed a bit, then one has mints.
163. Timing of Chest Pulse. In a great many languages (perhaps in all) the
pressure of air from the lungs not constant, but occurs in a series of pulses.
is

The timing of these pulses relative to other articulatory motions can be very
important. It may be that one of the main differences between an aim and a name
(for those who say them even in moderately rapid normal speech)
differently
is this particular timing factor: the pulse begins after the n in the first, but with
it for the second.

164. Affricates. The

recognition of one variety of articulation, the kind in-
volved in seems to turn largely on matters of timing, though the articu-
affricates,
latory details are by no means clear. In (i)t shouldn't), one has a sequence of
apico-alveolar voiceless stop, usually not audibly released, followed by lamino-
alveolar voiceless spirant; the beginning of the latter is accompanied by beginning
of a chest pulse. In a rapidly spoken (I'm going to hi)t y(ou), the last word usually
begins with a voiceless lamino-alveolar spirant, rather than with a y; this, with
the preceding t, which may often be assimilated to lamino-alveolar position, still
constitute two clearly successive articulations, even when the chest pulse for the
last syllable is very slight. But in (pi)tch(er) one does not have what can in any
sense be identified with these two versions of stop-spirant sequence; one has a
unitary articulatory motion which begins with a complete stop closure, very
short, in lamino-alveolar position, and passes immediately into spirantal closure
in the same position, also held for a very short time. Such a close-knit stop-spirant
sequence sharing a position of articulation is an affricate.
An efficient phonetic notation needs separate symbols for affricates, since they
occur so very commonly. The alphabet of the IPA has never supplied them, which
is an interesting reflection of the history of the IPA their original semi-phonem-
:

icized work was done under the very large influence of French, in which affricates
do not occur as unit phonemes, and of German, in which the same is probably
true though many affricates occur phonetically the presence of distinct affricates
;

enough weight to tip the scales of general IPA theory.

in English did not carry
If the IPA movement had taken place in the Caucasus, or in aboriginal north-
western North America, the story would have been otherwise. A uniform source
 A MANUAL OF PHONOLOGY 41

of unit symbols for affricates is to modify a spirant symbol with some constant
diacritic, say a "hat" (*). This is useful in the case of an affricate for which no
other symbol is available; for example, apico-interdental or apical-dental af-
fricates with slit-spirantal offglide are very rare, but do occur, and there are no
better symbols than those just described. But for apico-dental or apico-alveolar
with rill-spirantal offglide the symbols [c] and [3] are available; these with an
underposed dot serve for apico-domal. Lamino-alveolar affricates are most often
written as [<$] and [3], to which, once again, an underposed dot can be added for
lamino-domal. Affricates with lateral release can be represented by [X] and [X].
Affricates not involving some part of the front of the tongue as articulator are
much rarer, but the "cap" diacritic is always available if a symbol is needed.
Affricates seem to be subject to all the modifications by pharyngeal and glottal
functions which are found for stops: paired symbols are available to indicate
voiceless and voiced, and the diacritics used for stops can be used for other modi-
fications.

165. Glide and Peak Vocoid. Differences of timing are also fundamentally
(though perhaps not exclusively) involved in the distinction between glide vocoids
and peak vocoids. In bird, as pronounced in the Middle West, one has a retroflex
peak vocoid, a vocoid preceded and followed by articulations that last less long
and which produce less clearly audible acoustic effects. But in red the same retro-
flex vocoid is pronounced quite quickly, and with the chest pulse only half-
strong, followed by a non-retronex vocoid which lasts longer and for which the
chest pulse has gained full strength; the initial [r] in this case, then, is a glide
vocoid. The same difference applies to the lateral vocoid in the second syllable
of battle and that initial in let, or to the high front vocoid in bit and the y of yet.
One can prolong the y of yet and obtain what sounds like ee-et or ee-yet, or pro-
long the r of red and get what sounds like rr-red or rr-ed. (This is a rather artificial
phonetic experiment, of course, but well worth performing in articulatory analy-
sis.) In parade some speakers have a clear nonretroflex vocoid before the glide
vocoid [r], whereas others have simply a peak vocoid [r], contrasting with the
glide vocoid [r] in, say, pray. Much the same contrast appears in are (with glide
vocoid[r] after the a) versus a conceivable ah-er 'one who says "ah" all the time'.

The development of phonetic understanding and notation has been such that
some of our symbols for vocoids imply glide vocoid unless specially marked,
whereas others imply peak vocoid unless the contrary is indicated, "r", "1",
and symbols for nasal vocoids like "m", "n", "rj", all in the first instance imply
glide vocoids (verging on contoids, as was indicated in 150); to indicate a peak
vocoid, some diacritic may have to be added. Common for this is a small circle
under the letter. Vowel letters like "i", "a", or "u" imply peak vocoid unless
specially marked: a small curve, concave downwards, under the letter, is often
used (e.g., [i]). But for what might be indicated with [i], [u], and [y], special letters
are available: [y q]. w
Outside of the United States the letter "j" is far more widespread than "y"

unrounded glide vocoid, and "y" is used for a high front rounded
for the high front
peak vocoid (instead of the symbol [iij). The writer tends to prefer the latter
42 A MANUAL OF PHONOLOGY

usage, but in this monograph we shall follow the more prevalent current Ameri-
can practice.

166. Timing in Longer Stretches. The above paragraphs cover the most im-
portant ways in which the factor of timing and coordination makes differences
between short stretches of otherwise highly similar articulatory movements.
Over longer stretches, the same factor also plays a part, but by and large in less
subtle ways, easier to describe. One example will suffice here. In answer to the
question Have you been to the store? one might say Yes, and I bought a cake. In
answer to the question Is there any soap? one might say Yes, I just bought a cake.
The terminal portions of these two utterances, from bought on, do not sound
alike. In the first, the pitch is of moderate height through bought a, then jumps
upwards at the beginning of cake and falls during the word cake to quite low.
In the second, the pitch is high at the beginning of bought and falls to quite low
during the whole sequence bought a cake, more rapidly during bought and then a
bit more slowly. In the first, cake begins with a slightly stronger chest pulse
than bought; in the second, this is reversed. As can be seen, much the same articu-
latory motions are involved, even to the changing rates of vibration of the vocal
cords and the strength of the successive chest pulses but the precise sequence in
;

time of these last two factors makes an essential difference between the two
fragments. It is a little more difficult, but perfectly possible, to illustrate whole
utterances which differ in just such ways.
The matter of the overall rhythmic effect of longer stretches of articulation
cannot be overlooked, and yet it is exceedingly difficult to cover in the sort of
essentially timeless topographic survey with which we have been primarily con-
cerned. Rhythmic differences result from different temporal sequences of chest
pulses of varying strengths, from varying speeds of articulatory motion, from
varying arrangements of vocoids and contoids in sequence, from varying pitch-
arrangements, and so on. The two examples contrasted in the preceding para-
graph differ in rhythm, resulting from the factors there specified. Languages
differ very greatly in their characteristic rhythms, a matter of some importance
when it comes to the problem of establishing a taxonomy of phonologic systems
(2). For example, it has been pointed out that while English and Spanish both
have distinctive contrasts of stress, in English there is an essentially stress-
timed rhythm, whereas in Spanish one has a syllable-timed rhythm. This means
that in English higher stresses in an utterance are normally separated by about
equal lengths of time, no matter how many syllables with lower stresses may
intervene Find a board for me and Interpret this poetry for me each have three
:

higher stresses, and the interval of time between find and board in the first is
not much shorter than the interval between -ter- and po- in the second. In
Spanish, on the other hand, an utterance of ten syllables takes approximately
twice as long to say as one of five syllables, regardless of the number and location
of higher stresses in either.

2. A Typology of Phonologic Systems. When a particular language is subjected

to phonologic analysis, we find in due time a stock, not only finite but fairly
 A MANUAL OF PHONOLOGY 43

small, of ultimate phonologic constituents, each one a "target area" of articulatory

motion, such that any utterance in the language, short or long, consists of some
arrangement of elements selected from the stock. In English (and in a great many
languages, though not in all) momentary closure of the lips is one ultimate

phonologic constituent: in the sentence Pay the man, Bill, this particular ele-
ment occurs three times, once at the beginning and twice medially, variously
environed by other elements. The stock of ultimate phonologic constituents, and
the arrangements in which they occur relative to each other in utterances, con-
stitute the phonologic pattern of the language. As we pass from one language to
another, both the stock of elements and the occurrent arrangements vary, in
many different ways. Articulatory motions which are not in the stock of a par-
ticular language may nevertheless occur sporadically; they represent, so to speak,
"bad aim" at a "target area" and are normally compensated for by the hearer
without conscious effort.
Ultimate phonologic constituents do not occur in an utterance as the individual
bricks occur in a row of bricks. Rather, they occur in clusterings, these occur in
still larger clusterings, and so on, up to the level of the whole utterance. That is,

the phonologic structure of an utterance shows a hierarchic organization, in-

volving units of various relative size-levels: the units at any size-level save the
smallest consist of arrangements of units of the next smaller size-level. In
theory, the patterns involved in such hierarchic structures can be described in
either of two ways: beginning at the lowest level (the ultimate phonologic con-
stituents) and working up the scale, or beginning at the highest level and working
down. Either choice entails certain difficulties, but we shall choose the second
procedure, since it seems to afford a rather better basis for an effective com-
parison of different phonologic systems.
In the following we shall have occasion constantly to cite forms or parts of
forms from various languages, in a notation which in each case is supposed to
show only the phonologically relevant factors. Following current custom, we
shall in general enclose such notations between slant lines (/. ./). The symbols
.

used in this way are for the most part those introduced in 1; any others are
described as they occur. We cannot call this notation "phonemic," at least for
the present: the term "phoneme" will not be officially introduced until 242.
However, the distinction between the sort of notation indicated by slant lines
and that indicated by square brackets is roughly the distinction customarily
shown in this way in current linguistic work. In centered tables, slant lines will
be omitted.
21. Pause and Macrosegment. When we decide to proceed, in our survey,
from large to small rather than vice versa, we force on ourselves at the very
outset the problem of determining how far upwards the hierarchic pattern goes.
How long, in other words, can a speech-event be and still count as a phonologic
unity of some sort? Since the vast bulk of linguistic investigation so far has been
we cannot be sure
concerned primarily with relatively short stretches of speech,
that we answer to this question. That is, we cannot be
shall find the definitive
certain that the larger unit which we shall take as point of departure for our
44 A MANUAL OF PHONOLOGY

large-to-small survey is actually the largest unit of phonologic relevance. All

that we can do is to make a tentative guess the alternative is to abandon the
survey altogether.
If a single individual proceeds from silence to speech, speaks for a relatively
short time (say a second or so), and relapses into silence, we can be reasonably
sure that he has produced some sort of phonologic unity. If the duration of the
speaking is considerably greater, then very often it includes perfectly clear
articulatory pauses. In a good many of the languages which have been investi-
gated in sufficient detail, that which occurs between successive pauses (or be-
tween the beginning of the utterance and the first pause, or between the last
pause and the end of the utterance) has very much the kind of internal organiza-
tion which is to be found in a whole very short utterance with no internal pauses.
We therefore define such a stretch of speech to be a macrosegment, and say that
an utterance in such a language consists of a sequence of one or more macro-
segments. We shall assume that something much like this is the case in every
language.
In English, a macrosegment is known to consist of two immediate constituents.

One of an intonation; for the other there is no good term, and we will
them is

call it simply a remainder. There are a relatively small number of different in-
tonations (certainly less than a hundred, though the precise number is not
known), but a very large number indeed of different remainders for all intents
and purposes the number is transfinite. In a moment we shall deal with English
intonations in greater detail, showing their internal hierarchic organization.
The characteristic features which mark the end of an English intonation supple-
ment pause as a mark of the boundary between successive macrosegments, and
may very well in some instances replace it; the structural unity of the macro-

segment is none the less well marked perhaps even better and the validity of
the term is reinforced rather than weakened.
It is not known whether all languages have this same binary structure for
macrosegments. Many reports on different languages pass over the matter of
intonation in complete silence. A few specifically state that there are no in-
tonational differences which can be subsumed within the description of the
linguistic system, even though there are ups and downs of pitch (or of volume)
which seem to be semi-organized culturally, at least to show some correlation
with speaker's mood. Since detailed and effective intonational analysis is rela-
tively recent, statements of the kind are not to be trusted; more thorough work
with such languages may reveal full-fledged, if simple, intonational systems. If,
indeed, there are languages in which no distinctive intonational differences are
to be found, then this affords us a typologic criterion.
Not all utterances in a language conform neatly to the macrosegment-pause-
intonation-remainder scheme. Almost always one is forced to recognize that
some utterings are broken off before they reach a normal boundary between
macrosegments. If a man is shot, or has to sneeze or hiccup, in the middle of a
sentence, it is easy enough to regard the linguistically relevant event as having
been cut off by an intrusive agent, and to discard the particular event as irrele-
 A MANUAL OF PHONOLOGY 45

vant for linguistic analysis. But normal process of speaking, such inter-
in the
ruptions, or medial hesitations, occur with no obvious intrusive agent, and these
events cannot be regarded as unpatterned. There are even hesitation-forms,
varieties of speech-sound made while "pausing to think," which differ greatly
from language to language. These matters are not marginal in any statistical
sense: they occur very frequently, though more in the speech of some individuals
than in that of others. If we seem to disregard these for our analytic purposes, it is
because they are more easily described in terms of uninterrupted speaking as a
frame of reference.
If we were proceeding in accordance with what some investigators would call a
"completely objective" approach, then we would not be able to set broken frag-
ments aside, since we should have no "objective" way to determine whether a
given bit of speech was a broken fragment or was "complete." But there seems
to be nothing to recommend this degree of so-called "objectivity": in the end,
it can at most supply us with exceedingly complex means of obtaining results

which can more quickly and just as accurately be obtained by taking into
consideration a certain amount of the behavior of people other than their speech
behavior in the narrow sense. When we listen to a language we do not know, we
often cannot distinguish between fragment and complete utterance; but if we
know the language, we can, and if we do not, someone who does know the
language can usually tell us. Actors often have to try to deliver lines which in-
volve fragments, and one of the most difficult things in the realistic recitation of
the lines of a play is to make fragments sound the way they naturally do. As a
speaker of the language, the actor has all the drive towards utterance-"closure"
that any native speaker does, and the intonational and other mechanisms which
produce such "closure" are very hard to avoid when trying to deliver any pre-set
utterance or part of an utterance.
211. English Intonation. English intonations are built out of seven ultimate
phonologic constituents, which occur only in certain arrangements relative to
each other and relative to the remainder of the macrosegment. There are three
'pitch levels (for short PLs) three terminal contours (TCs) and one feature of the
: ; ;

all-or-none type which we shall simply call "extra height." (The number of PLs
is usually put at four rather than three, but the topmost of the usually-enumer-

ated four functions in a different way; we provide for it with our "extra height.")
The PLs are of course relative, not absolute in terms of cycles per second: a
man's voice produces the three in a lower register than a woman's, and the same
individual speaks now in one "key," now in another, with varying intervals
between the PLs.
We shall represent the three PLs with superscript numbers, from /*/ for the
lowest through/ 3 / for the highest. One of the TCs is neutral, involving simply
articulatory pause (which can be extremely short), without any rise or fall in

pitch; this one we shall write with a vertical line /{/ which, because of the sym-
bols we choose for the other two PLs, can be taken as an arrow shaft without
any point. A second involves normally a lowering of pitch from the last PL in
the macrosegment, accompanied by a relaxation of force of articulation
 : :

46 A MANUAL OF PHONOLOGY

sort of "dying-away" effect. This one we shall represent by an arrow pointing

downwards: /[/. The third involves a rise of pitch from the last PL; the amount
of rise varies a good deal without contrast, and is accompanied, not by relaxation
in force of articulation, but either by relatively steady force or, perhaps, by a
slight increase. Our symbol is an arrow pointing upwards: /|/.
The above are illustrated by such examples as the following (where we tran-
scribe only the PLs and TCs, using conventional orthography otherwise)
(normal colorless statement:) 2
It's three o ni clock[
2 ,i3
(question:) /s it three o clock]
32 32
(a quiet call:) John\ or Johri\
(concessive, but with objections to be added:)
31
Fest
(quiet approval of a question indicated by gestures rather than by words :)
22
|
31
Fea/i| i
IVs an ash tray\
The parenthesized comments are not intended primarily as descriptions of the
meanings of the intonations; rather, it is intended that by following the clues
afforded by the comments, and also trying to follow the intonational marking,
the English-speaking reader can stand a good chance of correctly identifying
the intonations which the transcriptions are supposed to represent.
It will be noticed that in each of the transcribed macrosegments two PLs are
written together somewhere in the stretch. This is a purely graphic convention,
and must be interpreted as follows: (1) The syllable where the two PL numbers
appear is the center of the intonation; this syllable is at least slightly more
prominent than anything before or after it. (2) The first of the two PL numbers
which are written together at the center represents the PL on which that par-
ticular syllable begins. (3) The second of the two numbers represents the reference
PL for the end of the macrosegment. This PL may be actually reached at the end
of the macrosegment, or it may be reached slightly before the end, the remaining
changing contour of pitch depending on the TC which is written at the end of
the macrosegment. It is therefore necessary to discuss conjointly the last PL
of the macrosegment (that is, the second of the two written together at the
center) and the TC
In the combinations /2|/ and /3|/, the pitch of the voice reaches the indicated
PL and dies out on that level.
In the combination /lj/, the pitch of the voice reaches the indicated PL
(usually, but not always, having fallen down to it from a different PL at the
center), and continues falling even lower, together with a dying-away of force
of articulation.
In the combinations /1|/, /2|/, and /3|/, the pitch of the voice reaches the
indicated PL and is cut off abruptly, without any dying-away.
In the combinations /It/, /2|/, and /3|/, the pitch of the voice reaches the
indicatedPL and then rises, the amount of rise apparently being non-distinctive.
Our notational conventions, as just in part explained, largely imply the ar-
rangements in which the constituent intonational features (PLs and TCs) are
to be found in English intonations. In a macrosegment there are two, three, or
 :

A MANUAL OF PHONOLOGY 47

four positions for distinctive occurrence of a PL, never fewer and never more.
There is always one at (or, phonetically, near) the end of the macrosegment,
and there is always a TC at the end. There is always another PL at the center.
There is always a PL at the beginning of the macrosegment, but this may coin-
cide with the center, in which case we have a two-PL intonation. If the center
and the beginning are distinct, we may have a three-PL intonation; or, as the
last possibility, one PL may occur somewhere between the beginning and the
center, giving a four-PL intonation.
The following examples show the same intonation and the same remainder,
but with the center variously placed:
2
I want to go n there[
2
I want to 31 go therel
2
I u want to go therel

The following is the same insofar as it can be, but as the center is at the beginning
there are only two PLs:
31
I want to go therel

The following examples contrast a three-PL and a four-PL intonation, on

the same remainder and with the center fixed
2
Pve been here five n minutesl
2
Fve been here 3five 31 minutesl

Such a fourth PL (second from the beginning, in terms of linear order) is neces-
sarily different than the PL at the beginning of the macrosegment; the syllable
on which it occurs is somewhat more prominent than those before and after it,
save, of course, for the syllable at the center. The PL at the beginning of a macro-
segment, when it is not also the center, implies no extra prominence of the
remainder at that point.
In addition to the limitation on sequence which requires that the first and
second PLs of a four-PL intonation must be different, there is another limitation
of importance: the last PL of the macrosegment is never higher than the one at
the center. This does not mean that the pitch of the voice cannot rise between
the center and the end, but when it does, the rise is interpreted as /f /, not as a
higher PL at the end. Thus in the notation of

2
Is it three o'^clock^

the the two / 3 /'s written at the center defines the pitch at that point,
first of

while the second, together with the /f/ at the end, mean: "pitch rises from the
center to the end, reaching a point higher than / 3/." The /\/ means "higher
than," and the second J 3 / specifies "higher than what." Thus in /33f / there is
steady rise from the center to the end; in /32f/ there is a dip from /3/ to /2/
and then a rise; in /31j/ and /2lf/ there is a dip down to /l/ and then a rise.
Taking these systematic limitations of sequence into consideration, we should
expect eighteen different two-PL intonations (/ll 21 31 22 32 33/, each with
48 A MANUAL OF PHONOLOGY

any of the three TCs); three-PL intonations; and 108 differ-'

fifty-four different j

ent four-PL intonations (the first two PLs being limited to /12 13 21 23 31 32/)
Not all of these actually occur. Most, or perhaps all, of the two-PL combinations]
n(

are attested, but some of the theoretically possible three-PL combinations, and x
a good many of the four-PL combinations, are lacking. The gaps, however, seem
to reveal no possibility of further simplification.
The best way to get at the "extra height" phenomenon is to imagine, for a
moment, that there are four PLs instead of three: the fourth, symbolized by
/*/, is higher than the other three. None of the intonations which we have so far
displayed or subsumed is affected by this tentative assumption, but we are able 1

to provide for a certain number of occurrent intonations which the scheme, as

so far described, cannot cover. We find, for example, such pairs as the following
2
It's three o m clockl :

2
IVs three o m clockl ;

2 S3
Is it three o'clock] :

2
Is it
u three o'clock] ;

32
John\ :

John\ ;

31
Fesl :

il
Yes\ ;

2
Are you Agoing] 3 said 33John] :

2
Are you Agoing] i said ^John] .

The second of each of these pairs differs from the first only in having /4/ wherever
(and only where) the first has /3/. Furthermore, the first of each pair has /3/

at the center. Ifwe consider intonations which have /2/ or /l/ at the center, we
find no contrasts between intonations without /4/ and intonations with /4/.
That is, we find a /3 21j/, but no contrasting /4 21j/.
This distribution renders it inefficient to treat /4/ simply as another PL.
Instead, we set up a feature of "extra height," which we shall symbolize by
/*/ directly before the PL-sign at the center of the intonation. This feature either
occurs, or does not occur, in a macrosegment. All intonations with /3/ at the
center may
occur either with or without this extra feature; intonations with /2/
or /l/ at the center do not show it. The phonetic effect of /*/ is to raise any and
all /3/'s in the macrosegment from [3] to [4]. Thus we can write:

2
IVs three o m clock\. :

2
IVs three o'*-
n clockl
;

2 33
Is it three o'clock] :

2 33
Is it * three o'clock] ;

and so on. Since there are (theoretically) nine different two-PL intonations with
/3/ at the center, twenty-seven three-PL intonations of this sort, and fifty-four
four-PL intonations, these are the outside limits on the number of different
intonations involving /*/; actually, fewer than this occur, but it does seem that
every occurrent intonation with /3/ at the center is attested with /*/ as well as
without it.
 A MANUAL OF PHONOLOGY 49

In the above we have dealt with English intonations directly in terms of the
ultimate phonologic constituents of which they are composed. It is necessary

now for us to show their hierarchic organization or IC (= "immediate con-

stituent") structure.
This can be done very briefly. Let us consider the most complex sort of intona-
tion, with four PLs, the /*>/ feature, and a terminal contour. Our first IC cut
separates /*/ from the rest, since every intonation with /*/ is matched by a
simpler one without it: /*/ is "satellite" and the rest is "nucleus." The second IC
cut separates the first two PLs from the one at the center, the one at the end, and
the TC: every intonation has the latter apparatus, but only some intonations
have the former. Therefore the first two PLs are "satellite" and the rest is
"nucleus." The satellite in turn consists of the macrosegment-initial PL as
"nucleus" (since always occurs if the center is not at the beginning of the
it

macrosegment) and the other PL as "satellite." The final two PLs and the TC,
lastly, must be taken as standing in a three-part "coordinate" construction, since
every intonation involves all three. Thus, the intonation of
2
Is it Hhree o'^clock]

is /2 3 *33|/; this and nucleus /2 3 33f/; this is satellite /2 3/

is satellite /*/
(consisting of nucleus /2/ /3/) and nucleus /33|/; and the latter is a
and satellite
coordinate construction of the three ultimate constituents /3/, /3/, and /f/.
The IC structure of any simpler intonation is clearly implied by the above de-
scription of that of the most complex intonation.

212. Other Intonational Systems. Sierra Popoluca has an intonational system

which at first appears remarkably one
might even say suspiciously similar
to that of English. Thus the system involves four PLs and three TCs, plus one
"terminal feature" which is of the all-or-none type of English /*/ However, the
precise nature of these features and the arrangements in which they occur differ
strikingly from the English system. Each macrosegment ends, by definition, with
one or another of the three TCs, and this may either be accompanied or not by
the "terminal feature." Apart from this, intonations fall into two sets: one-PL
intonations and two-PL intonations. A one-PL intonation accompanies a re-
mainder in which there is only one occurrence of the phoneme of stress, and
the PL occurs distinctively on the syllable which bears that stress. A two-PL
intonation accompanies a remainder in which there are two or more occurrences
of stress: the first PL occurs on the first stressed syllable, and the second PL
occurs on the last. In two-PL intonations, any of the four PLs may occur in the
first and any of the four in the second. The TCs are directions of glide
position
of pitch from the last PL in the macrosegment to the end: one is an upwards
glide, one a downwards glide, and one a suspension of the same pitch level. The
"terminal feature" consists of a dying-away effect, often accompanied by whis-
pering, but this may appear with any of the TCs, rather than (as in English)
only as an integral part of one of the TCs. Certain types of macrosegment-
terminal vowel-and-consonant phenomena preclude the appearance of the
"terminal feature," which is thus limited just as is /*/ in English, but in a very
different way.
50 A MANUAL OF PHONOLOGY

Preliminary investigation seems to show that many or most of the languages

of Europe have intonational systems more or less like that of English. Many of
the languages of northern Europe (French, German, and northern Slavic) seem

to have three PLs and a feature comparable to English /*/ or, it may be, simply
four PLs. Many of the languages of southern Europe, on the other hand (Ro-
mance except for French; Greek, Turkish, southern Slavic) seem to have systems

with just two PLs plus something like English /*/ or else, it may be, three PLs
and no /*/
Very little is known of Fox intonation, but it is clear that there is at least a
two-way contrast of TCs (or something more or less like TCs). One, the more
frequent, is marked by voicelessness of the last syllable of the macrosegment,
which always ends with a short vowel. The other is marked by voicing and high
stress (and pitch?) on the last syllable, which, in the case of a certain number of
interjection-like words, ends in a long vowel, otherwise in a short vowel. Nothing
is known of the PLs of Fox
if any. Work in the summer of 1952 with two re-

lated languages, Shawnee and Arapaho (the first quite close to Fox) seemed to
show only two PLs, and this may be the case for Fox as well.
There is, of course, no reason why an intonational system must always be
built out of the types of elements which it is convenient to recognize in English
and in Sierra Popoluca, and certainly no reason why the pitch of the voice should
function as the principle phonetic raw-material for every intonational system.
On the other hand, a language which makes phonologic use of pitch in other ways
is not necessarily precluded from using it intonationally. Two examples deserve
brief mention. In Mandarin, the pitch contours required by intonation are
spread through the succession of syllables which individually carry pitches or
contours of an accentual system (2321), so that the actual pitch at any point is
a sort of geometrical sum of the two factors. Thus the tones and stresses are
identical in /tui ma?/ 'Is that right?' and in /tui le./ 'That's right,' both of
which have been falling tone, and loud stress, on the first syllable, no tone and
no stress on the second. But in the first of the two the fall of pitch is only to
middle register, and the second syllable is pronounced in the middle register;
in the second of the two, the fall is all the way to low register, and the second
syllable is pronounced in the low register; this difference is intonational. In
Mazahua Otomi the situation is quite different. Roughly speaking (a definitive
purely phonologic analysis is not yet available), the last syllable of each "word"
is reserved for pitches which constitute part of the intonational system, while all

preceding syllables of each "word" carry pitches which are part of an accentual
system.
All-or-none features, logically like the English /*/, are known in some instances
where the other details of the intonational system remain obscure. The French
phenomenon traditionally called the accent d'insistence is of this sort. This
manifests itself, when present, on the first macrosegment which
syllable of a

begins with a consonant thus not necessarily on the initial syllable, which may
not begin with a consonant; it lengthens the consonant and places greater
articulatory force both on the consonant and on the following vowel. Thus the
 A MANUAL OF PHONOLOGY 51

word impossible, pronounced as a macrosegment with this feature, has a length-

ened p,and the -po- is stressed and (probably always) has higher pitch.
Some of the above information is, of course, not very reliable. Investigators
have been attempting intonational studies for a fairly long time, but earlier
reports are so different in approach that they are hard to interpret, and the most
recent studies show a variety of technique which clearly reflects differences
among the investigators as much as, or more than, differences among the lan-
guages studied. This is all to the good, since only by trying many alternative
techniques of analysis can we ever hope to develop a unified technique which will
both yield valid results in individual cases and also render cross-language com-
parisons easier.One of the most interesting of the newer approaches is that used
by Martin for Korean he simply precedes each syllable by /\ / if it is higher in
:

pitch than the preceding syllable (or, if directly after pause, if it contains a rise
in pitch) and by /[/ if it is lower in pitch than the preceding (or, if directly
after pause, if it contains a fall in pitch). It is impossible to judge, without ex-
tensive independent work, whether this device actually covers all the intona-
tional contrasts of the language. In any case, quite obviously the time for a
general typology of intonational systems has not yet arrived.

22. Syllables and Juncture. In 210 we made our initial IC-cut of macroseg-
ments (presumably in any language) into an intonation and what we informally
called a "remainder." The term "remainder" is so awkward that we shall now
make a terminological shift, using henceforth the term "macrosegment" itself
for what we have previously been calling a "macrosegment" minus its intona-
tion. In theory, the material with which we are now to deal has been completely
extricated from its accompanying intonational matrix, and we cannot be con-
cerned even with any "side-effects" of the latter. In practice, it is of course im-
possible to "lift off" intonations completely unless the intonational system has
been thoroughly analyzed, so that the intonationless macrosegments with which
we shall now deal may in some cases include unrecognized features that are
actually parts of intonation. This is the sort of risk which must be entailed if
we are to proceed at all.

In every known language, a macrosegment (now and henceforth in the new,

more limited, meaning) consists of one or more smaller structural units to which,
by a generalization of its meaning, we shall assign the term syllable. It is not to
be assumed that "syllables" in all languages are comparable to those of English
or to what are usually interpreted as syllables in English. In our generalized
sense of the term, it will turn out that the so-called syllables of English indeed
conform to the broader definition, but in other languages we will find strikingly
different units which also conform to the definition.
A preliminary survey of syllables in English will serve to introduce a number
of terms which will be needed in our survey of types of syllable structure.
Consider the following utterance, which in our traditional conception of Eng-
lish would be called monosyllabic:

31
Hotl / hatj/


52 A MANUAL OF PHONOLOGY

Lifting off the intonation leaves what we can transcribe (though not pronounce)
as /hat/ (the omission of a stressmark is intentional: the transcription subsumes

what is customarily called "primary" or "loud" stress; see 2311). This is a

single syllable,and consists of three constituents: an onset, /h/; a peak, /a/;
and a coda, ft/. Whether these three constituents are to be regarded as struc-
turally on a par with each other (three ICs), or whether some two of them go
together more closely and the combination in turn goes with the third, is a mat-
ter which will be taken up in 2222. As a second example, consider

Hot day \
2
/ hat -f
31
dey|/

Here there are two syllables, and each consists of onset, peak, and coda. Be-
tween the two peaks occurs a sequence of material consisting of a coda which
goes with the first peak and an onset which goes with the second, and, further,
an internal open juncture (or simply juncture) which marks clearly the line of

demarcation between coda and onset the point of division between the suc-
cessive syllables. Successive peaks in English are not always so separated, as
can be seen by comparing the following two familiar disyllabic utterances with
each other:
31
Nitrate. / naytreyt|/
Night-rate. 31
/ nayt + reytj,/

In both, the first peak is /ay/, the second /by/. In the second example, the first

peak is followed by a coda /t/, and the second is preceded by an onset /r/, with
an intervening juncture. But in the first example, the consonant sequence /tr/
cannot be broken into a combination of coda and onset. Rather, it constitutes
another type of structural unit which we must recognize in English: an interlude.
An interlude is coda-like and onset-like at the same time, and structurally it
belongs both to the syllable which contains the preceding peak and to that which
contains the following peak. When two successive syllables in a language like
English are finked by an interlude, there is no "point of syllable division" be-

tween them. Typographical rules governing how to break a word between the
end of one line and the beginning of the next are largely arbitrary in such cases
at least in their relation to pronunciation. Certain special styles of singing in-
troduce a clear break; for example, in singing such a two-syllable sequence on
two successive notes, one would usually sing nitrate as ni-trate or as nite-trait.
But what this special style does is to introduce a juncture, at the indicated point,
which is not there in normal speech.
The terms of importance introduced in the above brief discussion are "onset,"
"peak," "coda," "interlude," and "juncture." The best starting-point for a
classification of phonologic systems as to syllable type seems to be to divide
them all into two groups: those having juncture and those not having any such
phenomenon. The absence of a juncture necessarily implies that there can be no
contrast between medial interlude and medial coda-onset sequence; however,
we shall see that it is in some cases more convenient to speak of interludes (to
the exclusion of coda-onset sequences) and in other cases the reverse; in still

others, the choice of terminology seems not to matter.

 A MANUAL OF PHONOLOGY 53

In the treatment of specific examples, we shall have to supply terms for struc-
tural classes of units: for example, in discussing Mandarin, we shall use the term
"semiconsonant" for a set of three units which may either constitute a whole
syllable (except for tone and stress, if any), or occur as onsets in syllables that
include other material as well. Later on, in 242, a number of terms of this sort
will be assigned rather sharp meanings for purposes of cross-language discussion
and comparison. For the present, however, the terms will be defined separately
for each language, with no necessary consistency in their precise use from one
language to another.

221. Systems without Juncture. Systems of this kind fall into four subtypes:
the peak type, the onset-peak type, the onset type, and, standing rather apart
from these first three, the duration type. The difference (for the first three) lies

in the constant or most nearly constant constituents of syllables: in the peak

type, for example, every syllable, or every syllable save for a definable and limited
subclass, contains a peak which contrasts positively with other possible peaks.
2211. Peak Type. Systems of this type are further differentiated by the nature
of the phonologic material which defines (and constitutes) a peak. The following
varieties, at least, are worth distinguishing: peak defined by vocoid; by tone; by
vocoid and stress; by vocoid and duration.

22111. Peak Defined by Vocoid. In Fox there are eight contrasting peak units:
/ii* e e* a a* o o*/. Every syllable peak consists of one of these eight vocoids, and

every occurrence of any one of the eight constitutes a syllable peak. All other
vocoids which occur in Fox are of the glide type (165). This exact matching of
syllable-peaks and a specific set of units, all of them vocoids, is what leads us to
say that peaks are definable in terms of vocoids.
Initially (in a macrosegment) a peak may be preceded by no distinctive onset,
or by one of the set /p t c k s m
n/, or by one of the pair /w y/, or, with some
limitations, by an onset consisting of one of the first set followed by one of the
pair. In final position, only the four peaks /i e a o/ occur with the most frequent
TC (involving terminal devoicing; 212), but all eight seem to occur with the

other TC. No have codas. Successive peaks are in all cases sepa-
final syllables
rated at least by an onset for the second. This onset may be any of those which
occur initially, or /h hw hy/. There are also interpeak sequences consisting of
/h/ plus /p t 6/ or /k/, with or without a following /w/ or /y/, and of // plus
/k/, with or without a /w/ or /y/. Phonetically, such sequences with /h/ as the
first element strike ears accustomed to English as having /h/ as a coda for the

preceding peak, the remainder as an onset for the following. But structurally,
since there is no possible contrast between coda-onset sequence and interlude, it
does not matter which terms we use.
Senadi represents a rather different variety. Here there are seven vocoids every
occurrence of which constitutes a syllable peak: /i e e a o o u/ There are also
four nasal continuants, /m n n y n/, which do not constitute peaks when im-
mediately before a vocoid of the first set, but do otherwise. (These nasals are
reanalyzed in 252212.) Peaks of the latter type are never preceded by distinctive
onsets, but those of the former type may be so preceded or not. For five of the
54 A MANUAL OF PHONOLOGY

vocoids, /i e a o u/, when preceded by an onset which is not itself nasal, there
is a further contrast between oral and nasal after onsets
; /m n n yis no rj/ there
contrast, all seven of the vocoids occurring and being nasal.
Unlike Fox, peaks may occur in immediate succession, with no intervening
onset there are no distinctive codas at all. In addition to the above, there are
three tones, respectively high, mid, and low. The number of tone-occurrences
in a macrosegment, however, does not agree with the number of syllables,
since although each syllable peak bears at least one tone, a few are accompanied
by a succession of two (different) tones.
Senadi borders on the variety of syllabic system in which peaks are defined by
duration (or at least jointly by vocoid and duration, 22114), and might be so
classed. Thus the essential difference between a two-peak sequence /ha,/, the
first with low tone and the second with high, and a single peak /a/ with two suc-

cessive tones, lies in the fact that the former takes approximately the same length
of time as a sequence like /taka/, while the latter takes approximately the same
length of time as /a/ or /a/.
22112. Peak Defined by Tone. Bariba apparently has a system which is a kind
of reversal of that in Senadi. The vocoid system (other than nasal continuants)
is much the same as in Senadi. Nasal continuants /m n/ occur both as peaks and
as onsets, but which is the case is not predictable in terms of what precedes and
follows: /m
n/ are peaks when accompanied by a tone, otherwise not. The es-
sential ingredient of a syllablepeak is thus a tone, of which Bariba has five, high,
mid, low, rising, and falling; all take about the same length of time to produce.
There are as many syllables in a macrosegment as there are tone-occurrences.
Each peak includes, in addition to the tone, a nasal continuant /m/ or /n/ or
one of the seven other vocoids. Peaks not consisting of tone plus /m/ or /n/ may
have no distinctive onset or any of a number of contrasting onsets; there are no
codas.
Thus both in Bariba and in Senadi sequences such as /aa/ and /a/ contrast.
But while in Senadi the former and the latter both include two tone-occurrences,
spread differently over vocoids the presence of which marks the occurrence of a
syllable peak, in Bariba the former has two successive tone-occurrences, ac-
companied by the same vocoid, while the latter has a single tone (the rising one),
and it is the number of tone-occurrences which tells how many syllables there are.
22113. Peak Defined by Vocoid and Stress. In Spanish there are five stressed
vocoids, /i e a 6 u/, which* constitute syllable peaks wherever they occur. There
are also five unstressed vocoids, /ieao u/, two of which constitute syllable peaks
only in certain environments. Unstressed /e a o/ occur only as peaks. Unstressed
/i/ and /u/ constitute peaks except in the following cases: (1) when flanked, on
either side, by a stressed vocoid or by /e a o/; (2) in the sequence /hi/ or ui/.
In this latter case the sequence counts as a single peak, but it is very hard to tell,

phonetically, whether the first or the second is the more prominent. That is, it is

hard to whether a form such as muy 'very' (/mui/), in a context in which

tell

neither vocoid is stressed, should be interpreted as [muy] or as [mwi]. From the

point of view of counting syllables, this indeterminacy does not matter.

 A MANUAL OF PHONOLOGY 55

The conventions of Spanish poetry introduce a complication into this picture.

If these conventions can be regarded as sufficiently artificial, then their implica-
tions can be ignored, but this conclusion requires careful investigation. In verse,
a sequence /ea ae oa ao/ or of /e/ or /o/ flanked on either side by a stressed
vowel, counts as two syllable peaks if within a single word, but when a word
boundary occurs between the /e/ or /o/ and the /a/ or stressed vowel, the /e/ or
/o/ not counted as a separate syllable. Thus veo (/beo/ T see') is two syllables
is

in poetry, but se olvidar (/seolbidar/ 'I know how to forget') would count as
only three, not as four. Now there is no convincing evidence for an internal open
juncture in Spanish. Thus if such contrasts are genuine for ordinary speech, it
becomes necessary to distinguish between peak unstressed /e/ and /o/ and glide-
vocoid /e/ and /o/, even though no such contrast exists for the high vowels /i/
and /u/.
22114. Peak Defined by Vocoid and Duration. In Fijian there is a set of five
vocoids, /ieao u/, which occur only and always as peaks; and every peak con-
sists ofone or another of these five. Any peak, initial or medial in the macroseg-
ment, may
be preceded by no distinctive onset or by one of sixteen contrasting
onset consonants. The element of duration enters because the only difference
between /a/ and /aa/, or /[/ and /ii/, or the same contrast with any other
vocoid, is that the second takes approximately twice as long to say as the first.
That is, the word /jSaa/ 'four' takes about as long as /sajSa/ 'clean', where each
peak is accompanied by an onset, or //?ia/ 'wish', where the two peaks differ in
vocoid quality; but //3aa/ has approximately the same unchanging vowel color
throughout after the //?/, with no pulsation between the two /a/'s, and this vowel
color is approximately the same as that of a single /a/ followed by a syllable
beginning with an onset consonant.
The between Fijian and Fox is instructive. Fox has both short and
difference
long vocoids, but the structure is such that a long vocoid clearly constitutes the

peak of a single syllable, not two successive peaks of the duration type. In Fijian,
a sequence of two identical vocoids is structurally quite like a sequence of two
different vocoids. The Fijian macrosegment is marked by some differences of
stress from syllable to syllable, seemingly determined in a mechanical way in
terms of the positions of the various syllables relative to beginning and end of
the macrosegment. In particular, the next to the last syllable is stressed (if there
are at least two in the macrosegment) if the last two syllables contain identical
;

vocoids not separated by an onset consonant, then this stress stretches through
both vocoids. This statement about the location of stress in the macrosegment
can only be made efficiently in terms of the syllable-peak analysis given above.
To treat Fijian as having "vowel length" like Fox (or, worse, like our current
impression, perhaps in part false, of Latin) is to deal with the facts very ineffi-
ciently.

2212. Onset-Peak Type. In all the systems so far discussed the only syllable-

constituent constantly found in all syllables has been a peak, itself of one or
another structure. Because of this, we would say that in each of these systems
the peak is the nucleus of the syllable, and that any other syllable-element
56 A MANUAL OF PHONOLOGY

present in one or another instance is a satellite. Thus in Fijian a syllable-onset

consonant is a peak; in Cree, which resembles Fox except
satellite to the nuclear
that some syllables have codas, a coda is a satellite to the nuclear peak.
In systems of the onset-peak type, on the other hand, every syllable contains
at least both an onset and a peak. In this case, neither of these two can be in-
terpreted as satellite to the other; the construction is, so to speak, coordinate.
Codas, when they occur, are satellites to the onset-peak combination.
The simplest variety of onset-peak system one in which there are no
is

codas, so that macrosegments uniformly consist of one or more syllables each

including onset and peak. Examples are not very common, but Mazateco has
such a system.
Onsets are simple, consisting of one or another of some eighteen or twenty
consonants (fewer before loan words were introduced from Spanish), or complex,
consisting of certain clusters of some of these consonants. Onset clusters may
contain as many as three consonants, but in most of these the cluster seems to
consist of two immediate constituents, one of which in turn consists of two con-
sonants. Peaks always involve at least one of four vocoids and one of four tones,
so that the combination of vocoids and tones is coordinate; in addition, the peak
may be nasalized (not distinctively after nasal onset consonants or certain other
simple or complex onsets), and nasalization is to be interpreted as satellite to the
vocoid-tone combination. The vocoid part of the vocoid-tone combination may
consist of up to three vowels in succession, and the tonal part may consist of up
to three tones in succession, but there is little coordination between the number of
vowels and the number of tones. Regardless of complexity, a syllable lasts ap-

proximately the same length of time the duration of peaks being somewhat more
consistent, perhaps, than that of onsets.
Yawelmani presents a system with far fewer layers of immediate constituents,
but with codas. Peaks are one or another of a set of ten vocoids, five short /i e
a o u/ and five long /i- e* a- o- u-/; these vocoids occur only as peaks. Onsets
are one or another of thirty-three contoids or glide vocoids; any of these may also
occur as a coda, though a few are very rare in that function. Thus every macro-
segment begins with an onset unit followed immediately by a peak unit medially, ;

one consonant (contoid or glide vocoid) between two peaks constitutes an onset
for the second peak and the preceding syllable has no coda, while two consonants
are split between the two syllables; in final position, one may have either a peak
or any of the possible codas. Although there is no contrast between an interpeak
coda-onset sequence and an interlude, the occurrent material between successive

peaks is more easily described and with complete phonetic accuracy in terms
of codas and onsets.
A great many phonologic systems resemble that of Yawelmani in all details
except for the exact matching of onsets and codas. Often, only some of the con-
sonants which occur as onsets also occur as codas; sometimes codas include, in
addition, some sequences of two or more consonants. Except by virtue of such
complex codas, it is rare to find a greater variety of codas than of onsets. Thus in
Taos there are twenty-seven different onsets; nine of these are voiced, and occur
also as codas, while the rest do not; there are no complex codas at all.
 A MANUAL OF PHONOLOGY 57

Nootka resembles Yawelmani save for the occurrence of long codas. Of the
thirty-four consonants that appear as onsets (initially or medially, whenever a
vowel immediately follows), at least twelve occur neither as nor in codas. But the
consonants which do so occur turn up singly or in long strings with relatively few
limitations on sequence. It may be, therefore, that Nootka should be classed with
the onset type, which we are about to discuss; this possibility will be considered
after an unambiguous instance of the onset type has been presented.

2213. Onset Type. This type of syllable structure is most unusual from our

Western point but seems to be undeniably attested at least in one case:

of view,
Bella Coola. This language was the subject of a paper by Boas half a century ago,
which is said to have been rejected by the editor of a German philologic journal
on the grounds that, as "everyone knows," it is impossible to have words
without vowels. Bella Coola proves that this nineteenth-century belief is false.
The key feature of Bella Coola syllable structure is that while every (or al-

most every) syllable has a distinctive onset, many syllables contain no other
syllable-element distinctively. Some contain a distinctive peak, some a distinc-
tive coda, and some both. Whatever occursin a syllable in addition to the onset
is to be regarded as satellite thereto;both peak and coda are present, they
if

stand in a coordinate construction, and the combination is then satellite to the

onset.
Thus there is a set of units, full consonants, which constitute onsets
wherever they occur, and are therefore nuclei of syllables on every occurrence.
One or two of these full consonants (which total thirty) are defective: they occur
only in syllables containing also a distinctive peak or coda. There is a set of three
semiconsonants, /m n 1/, which function now as onsets, now as codas, depending
in a fairly complex but perfectly determinate way on the environment. Finally,
there is a set of vowels, /i a u/, which occur singly or, it would seem, in sequences
of two, and which constitute (singly or in pairs) syllable peaks wherever they
occur.
We give some examples of Bella Coola macrosegments to show how these
differently constituted syllables can occur in sequence relative to each other. A
hyphen is inserted to show the point of syllable division; the hyphens are not part
of the systematic notation, since their position is predictable in terms of the
other symbols used, and they are included only for clarity of exposition. In
/l-k' w -t-x w / 'make it big!' there are four syllables, each consisting of onset un-
accompanied by any distinctive peak or coda. Furthermore, this entire macro-
segment is voiceless. The same is true in /k'-x-l-c/ 'I looked'. In /s-k'1-xl-x-c/
'I'm getting cold', two of the syllables have a distinctive coda (/l/) ; an automatic
murmur vowel between the onset and the /l/ functions phonetically as the peak,
but is phonologically predictable in terms of the onset consonant and the coda.
In /ka-c'an/ 'tail' the second syllable contains both distinctive peak and dis-
tinctive coda. In /nm-nm-k'/ 'animal' and /mn-k/ 'excrement' we see semi-
consonants functioning as onsets (those written initially, or immediately after a
hyphen) and as codas (those written before a hyphen) here, also, there is a pre-
;

dictable murmur vowel as peak. The first syllable of /m-tm/ 'sea-egg' is the one
variety in Bella Coola which seems not to involve a distinctive onset: that is, a
58 A MANUAL OF PHONOLOGY

syllable consisting of a preconsonantal semiconsonant initially in the macro-

segment. In this connection there remain some problems of analysis; it may be
that here, too, a distinctive onset has to be recognized (perhaps /?/)
The clinching factor in forcing us to this interpretation of Bella Coola is the
occurrence of whole macrosegments in which all the syllables consist, structurally,
of distinctive onset only (cf the first . two examples given in the preceding para-
graph). Supporting this is the fact that if we interpret successions of such
syllables simply as "consonant clusters," then there seem to be virtually no lim-
itations on the sequences in which consonants can occur in clusters. In every
language where the peak or peak-onset interpretation suggests itself, there are
such limitations, few or many (save for an occasional case of complete match-
ing of simple onsets with codas, as in Yawelmani).
Yet in some cases where the second situation occurs, unaccompanied by the
first, we may still prefer the onset interpretation. Let us describe Kota as though

this interpretation were to be accepted. Kota has a set of eighteen full con-
sonants, which occur only as onsets, and five semiconsonants /m n n rj s/ which
occur also as codas (// only as coda, after a vowel before a retroflex stop /\ d/).
There are also onsets consisting of one or another consonant plus /y/. There are
five vowels, /ieao u/, and five long vowels /v e- a* o* u /, which occur only
-

as peaks; all peaks consist of one of these ten, but there are syllables with a
peak and syllables without. Initially in a macrosegment, syllables occur consist-

ing of a peak without an onset that is, vowels occur initially and this is one
structural difference from Bella Coola. Any full consonant followed immediately
by another full consonant constitutes a syllable of the isolated-onset type.
Phonetically, such a consonant is followed by "loose transition" to the next,
producing a murmur vowel, an aspiration, or the like; phonetically, but not
phonologically, this may be taken as a syllable peak. By this interpretation, the
form /an3rcggvdk/ 'because of the fact that (someone) will cause (someone)
to terrify (someone)' would not consist of one syllable, containing a peak followed
by a long and complex coda, but of ten syllables, /an-3-r-6-g-6-g-v-d-k/, most
of them of the isolated-onset type.
The major structural difference between this and Bella Coola lies in the fact
that every Kota macrosegment contains at least one syllable with a distinctive
peak. This is apparently true also of most of the Salishan languages other than
Bella Coola, which otherwise resemble the latter, particularly in having long and
complicated sequences of consonant phonemes with few limitations. (Georgian
also belongs here; no doubt there are others.) Such a situation implies another
factor of hierarchic structure in the macrosegment: since every macrosegment
contains at least one syllable with a distinctive peak, a syllable of that kind con-
stitutes a nucleus, and any other which are present are satellites to it.
syllables
This does not tell us, of course, what macrosegments which
to do in the case of
contain several distinctive-peak syllables. But this structural situation is very
familiar though it usually turns up in the context of rather different phonologic
raw-material. For example, it often happens that at least one syllable in a macro-
segment must bear stress, so that in a macrosegment containing one and only
 A MANUAL OF PHONOLOGY 59

one stressed syllable, any other syllables are satellites to it; if the macrosegment
contains more than one stressed syllable, then we are faced with just the prob-
lem This problem will be dealt with in 23.
of IC-analysis stated above.
Now we can turn back to the Nootka system. Nootka seems to occupy an
ambiguous position between the onset-peak type with some codas (Yawelmani)
and the onset type of Kota (if we so interpret it) rather than that of Bella Coola.
There are rather more limitations on "consonant clusters" than in Kota or a
Salishan language, so that it might not be too difficult to describe what occurs in

terms of long and complex codas; yet certainly the relative complexity of these
"clusters," despite the limitations on sequence, is reminiscent of the typical
situation in Kota or Salishan. No doubt a further refinement of our typology can
provide a neater categorization.
2214. Duration Type. Japanese has a set of vocoids, /i e a o u/, which occur
only and always as syllable peaks. The statements which we made about Fijian
(22114) apply in part here: the sequence /aa/ is two syllables, rather than a
single syllable with a long peak /a /, even though phonetically what one hears
-

for /aa/ is [a*]; the reasons are quite comparable to those stated for Fijian. In
addition, however, there is a nasal continuant /n/ (the diacritic is arbitrary),

which varies in position of oral closure depending on environment, but which is

always a syllable peak: a syllable with this peak has neither onset nor coda.
And, finally, there is a set of voiceless stops and spirants, /p t k c 6 s x/, and
an /h/, which occur both as syllables in their own right and as onset consonants.
The members of this set are syllabic in the position before any other member
of the same set (not all of the sequences occur) otherwise they are onsets. Thus
;

/nippon/ 'Japan' takes just about the same length of time to utter as does
/sayonara/ 'goodbye', where all the syllables contain onset and peak; the syl-
lables of the first are /ni/ (onset plus peak), /p/ (acoustically a silence of ap-
proximately syllabic duration), /po/ (onset plus peak) and /h/ (syllabic nasal).
We cannot class this Japanese system in any of th? other three types (peak,
onset-peak, or onset), because the Japanese syllable is defined fundamentally in
terms of duration and nothing else. A syllable consisting of a lone /p/ or /s/
cannot be broken down into anything like onset and peak; it simply lasts the
proper length of time to fulfil the requirement of syllabicity.
In terms of hierarchic pattern, a Japanese syllable consisting of /h/ or of a
voiceless stop or spirant is a satellite, since every macrosegment contains at least
one syllable with a vocoid peak, /n/ occurs only initially before an onset conso-
nant or medially after a vocoid; stop and spirant syllables occur only between
successive syllables with vocoid nuclei, of which the second must have an onset
consonant.
222. Systems with Juncture. The recognition by the analyst of an internal
open juncture may be brought about in a number of different ways, and the
methodology will be discussed later (324). Even where juncture either must be
recognized or is merely extremely convenient to recognize (and we do not insist
that these two situations are distinct), there may or may not be any contrast
between medial coda-onset sequences and interludes. Where there is no such
60 A MANUAL OF PHONOLOGY

contrast, then we do not speak of interludes at all, and the juncture may be
called syllable juncture. We shall take up this sort of situation first, and then
turn to that in which interludes must be recognized.
By definition, we shall assume that the boundary of a macrosegment entails
open juncture, since this renders description easier. In some instances this may
be a mode of speaking rather than a phonemic conclusion.
2221. Systems without Interludes. Our
example here is Cantonese. In a
first

macrosegment there are as many segments bounded by

syllables as there are
successive syllable junctures. Each syllable contains a single peak, which may or
may not be preceded by distinctive onset, and may or may not be followed by
distinctive coda. Furthermore, each syllable includes one tone.
A number of vowels, including /a/, occur only as peaks. Two units, /[/ and /u/,
occur as peaks, as codas, and as onsets (or as parts of complex onsets). Three
units, /m n rj/, occur as onsets and as codas, and also as peaks, but in the latter
case only when they stand alone in the syllable except for the tone (this limitation
does not apply to /i/ and /u/ as peaks). Of the consonants, /p t k/ occur both
as onsets and as codas, while the rest occur only as onsets. Writing a hyphen for
syllable juncture, it is possible to get such contrasts as the following:

/. . .an-a. . ./

/. . .a-na. . ./

/. . .a-n-a. . ./

to say nothing of /. . .an-na. . ./,/.

./,/. .a-n-na.
. ./ and the like.
.an-n-a. . . .

In the first three, we /n/ between two successive vowels is a

see that whether
syllable, or a coda, or an onset, depends on the number and location of syllable
junctures.
This is one of the distinctive things about the Cantonese system. Since a syl-
lable is defined by its boundaries, and since each syllable contains one and only
one peak, it turns out that there is no contrast between, say, glide-vocoid [y]
and peak-vocoid [i], or between onset [n], coda [n], and syllabic (peak) [n]. The
unit /i/, for example, will be an onset, or part of a complex onset, if it is followed
by a vowel like /a/; a coda if it is preceded by such a vowel; but will itself be
the peak if the syllable which contains it does not also contain a vowel. The
point of syllabic division is not predictable from the sequences of vocoids and
contoids, as it is in Yawelmani; but, contrariwise, given the points of syllable
peaks is predictable.
division, the location of
However, this property of Cantonese is not exclusively to be found in systems
with syllable juncture. The occurrent sequences of vocoids and contoids in some
languages without open juncture, perhaps for example in Ojibwa, may be such
that [y] and [i] are not in contrast. Nor do all systems with syllable juncture
necessarily have this property. Burmese, which is in some ways much like Can-
tonese, has distinct and contrasting /y/ and /i/, /w/ and /u/.
The Apachean languages represent a different variety of system with syllable
juncture, particularly in the functioning of tones. The fundamental description
of the Cantonese syllable applies here as it does, necessarily, in any syllable-
 A MANUAL OF PHONOLOGY 61

juncture system. We supply details for Chiricahua. A syllable may have no onset,
or any of twenty-nine simple onsets, or any of four complex onsets /mb nd sd
d/. It may have no coda, or any of eight simple codas /? h z s & s 1 1/, or any of
three complex codas / ? s ?s ?!/ The three complex codas occur only before
syllables beginning with an onset. The peak consists of one vocoid /i e a o/ or of
certain sequences of two of these, including the geminates /ii ee aa 00/ and /ea ai
ei oiao eo/. The whole peak can be oral or nasal. Each individual peak vocoid is
accompanied by one or the other of two tones (high and low pitch). There are
also syllables with no onset or coda consisting of the consonant /n/ or /m/ with
one of the two tones, or of the geminate cluster /nn/ or /mm/, each with one of
the two tones. A syllable containing two vocoids lasts about twice as long as a syl-
lable containing one. We may therefore introduce the term mora: a syllable con-
tains one or two moras. There is no need for this term (in addition to "syllable"),
save when a language shows both types of unit, as Chiricahua seems to.
The tones here function differently from any case cited heretofore. There is an
exact matching between number of tone-occurrences and number of vocoid or
syllabic nasal occurrences, and the duration of a syllable depends on both
equally. There are more limitations as to what variety of syllable can precede
or follow what other type than there are in Cantonese; one such limitation was
stated in the preceding paragraph. The position of the syllable juncture is in
many cases predictable in terms of the sequence of tones, vocoids, and contoids,
but there is a sufficient body of exceptions to justify recognition of the syllable
juncture as a separate functioning entity. Thus, in such a sequence as /. . .V-
tV. . ./, there has to be a syllable juncture and it necessarilycomes before the
/t/; therefore a notation which left out any mark for the juncture would still be
unambiguous. But in /li-ba-i/ 'those who are gray' versus /sal/ 'sand', one has a
contrast between /a/ and /i/ in two successive syllables and the same sequence
within a single syllable, and this contrast would be obliterated if the symbol for
syllable juncture were omitted from the notation.
The writer knows of no case of a system with syllable juncture which does not
also have tones of one kind or another. Burmese, Thai, Vietnamese, many dialects
of Chinese (but not all), and a good number of less well-known languages re-
semble Cantonese both in the presence of syllable juncture and in having tones
with whole syllables as their domain. The Apachean situation may recur in the
Northern Athabascan languages. Certain of the West Coast Athabascan lan-
guages are reported to have no tones; if, apart from this, they have the kind of
syllable structure typical of Apachean, then they illustrate the possibility men-
tioned at the beginning of this paragraph.
2222. Systems with Interludes. A system which has both interludes and internal
open juncture necessarily has some segments flanked at both ends by that junc-
ture, or by margin of macrosegment, but containing more than one syllable.
Such we shall call a microsegment, and the variety of internal open juncture which
bounds microsegments we shall call microjuncture. Syllable juncture is, then,
simply a variety of microjuncture; we use the former term when microsegments
and syllables are identical.
62 A MANUAL OF PHONOLOGY

Mandarin somewhat marginally, into this type. In many ways Mandarin

falls,

is like The tones of Mandarin, like those of Cantonese, have as their

Cantonese.
domain whole microsegments. Unlike Cantonese, some Mandarin microsegments
carry no tone; since in any macrosegment (with rare, interjection-like exceptions)
there is at least one microsegment bearing a tone, toneless microsegments are
satellites to thosewith tones. Also unlike Cantonese, there are a fair number of
microsegments in Mandarin containing two peaks instead of just one, with an
intervening interlude. Thus /srama/ 'what' carries a single tone (rising), and
is flanked by microjuncture but contains none; but it has two peaks, both con-

sisting of the mid vowel /q/, the first preceded by an onset /sr/, and the two
separated by an interlude /m/. This /m/ goes with both the preceding and the
following peak in a way which is, phonetically as well as structurally, much like
the way the medial /m/ of English hammer goes both with the preceding stressed
vowel and the following unstressed vowel.
Mandarin and they are not numerous; in the
interludes are all fairly simple,
flow of speech /m/
probably the most frequent. Furthermore, there are ap-
is

parently never more than two peaks in a single microsegment. Despite this more
complex structure, it is possible, as in Cantonese, to predict the location of peaks
in terms of the location of microjunctures and the intervening sequences of
contoids and vocoids. Thus there is a class of two vowels, /a a/, which are always
peaks; there is a class of semivowels /i u r u/, which occur as peaks in syllables
containing no vowels, but as parts of onsets or of complex peaks when flanked by
a vowel; there is a class of semiconsonants /s c c'/ which occur as onsets with a
distinctive peak, but which also occur as the only constituent of a microsegment
save for the tone (if any), and which in the latter case are followed by an auto-
matic peak vowel of the [i] type; and there is a larger class of consonants, which
occur only as onsets, parts of onsets, or (in a few instances) as parts of complex
peaks.
There is some evidence to suggest that Mandarin has an internal open juncture
of
an opener variety than microjuncture call it, say, mesojuncture, since if it
really exists and occurs within macrosegments its occurrences break a macro-
segment into one or more mesosegments, each in turn composed of one or more
microsegments. But it is possible that a more thorough investigation of Mandarin
intonation will show that the difference between this intermediate type of open
juncture and either microjuncture or macrosegment-boundary, and the location
of apparent mesojuncture occurrences, are both subsumable within the intona-
tional picture. In any case, this possible mesojuncture does not materially bear
on our problem here.
English is another instance of a system with both internal open juncture and
interludes; the details differ greatly from those in Mandarin. In the first place,
English microsegments have no unity-marking constituent comparable to the
Mandarin tones. In the second place, there may be any number of peaks in an
English microsegment, from none up to perhaps six or eight, though the upper
extreme is not common.
A microsegment with no peaks at all does not occur as the only one in a macro-
 A MANUAL OF PHONOLOGY 63

segment. Such a microsegment consists of a single consonant flanked on both

sides by microjuncture. In She's going with us (said with emphasis on she), there
is most usually no microjuncture between the she and the 's (/z/), but the latter
is normally followed by one. But if we replace the pronoun by a proper name,
say Bea, giving Bed's going with us, then there is frequently (though not always)
a microjuncture both before and after the 's. A closer contrast is between this
latter and what normally happens in Bea's going with us was a surprise, where
the "genitive particle" /z/ added to Bea seldom if ever separated from it by a
is

microjuncture. In He owed me we have sequences /owd/ and /ad/,

ten-odd dollars
flanked by microjunctures; in one common way of saying He oh'd and ah'd we
have, instead, the sequences /ow-d/ and /ah-d/, so that the /d/ in each of the
latter instances is bounded by microjunctures. In yacht one has /ya /; in . . .

Well, y'oughta one often hears /y-o /.

. . .

We can say either that this marginal kind of microsegment contains no syl-
lables at all, or we can say that it consists of a single syllable of a special type,
the isolated consonant type. The latter is probably systematically preferable,
since it permits us to say that a microsegment includes one or more syllables;
the former, however, makes it easier to state some of the things that need to
be said about all other microsegments, and we shall therefore adopt it.
Every English syllable contains a peak, either simple (one of a number of
vowels) or complex (a vowel plus a semiconsonant /y w h/, perhaps in some
cases /r/). The vowels occur only as, or as part of, syllable peaks, so that, as
in Fox, syllables are defined by peaks and peaks by vowels. Initially in a micro-
segment a syllable may have a distinctive onset (man, snow, scrimp), or none
(each, ouch, apple). Finally in a microsegment, a syllable may have a distinctive
coda (man, scrimp, glimpse, sixths) or none (day, idea). Such a final syllable, if
stressed, has either a complex peak or a coda or both; if unstressed the peak
may be simple and not be followed by any coda.
Onsets consist of from one to three consonants and semiconsonants; the clusters
of two and three are relatively limited. Codas consist of from one to four con-
sonants (not semiconsonants): thing (1), think (2), thinks (3), jinxed (4). One

consonant which occurs as a coda /n/ does not occur as or in any onset;

another //, rouge is very rare as an onset. All the consonants which occur
as onsets, however, occur also as codas.
Interludes range from zero (layer), providing the peak of the preceding syl-
lable is complex, through one (manner, hammer, fatter), two (faster, milking,
asking), three (pantry, empty), to four (subscribe), and possibly, very rarely,
more. In the case of the simplest interludes, the contrast between an interlude
and a coda-onset sequence is quite clear: minus slyness, nitrate night rate
: : :

dye trade. With longer interludes the contrast is not so clear, probably largely
because the forms which may contain them are pronounced now with, now with-
out, a microjuncture at some point, even by a single speaker, with no particular
functional importance.
All interludes can be divided, in at least one way, into two successive parts,
the first of which occurs also as a coda, the second of which occurs also as an
64 A MANUAL OF PHONOLOGY

onset. Thus the interlude /m/ hammer occurs

also both as coda and as onset,
of
and can be divided either into /m/ and zero,
so does zero, so that this interlude
or into zero and /m/. Similarly, the /ntr/ of pantry can be divided either into
/nt/ and /r/ or into /n/ and /tr/. The /bskr/ of subscribe, on the other hand,
can be divided only into /b/ and /skr/, and the /ml/ of hemlock, which is at
least sometimes pronounced without microjuncture, can be divided only into
/m/ and As has already been pointed out, the possibility of such division
/l/.
does not mean that an interlude can be interpreted as a coda plus an onset;
thus the fact that the /tr/ of nitrate can be broken either into zero and /tr/, or
into /t/ and /r/, does nothing to eliminate the contrast between that interlude
and the coda-onset sequences of night rate and dye trade.
But phonetically the divisibility has an important implication. In interludes
where only one division of the kind described above is possible, the phonetic
point of syllable division is quite clear: everyone would agree that it falls between
the /b/ and the /skr/ of subscribe, between the /m/ and the /l/ of hemlock.
Where multiple division is possible, the phonetic point of syllable division is
obscure or variable. In pantry, one sometimes seems to hear the /n/ as syllable-
final and the /tr/ as initial, whereas at other times the point of syllable division
seems to fall in the middle of the /t/. Now, structurally, the point of syllable
division in an interlude is irrelevant. The purpose of this and the preceding
paragraph is to show that as a phonetic phenomenon, even if partly impres-
sionistic,both unambiguous, and ambiguous or variable, points of syllable divi-
sion at interludes can be accounted for in structurally relevant terms.
Other Germanic languages have syllable patterns largely similar to that of
English, though English interludes and codas perhaps become somewhat more
complex.
Some of the Slavic languages also have partially similar systems. In Russian,
some microsegments, not
at least, the similarity reaches the point of having
standing alone in the macrosegment, which consist of an isolated consonant;
e.g. /v-g6rad yi/ 'in the city', /v-akn6/ 'through the window', /s-utk y i/ 'from a
duck', /g-g6radu/ 'towards the city'.
Finnish has a microjuncture with a special kind of phonetic marking: micro-
juncture is a point of onset of loud stress. Every macrosegment thus begins with
loud stress, and any loud stress in the middle of a macrosegment starts at a
definite place and constitutes an occurrence of microjuncture. By "at a definite
place" is meant that in a medial sequence of vowel, consonant (say /n/), vowel,
there are three clearly contrasting situations: .VnV. ./, with no loud /. . .

stress during the particular stretch, /. where loud stress begins

. .V-nV. . ./,

with the /n/, and /. .Vn-V. ./, where loud stress begins after the /n/, with
. .

the second vowel. Microsegments consist of one or more vowel-peak syllables;

onsets (by definition involving microjuncture) are relatively limited, codas
even more so, but interludes achieve some complexity. Apparently the phonetic
point of syllable division in an interlude is normally quite clear; but it is not
possible, as for English, to divide every interlude into a part which occurs also
as a coda and a part which occurs also as an onset.
 A MANUAL OF PHONOLOGY 65

23. Accentual Systems. At a number of points during the discussion of 22,

we had occasion to refer to features present in every macrosegment of a language
(or in all save a few clearly marginal ones), or to a feature or type of structure
which occurs only in macrosegments which contain also some other given feature
or structure. Our aim now, for the moment, is to discuss this sort of thing in
greater detail.
The syllables of a system often fall into a number of mutually exclusive
structural types: e.g., in Bella Coola, syllables may have only an onset, or they
may include peak or coda or both; in English a syllable bears loud stress, or
medial stress, or none at all. If, in any such case, a whole macrosegment may
consist exclusively of (one or more) syllables of one type, that type is isolable;
otherwise it is non-isolable. A type is strongly isolable if a whole macrosegment
may consist of a single syllable of that type. In a language which has microseg
ments larger than syllables but smaller, on the average, than macrosegments
the same terms apply to mutually exclusive structural types of microsegment
Thus in Bella Coola both onset syllables and those involving other material
are isolable: witness the macrosegments /lk' w tx w / 'make it big!', /kac'an/
'tail'. But onset syllables are not strongly isolable, and it is doubtful that peak-

or-coda syllables are. In Nootka, if we interpret the system as of the onset-

syllabic type (2213), syllables with a peak are isolable, perhaps strongly so,
but syllables consisting only of an onset are not. In Yawelmani, where all syl-
lables have onset and peak, both the type with a coda and the type without are
isolable. In English, syllables with loud stress are strongly isolable, but those
with medial stress or with none are not isolable at all. In Cantonese, where all
syllables can be classed according to which tone is present, apparently all the
types are isolable. But in Mandarin, where some macrosegments have no tone,
only the type with a tone is isolable.

In some cases of this kind, we are confronted with an accentual system. In

other cases we are not. It seems very difficult to define formally the fine of
demarcation between the two situations. We should like a definition of "ac-
centual system" which would include the tones of Cantonese and Mandarin,
the stresses of Mandarin, English, Spanish, and perhaps the syllable peaks of
Nootka. But we should not like to include also the difference between micro-
segment-initial syllables with and without an onset in English, although those
constitute two clearly distinguishable and mutually exclusive structural types,
both of them strongly isolable. Whether or not the tones of Apachean, Mazateco,
Senadi, or Bariba should be subsumed under the definition seems uncertain.
The tones of these languages are encountered at a later point in the progressive
IC-analysis from large to small: they are not ICs of syllables as wholes. But
this is true also of the stresses of English and Spanish, which we definitely want
to include. Any definition of "accentual system" in terms of the kind of articula-

tor material involved say as systems involving contrasts of pitch, volume, or

duration would be of little use, since intonational systems involve such raw
material, and systems of the kind we want to call "accentual" sometimes in-
volve other types of raw material.
66 A MANUAL OF PHONOLOGY

Instead of attempting to give a formal definition of "accentual system,"

therefore, what we
shall do is to give, classified by several criteria, a number
of examples systems which probably ought to be subsumed under any defini-
of
tion finally formulated, and leave the formulation itself to the future.
We classify the systems to be described as linear and non-linear, and as systems
with or without zero.
231. Linear Systems. In a linear system, the contrasts which are involved
constitute points or regions along some single scale of articulatory or acoustic
property: for example, two or more different pitches, or two or more degrees of
prominence or stress, or, conceivably, presence versus absence of nasalization,
glottalization, or the like.

2311. Linear Systems with Zero. If one of the contrasting elements of a linear
system is isolable and the others are not, the system falls into this classification.

Beyond can be made into two-termed, three-termed, and so

this, classification
on (if there are in fact any systems of this kind with more than three terms),
depending on the number of contrasting points along the scale.
Spanish and Italian, for example, have isolable syllables with loud stress, and
non-isolable syllables with soft stress. Many languages are like Spanish and
Italian; others are like English and German, which have isolable loud stress,
non-isolable medial stress and soft stress.
In the case of a two-term system, we can define one of the two as involving
the presence of a particular structural unit, and the other as marked by the
absence of that same unit. In a three-termed system, we can define two of the
three as involving the presence of one or the other of two different units, and
the third as marked by the absence of both. The usual way to do this is to say
that loud stress in a language like Spanish is a positive unit, while soft stress is

its absence; or that loud and medial stresses in a language like English are
positive units, soft stress their absence. If we mark loud and medial stress by
accents /'/ and /V over vowels, then our marking stands in direct relationship
to this usual interpretation. Such a notational convention usually has the ad-
vantage of minimizing the number of accent marks which have to be written,
since unstressed vowels, in Spanish for example, are considerably more frequent
than stressed ones. And it is because of the possibility of such as interpretation
that we speak of "linear systems with zero."
However, it is also perfectly possible, and for some purposes more useful, to
regard the absence of loud stress as the positive entity, rather than the other
way around. (The notational conventions described above can still be retained:
they simply stand in a less direct relationship to the phonologic interpretation.)
In English, for example, a word such as cat as a whole macrosegment would then
consist entirely of its vowelsand consonants only in polysyllabic macrosegments
;

would the positive entities of secondary and weak stress make their appearance.
One advantage of this alternative is that accents as positive entities would only
be mentioned (and transcribed, if transcription followed phonologic interpre-
tation) in those environments in which contrasts are possible. There is no second-
arily or weakly stressed cat, as a whole macrosegment, to contrast with the one
 A MANUAL OF PHONOLOGY 67

with loud stress: secondary and weak stress are non-isolable. There are also, as
ithappens, some clear advantages of a morphophonemic and grammatic nature,
but we cannot strictly be concerned with them in the present manual.
tempting to propose a general methodologic rule at this point to prescribe
It is :

that, in any linear accentual system with zero, the isolable, or strongly isolable,
member be the one which is zeroed out. In our treatment of English in this
monograph we shall do this, and our phonologic notation for English will accord
with the interpretation: /a/ means the vowel with loud stress, /a/ the vowel
with medial stress, and /a/ the vowel with soft stress. But we are not actually
proposing the suggested methodologic legislation, because it involves a degree of
arbitrariness which we want to avoid. The typologic fact of importance is only
that in a linear system with zero such "zeroing out" of one term is realistically
possible, whereas in a linear system without zero it seems not to be; this will be
further demonstrated shortly.
In either a two-termed or a three-termed system, the occurrent arrangements
of the various types of syllables so differentiated may vary a great deal. In English
there is at least one loud stress in a macrosegment, and there may be more; there
is at most one loud stress in a microsegment, but there may be none. A micro-

segment may contain secondary stresses whether or not it also contains a loud
stress, and also weak stresses whether or not it contains either secondary stresses
or a loud stress. The possible arrangements are very numerous, though not
unlimited.
In Spanish there is at least one loud stress in a macrosegment; if the macro-
segment is four syllables in length or longer, then at least one of the last four
contains a stressed vowel, and most usually it is one of the last three. More than
two stresses in immediate succession is very rare, if it occurs at all, but two is

fairly common. Unstressed syllables are satellites to stressed syllables: those

preceding the first stress of the macrosegment are satellites to that one, and
those following the last stress of the macrosegment are satellites to that one,
but those between two successive stresses cannot be definitely assigned, on the
phonologic level, to either. Rather, they go with both, like the planets of a

double star.
In some languages the syllable of a macrosegment is always stressed;
first

in others the last, or the last but one, always bears stress. The location of other
stresses, however, remains distinctive. Finnish is not a system of this kind, since
in Finnish loud stress is part of what constitutes microjuncture. But some
languages differ from Finnish only in the following way: every macrosegment
begins with stress, and some stresses occur medially, as in Finnish; whereas in
Finnish there is a medial contrast between /. .V-nV. ./ and /. .Vn-V. ./,
. . . .

with the second V stressed, in these other languages there is no such contrast,
the consonant immediately before a stressed vowel necessarily syllabifying
with it. However, even in such a system as this we are at least able to say that
unstressed syllables are uniformly satellites to the nearest preceding stressed
syllable.
Often enough a system has predominantly a stress on every macrosegment-
68 A MANUAL OF PHONOLOGY

on every macrosegment-final syllable, or in some other statable

initial syllable, or

position, but with some small numbers of exceptions. Thus Hungarian always
has a stress either on the first or on the second syllable of the macrosegment,
and the former is the case in a much larger number of different macrosegments,
though the text-frequency of the latter is not particularly low.
2312. Linear Systems without Zero. If neither (or none) of the contrasting
elements of a linear system is isolable, or if both (or all) are, then the system
But these two situations need to be differentiated.
falls into this classification.

One may imagine a language with two stress levels in which neither is isolable
because the shortest macrosegment consists of two syllables, one of each kind.
Logically this situation is different from that in a linear system with zero, but
one suspects that in any such case the system might work, in all other respects,
like one of the latter. The cases with which we shall deal here are all of the other
kind: those in which both or all of the contrasting elements are isolable. Here,
as in the case of systems with zero, a further classification can be made into
two-termed, three-termed, and so on.
If we choose to interpret Apachean tones as an accentual system, then they
constitute a two-termed system (low versus high pitch) without zero. High tone
is strongly isolable: a macrosegment may consist of a single one-vowel syllable

with high tone, as (Chiricahua) /to/ 'water'. Low tone is likewise strongly iso-
lable: /li/ 'smoke'. In longer macrosegments we cannot regard occurrences of
either tone as in any way satellite to some occurrence of the other tone. How-
ever, there is doubt as to whether Apachean tones ought to be interpreted as
constituting an accentual system. Precisely the structural statements just made
about high and low tone would apply also to the contrast between high vowel
and low vowel, or to that between front vowel and back vowel: it is probably
preferable to regard the two tones, the two tongue heights, and frontness versus
backness of tongue position, as three contributing factors in the differentiation
of vowels. The only reason to doubt this is that the two tones do also occur as
differentiae with syllabic nasals, whereas the tongue positions occur only for
the vowels. On the level of whole syllable peaks, of one or two units length, it
would be equally valid to interpret oral versus nasal as an accentual system of
the two-termed type without zero.
Loma is representative of a large number of languages with a two-termed
accentual system without zero: the terms are high and low tone, as in Apachean,
but seemingly these appear at the level of syllables rather than at the lower
IC-level on which they function in Apachean. Senadi and Mixteco have three-
termed systems (low, mid, and high tones); Mazateco has four; Trique is re-
ported to have five, which is, so far, the maximum.
It is interesting to note that linear systems without zero seem uniformly to
involve tone, rather than stress, as the predominant raw material. There may
be fundamental reasons for this. Pitch levels, as used in phonologic systems,
are always relative matters: a woman's pitches are all higher than a man's, and
even in one person's speech some structurally low pitches can be higher than
some structurally high ones. Yet there is such a thing as "absolute pitch" in the

musical sense the ability of an individual to identify in terms of the musical
 A MANUAL OF PHONOLOGY 69

scale, and thus indirectly in terms of number of vibrations per second, a tone
that he hears, or to produce a tone of a prescribed pitch quite accurately. Differ-
ent voices, and even different concomitant circumstances of the kind that may
lead a speaker to raise or lower his whole register of pitches, fall into recognizable
categories,and a syllable of a tone language pronounced in isolation can usually
be identified as to its structurally relevant pitch level, not only by another
speaker of the language, but even by analysts who have had extensive experience
with languages of the kind. The loudness scale is apparently much more purely
relative, not only in speech but generally. The relative loudness of a syllable, or
of any other sound, can be measured and judged only relative to the noise-level
of the background in which it is presented. However, whether these considera-
tions prove ultimately relevant or not, it is none the less a fact that stress sys-
tems, so far as is now known, are always of the type with zero, while pitch
systems never are.
From a purely logical point of view it would be possible, in these systems as
in those discussed in 2311, to "zero out" one of the contrasting terms. As a
matter of notation, it is certainly possible to omit any positive mark for one of
the contrasting terms. In some cases even other notational conventions are
used, in order to economize on diacritics. Thus in some two-toned languages of
West Africa investigators have adopted the convention of marking the first

high tone in a sequence flanked by spaces in transcription (the spaces are usually
nonphonologic, which does not concern us here) with an acute accent mark, and
of writing acutes and graves thereafter only on a syllable which differs in tone
from the preceding one. A graphic segment all on low tone then needs no mark;
one all on high tone, or with no low tones after its first high tone, needs only
one mark.
But we must distinguish between possible notational devices and the phono-
logic pattern which is represented by them. In a linear system without zero
there is no structural basis for deciding which terms are positive, which term
negative: one term of the set is just as positive as another. This is all the meaning
that need be read into our labels "with zero" and "without zero." And this
difference remains even for those analysts who, following slightly different pre-
dilections, insist that English weak stress is just as much a "positive" phonologic

entity as medial or strong stress.

232. Non-Linear Systems. In a system of this kind the contrasts cannot be
lined up along some single scale of articulatory or acoustic property. Sometimes
two or more points along one such scale are involved, together with differentia-
tions of some other kind that partly intersect. If the intersection is complete or
nearly so, then instead of speaking of a single non-linear system it may be better
to speak of two intersecting systems, each linear or not as the case may be. But
complete intersection is rare.

A non-linear system with zero is one which involves contrasts along one scale,

at least, some ofwhich are non-isolable; that is, such a system looks like a partial
intersection of a linear system with zero and some second system. A non-linear
system without zero is then any other kind of non-linear system.
2321. Non-Linear Systems with Zero. In many Scandinavian dialects, including
70 A MANUAL OF PHONOLOGY

standard Norwegian and Swedish, there an accentual system largely reminis-

is

cent of that in English or German: and weak stresses, with much

loud, medial,
the same privileges of occurrence relative to each other as in the latter languages.
There is, however, a further complication loud stresses are of two types, differ-
:

ing in tonal contour. Loud stress with "simple" tonal contour (the phonetic
details vary from one form Scandinavian to another) is strongly isolable loud
of ;

stress with "compound" tonal contour is not isolable at all. Every macroseg-

ment includes at least one loud stress, but this may be that with either tonal
contour, save in the case of a monosyllabic microsegment, which, as already
indicated, will have the simple contour. In Norwegian, /bo-ner/ 'peasants' has
the simple-contour loud stress on the first syllable; phonetically the stress is

accompanied by a pitch lower than that on the following weak stressed syllable,
/bo mar/ 'beans' has the compound-contour loud stress on the first syllable;
phonetically the stress is accompanied by a downwards glide in pitch, starting
and ending at least somewhat lower than the pitch on the following weak stressed
syllable.
A system with zero can be graphed as shown on the left below; the sys-
linear
tem chosen is that of English, and parentheses mean non-isolability. The Scandi-

navian non-linear system with zero can then be graphed as shown on the right:
/
'
n
o
o o
iglish Norwegian

The transverse contrast at primary stress level in Norwegian does not force us
to "zero out" weak stress; we can still proceed as in English (2311), taking the
strongly isolable ['] as phonologically nothing, the other three as positive entities.
The type of contrast between ['] and /*/ still remains different from that among
['],/Y, and /V-
Such a diagram is the best way to present the much more complex stress-and-
tone system of Mandarin, though some of the symbols require description. For
the stress levels we use, in this diagram, the marks used above for English stress
levels. For the tones we use, here, superscript numbers, to the right of the stress
marks. / / is a level tone, not low; / 2 / is a rising tone, not low; / 3 / is a low tone,
l

rising terminally in some environments; / 4 / is a falling tone; / 5 / is a rising tone

which starts and ends lower than / 2 / and is accompanied by a diminuendo of
force, rather than the crescendo which normally accompanies / 2 /. /V is quite
limited, and for some speakers does not contrast with / 3 / or / 2 /. Absence of any
of these tones is represented by //:

'1 '2 '3 '4

C5)
CD 02) (
v
3) 04)
v
05)
y v
( i) ( 3) ( 4) 00)
It should be remembered that the domain of a tone in Mandarin is the micro-
segment, not the syllable; this is true also of the stresses.
 A MANUAL OF PHONOLOGY 71

Burmese has a simpler non-linear system with zero. There are four tones, and,
in context only,some toneless syllables, but apparently no stress contrasts.
The system is non-linear because the tones are not simply points along the
scale of pitch, but rather, as in Mandarin, fairly complex contours of level, rise,
and fall, the latter at different tempos. A diagram of the system would look
like this, where we number the tones arbitrarily:

12 3 4
(0)

the non-isolable weak type is put in a different line because of the non-linearity
of the system.
Another system of this kind is found in Taos, where the situation resembles a
partial intersection of two linear systems, one without zero (three pitch-levels)
and one with (three stress-levels). Numbering the pitches from bottom to top,
we can diagram thus:
'1 '2 '3

CD 02) 03)
CO)

The presence or absence of a syllable peak in Nootka can be interpreted as

an accentual matter; if so, then the different vowel qualities constitute an inter-
secting system without zero. We can interpret vowel length as part of the ac-
centual system too, and thus arrive at the following diagram:

-
i- e* a o* u-
i e a o u
(0)

Syllables of the non-isolable type (withno vowel as peak) occur only in se-
quences of one or more preceded by one of an isolable type, so that they can be
regarded as satellites thereto.
The point of this last example is not, of course, to insist on the interpretation,
but simply to show that the structural situation, despite its manifestation in
markedly different phonetic material, is quite comparable to obvious instances
of accentual systems.

2322. Non-Linear Systems without Zero. In Tangsic (a Wu-type dialect of

Chinese) there are three tones: level, rising, and falling. All are isolable; their
domain, as in Cantonese, is the whole syllable; also as in Cantonese, there are
no toneless syllables.
In various other languages one can find three, four, five, six, or more con-
trasting tones, all or most of them isolable, constituting non-linear systems with-
out zero. Bariba has five (22112); Vietnamese has six; some varieties of Can-
tonese have eight, others nine.
In any such case the possibility must always be considered of the sort of re-
duction achieved for Mazateco (2212). There, if one were simply to count the
72 A MANUAL OF PHONOLOGY

total number of tonal contours which occur distinctively on the peaks of syllables,
one would end up with a system of fourteen tones. Four of these are level; eight
more are rising or falling; two involve a rise followed by a fall. But the pitches

at the endpoints of the rises and falls can be identified with the four level pitches,
reducing the system to one of four tonemes which occur not only alone but also
in sequences of two and three.
This sort of reanalysis will not work for Tangsic, since there two of the tones
involve change of pitch, and only one is level. Nor would it work conveniently
for Bariba, since although there are three level tones to which the endpoints of
the rising and falling tones could be related, there is contrast between the rising
or falling tone, on a single vowel, and a sequence of two level tones, on two suc-
cessive vowels: the rising and falling tones are irreducible because they last the
same length of time as any of the level tones and contrast with (reducible) con-
tours which last longer.
There are also more complex systems for which such a reduction is, if not im-
possible, at least of no particular profit; and this seems to be true even of such
a many-termed system as Cantonese. However, it is quite possible that further
investigation into such more complex systems will change our interpretation.

24. Syllable Peak Systems. All languages have at least some syllables with peaks
composed of distinctiveand contrasting material. Quite apart from accentual
matters, the syllable peaks of a language constitute a system of one kind or
another, and the investigation of such systems is our program in this section.

First we shall deal more systematically with the matter of simple and complex
peaks: these terms were occasionally used in 22, but require refinement. Then
we shall be ready to introduce the term (segmental) phoneme and labels for struc-
tural classes thereof. Then we shall survey the various kinds of peak systems
which involve complex as well as simple peaks. Finally we shall classify vowel
systems.
241. Simple and Complex Peaks. In an English microsegment like bet, we
say that the onset is /b/, the peak /e/, and the coda /t/. In bent, we have the

same onset and peak, but coda /nt/. In bait /beyt/, however, we do not speak
of peak /e/ and coda /yt/, but rather of a complex peak /ey/, and coda /t/
as in bet. And in bay /bey/ we recognize the same complex peak, with no coda
at all. Similarly, in microsegments such as baying /beyin/, bating /beytirj/,
painting /peyntig/, we do not recognize interludes /y/, /yt/, and /ynt/ after
a simple peak, but instead speak of a complex peak /ey/ and interludes zero,
/t/, and /nt/.
The choice in such cases stems from a simple and well-known criterion in the
analysis of hierarchic structures. We investigate both possible cuts between
peak and coda, or peak and interlude, and see what the results would be if either
were adopted definitively. Suppose that in all English syllables we were to cut
immediately after the vowel. The codas (and interludes) thus tentatively recog-
nized would fall into four sets: those beginning with /y/, those with /w/, those
 A MANUAL OF PHONOLOGY 73

with /h/, and those with none of those three units; and the members of these
four sets would line up, with random exceptions, like this:

lj*l /wz/ /hz/ //

faze rose gauze 18

/yt/
beet
/wt/
loot
/w
ought
N
it

/ynd/ /wnd/ /hnd/ /nd/

fiend loaned spawned eric? .

On the other hand, we would discover the most whimsical limitations as to what
peaks were followed by what codas (or interludes). Peak /i/, for example, would
turn out to be followed quite freely by codas of the type listed in column four

above, and by codas beginning with /y/, but hardly ever for some speakers

never by codas beginning with /w/ or /h/.
If, on the other hand, we assign postvocalic /w/, /y/, and /h/ uniformly to

the preceding vowel as part of a complex peak, then the limitations on peak-coda
or peak-interlude sequences become far less whimsical. It is still perfectly true,
of course, that the sequence /iw/ or /ih/ is rare or nonexistent: analysis cannot
fill a pigeonhole which is in fact empty, or force symmetry where there is none.

But such limitations can now be covered in a simpler way. Instead of having to
make long lists or complex descriptions of codas which do not follow certain
peaks, one can describe all the limitations within the discussion of peaks them-
selves.
The criteria do not always give such clearcut answers. In the Middle-Western
American variety of English, postvocalic /r/ presents a problem. Something is
gained by considering the /ar/ of fur, first, furred, firze, Bert as a complex nucleus
rather than as involving codas beginning with /r/. But there is less neatness
in the overall picture when this is done than there is in the case of postvocalic

/y/> Ml an d /h/. Furthermore, the case is not at all convincing for other
>

sequences of vowel and /r/. Again in microsegments like cute, pure, few, spew,
one may hesitate between dividing into onset /kpf sp/ and peak /yu. ./, or .

onset /ky py fy spy/ and peak /u. ./. The former suggests itself because /y/
.

does not occur after a microsegment-initial consonant before all vowels, but only
(or largely) before /u/. On
the other hand, assigning the /y/ to the onset has
the advantage of pointing up the partial parallelism between onset clusters
with last member /y/, /w/, /r/, and /l/.
Such vacillations making a cut between ICs are
between alternative ways of
not unusual. They stem not solely from inadequate methodology (though there
is always room for further sharpening and clarification), but at least in part

from the tremendous complexity of the object of analysis a kind of object in
which we have no right to expect absolute neatness of pattern.
74 A MANUAL OF PHONOLOGY

In Russian, sequences such as /ay ey oy/ occur as they do in English. But

the same criteria give a different answer. The "privileges of occurrence" of such
sequences in Russian do not particularly resemble those of simple nuclei like
/a e o/; they are much more like those of such sequences as /et em ap an/.
Therefore we do not speak of complex peaks in Russian; instead, we assign
postvocalic /y/ to an interlude or coda.
In some cases it does not seem to matter which way we interpret the evidence.
In Cantonese, such syllable-final postvocalic consonants as /p k/ set us up with
t

a phonetic prejudice in favor of interpreting /. .ap/, /.

. . .ak/ and
.at/, /. .

the like as peak plus coda, rather than as complex peak: having made this
decision, then other post-peak elements must also be called codas (e.g., the
/m/ of /. .am/ or the /i/ [y] of /. .ai/). In Mandarin, our preference is rather
. .

for treating the second elements of prejunctural /ai au ar an an am ai au ar

an an am in in tin un/ uniformly as parts of complex peaks there are no syl-
lable-final stops. But in both cases either terminology would do, since there is
no contrast. (In 22 we followed the preferences just indicated.)
Distinct from these instances where the choice of terminology does not matter
are cases where, logically, the question should simply not arise. For example,
the presence in Fijian of articulations which sound to our ears like long vowels
(such as [a ]), or like diphthongs (such as [ay], does not present us with a problem
-

of complex peak versus peak-plus-coda. The status of these segments is solved

by logically prior criteria: they turn out to consist of successive vowel units,
each one of which is the peak of a syllable. In the actual chronology of analytical
work, of course, one may be led to a temporary assumption that things like
[a] and [ay] in Fijian are within single syllables, and so attempt to determine
whether they are peaks plus codas or complex peaks. But in due time their
structurally disyllabic status should become clear, and the irrelevant question
be abandoned.
242. (Segmental)Phonemes and their Structural Types. As most usually used
in American term "phoneme" subsumes several classes of units:
linguistics, the
junctures, accents, and intonational features such as PLs and TCs, are all called
phonemes, along with units which are of the vowel-or-consonant type. Phonemes
which are not junctural, accentual, or intonational are grouped together as
"segmental." Segmental phonemes are structural units of larger average size
than ultimate phonologic components, but smaller average size than syllables.
Generally speaking, a "simple" onset, peak, coda, or interlude consists of what
is ordinarily called a single "segmental phoneme," whereas a complex one con-

sists of two or more. Thus in English /bet/, the onset /b/, the peak /e/, and
the coda /t/ are all single "segmental phonemes"; in /strerjkfl/ the onset and
the coda each consist of three "segmental phonemes."
European usage sets up essentially the same sorts of different classes of units,
but the term "phoneme" is used in a different way. The European "phoneme"
equates, by and large, with what in America is called a "segmental phoneme,"
and American "suprasegmental" or "non-linear" phonemes are given other
terms.
In this manual, we shall follow the European usage, not because it necessarily
 A MANUAL OF PHONOLOGY 75

has any greater merit for linguistic purposes in general, but because the particu-
lar nature of our discussion renders it more convenient to use the one- word term
"phoneme" instead of the longer phrase "segmental phoneme." We have already-
dealt with all the phonologic features of the sort which in American usage would
be called "suprasegmental" or "non-linear" (21, 22, 23), and when subse-
quently we have occasion to refer to them it will always be possible to speak
specifically of a juncture, an accent, a PL, or the like. Furthermore, our cover
term "ultimate phonologic constituent," taken in the context of the thorough-
going IC approach which is here being attempted, and in the context of a sort
of relativism of phonemic units (323), tends to render irrelevant most of the
arguments by which proponents of either the American or the European usage
have supported their ways of speaking.
The phonemes of a language fall into various structural classifications, based
on similarities and differences of privilege of occurrence. It is impossible to
supply any general classificatory frame of reference from which terms can be
drawn in a completely consistent way for the discussion of every individual
language. But some approach to this can be attempted. In a language which
has only simple syllable peaks, it is usually important to know whether a given
phoneme occurs only as a peak, only otherwise, or as a peak in some environ-
ments and otherwise in others. Instead of "otherwise," we can say "as margin,"
which is then shorthand for occurrence as, or as part of, either onset, coda, or
interlude. General terms for the possible types of phonemes in such a case are
the following:
occurs as:
peak margin
vowel yes no
consonant no yes
semivowel yes yes .

Fijian phonemes are all vowels or consonants; Senadi has also semivowels
/m n n y rj/.
If a language has both simple and complex peaks, then a given phoneme in a
given environment may
be a simple peak, or the nucleus of a complex peak, two
situations which we will class together; or it may be a satellite in a complex
peak; or it may be marginal. A given phoneme may be restricted to one of these
three functions, or, in different environments, it may have differing functions.
The following set of terms includes three neologisms, of which one at least is

not particularly apt; but some terms are needed:

occurs as:
peak peak margi
nucleus satellite
vowel yes no no
covowel no yes no
consonant no no yes
semivowel yes yes no
demivowel yes no yes
semiconsonant no yes yes
omnipotent yes yes yes
 ;

76 A MANUAL OF PHONOLOGY

English has vowels consonants (e.g. /ptm n/), and semiconsonants

(e.g., /i e/),

(/y w h/). German is but has only two semiconsonants, /y h/; /w/, as
similar,
in flaws /haws/, is a covowel. Mandarin has vowels /a 9/, consonants /p t/
etc., semiconsonants /m n n/, a demivowel /ii/, and three omnipotents /i u r/.
(Our generalized terminology, it will be noted, supplies no term for the special
class /s c c'/ mentioned in 2222.) Spanish has vowels /e a o/, consonants, and
semivowels /i u/ providing we interpret such sequences as /ei/, A^/, /u^i/
(rey, tierra, buey) as complex peaks.
This terminology inadequate for a few rare systems in which one finds
is

what might be an example (Totonac) will be described

called hyper complex peaks:
in the next section. It is likewise unsatisfactory for systems in which complex
peaks involve a coordinate construction of two units rather than a construction
of nucleus and satellite: this is the interpretation which seems preferable for
the double vowels (and triple vowels) in Mazateco (2212). But in such cases
a satisfactory terminology can usually be obtained by a slight extension of mean-
ing of the term "vowel," throwing out the terms "semivowel" and "covowel."
243. Systems with Complex Peaks. If complex peaks involve nucleus and
satellite, then three situations can be distinguished in theory: those in which a
satellite always follows the nucleus; those in which it always precedes; and those
in which it either precedes or follows, though not necessarily both in the same
complex nucleus. The second of these does not seem to be attested. A third
actually attested situation has coordinate construction in complex peaks.
Some systems can be interpreted in more than one way. Many systems have
"long vowels," for example, which can be interpreted either as short vowel
phoneme plus satellite length, or as a coordinate construction of two identical
short vowels in sequence. Sometimes there is clear indication as to which of
these interpretations is preferable, but sometimes there is not. When there is
not, then the fact that either interpretation is possible is the typologically
relevant consideration. However, it is easy to jump to the conclusion that choice
of interpretation is irrelevant when a more thorough examination of the facts
would show that it is not. Such conclusion-jumping is naturally to be avoided;
when, in the examples given below, there seems to be reasonable doubt, this is
indicated.

2431. Satellites Follow Nucleus. An is that with

extremely widespread system
exactly equal numbers of simple and complex peaks,
which the phonetic in
difference lies primarily in quantity: simple peaks are short, complex peaks are
monophthongal and long. In such a system there is approximate matching of
vocoid color between a paired short and long, though there may be minor differ-
ences. For example, the four short vowels /i e a o/ of Fox are roughly [i e a u]
the corresponding longs are [i- ae- a- o-]. The proper pairing is quite obvious:
two of the shorts are high, and two low; two of them are front, and two back;
and the same classifications apply to the longs. The short high front and the
long high front, then, belong together.
This still does not tell us, however, whether [r] should be interpreted as /i /, -

with a covowel //, or as /ii/, a presumably coordinate construction of identical

 A MANUAL OP PHONOLOGY 77

vowel phonemes. If the former interpretation can be justified, then Fox belongs
in the category now under discussion complex peaks with satellite following
nucleus. If it cannot, or if there is no more evidence for one interpretation than
for the other, then some other classification is preferable.
If we take the longs to be the shorts plus a covowel phoneme of length //,
then the latter has what can be described as two phonetic shapes. Since [i-]
differs from [i], which we are asserting it contains as one constituent, by being
longer and higher, the phonetic shape or allophone of // when it accompanies
[i] is that of a raising lengthener: both the greater height and the greater length
of the longpeak are assigned to the //. But with the other three shorts, //
appears as a lowering lengthener, since [se- a* o*] are not only longer but also
lower than the corresponding shorts. We are justified in saying that the //
"follows" the vowel it accompanies, not just in terms of symbolism but also
structurally, since there is one structurally definable position final in nor-
mally-intoned macrosegments in which only the shorts occur, not also the
longs. By saying that // follows the vowel, we are able to state that the limita-
tions on consonant-vowel sequence are the same everywhere. If we wanted to
interpret the covowel as preceding the vowel, then we would have to say that
a consonant-covowel-vowel sequence occurs only non-finally, and the limitations
on sequence (e.g., the non-occurrence of /y/ before both /if and /i*/) would
be harder to state. It is not // which does not follow /y/; it is /i/, whether
itself followed by // or not.
we take the longs to be geminate (coordinate) clusters of vowels, then it
If
is important to note that only geminate clusters of vowels occur as peaks, not
also clusters of different vowel phonemes. Each vowel, in this interpretation,
has two allophones: /i/ is somewhat higher when flanked by itself, on either
side, than when not; the other three vowels are all somewhat lower as members
of geminate clusters than otherwise.
Still a third interpretation is possible: the contrasts of length of nucleus can

be taken as accentual. Since both short and long are isolable (though not strongly
so), the system would be a two-termed linear system without zero.
So far as the writer can see, there is no convincing evidence in favor of any of
these interpretations, and if this is true a typologic fact
then that in itself is

about Fox. An uncritical reading of the literature would yield far more examples
of this type of system than are actually to be so classified. No English dialect is
of this type; nor is German, nor any other modern Germanic language; nor is
Latin or Greek or Sanskrit; nor is Czech or Polish or Finnish; nor is any Poly-
nesian dialect, nor Fijian; although the literature on many of these languages
and families of languages is full of talk of "short and long vowels." But even
when taken there are still many genuine instances: Alabama,
greater care is

Badaga(?), Chitimacha, Choctaw-Chickasaw, Chontal(?), Creek-Seminole,

Delaware, Fox, Hitchiti-Mikasuki, Hopi(?), Hungarian(P), Karok, Koasati,
Kota, Kutenai, Miwok (Sierra and Western), some Nahuatl (Matlapa), Nootka,
Oneida, Seneca, Shawnee, Tojolabal, Tonkawa, Tubatulabal, Wichita(P),
Yokuts, Yuchi, Yuma, Zuni.
78 A MANUAL OF PHONOLOGY

Sierra Popoluca has amore complex, but very symmetric, system. A simple
peak is one vowels /i e a o u i/. A complex peak has one of these vowels
of the six
as nucleus, followed by one of two satellites: a covowel // or a semiconsonant
/?/. The parallelism of function of these two leads one to interpret long vowels
as short vowel plus a covowel, and to say that the covowel // structurally
follows its vowel. Virtually the same codas and interludes are possible after a
simple peak and after either type of complex peak. A single /?/ between vowels
isonset (or interlude), but /?/ after a vowel, before a consonant or at end of
macrosegment, is part of a complex peak. The following three examples show
minimal contrast and part of the parallelism: /?ihikpah/ 'he dries it up' :

/?ihi-kpah/ 'he pulls it' : /^ihi^kpah/ 'he permits it'. Note that coda /y/ can
follow complex peak with /?/ : /?fskuy/ 'eye' : /kni'i'ycih/ 'quail'.
Totonac has a more complicated system. Simple peaks consist of one of the
three vowels /i a u/. Complex peaks consist of any of these plus a covowel
//, satellite to the vowel. Hypercomplex peaks consist of a nucleus which is
either a vowel or a vowel plus //, and a following satellite /?/, which, as in
Sierra Popoluca, is a semiconsonant, since it occurs also as an onset. The codas
and interludes which can follow these various types of peaks are subject to
very few limitations in terms of what type of peak precedes. Phonetically, /?/
as a satellite in a peak colors the preceding vowel or vowel plus //, which is
pronounced with glottal stricture.
All the systems so far described in this section manifest a kind of symmetry
which is not by any means always to be found. For example, if we interpret the
Fox system with a covowel //, then the peaks can be displayed in a table:
simple complex
i i-

e e-

a a-
o o- .

Totonac peaks can be shown in a table which, though it has more pigeonholes,
still has all of them filled:

simple complex hypercomplex

i i* i? i-?

a a- a? a ,?
u u* u? u-f

But now let us consider Plains Cree. We begin with a table, which shows clearly
the lacks of symmetry, and then describe the phonetics:
simple complex
A v
i i i iy
v
a a ay
Y
u u uy .

The three vowels, as whole peaks, are approximately [i a u]. /*/ is a raising
v
lengthener (covowel), so that /i*/ is [i-]. a lowering lengthener (covowel),
/ / is
v v v
so that /i a u / are [e- a- o-]. /y/ is a semiconsonant: /iy ay uy/ are [iy Ay uy].
 A MANUAL OF PHONOLOGY 79

Or consider Menomini, which has six vowel phonemes and eight complex
peaks, and allows of three differing interpretations. Phonetically the vowels,
as whole peaks, are [i i e a u u]; the complex nuclei are [i- e- se- a- o- ir ia ua].

These can be lined up as follows:

simple complex
i r ia
e e-

e e-

a a-
o o-
u u- U9

in which // is a covowel with two allophones: a scalar lengthener (neither

raising nor lowering) after /i u/, and a lowering lengthener with the other four;
and /a/ is a covowel which appears only as a centering glide. Alternatively,
the second allophone of // could be grouped with the centering glide, which
still gives two covowels. A third alternative interprets the eight complex peaks
as two-vowel clusters: /ii uu/ and /ia ua/. This third is not very
ee ee aa oo
desirable, since it chooses thevowel /a/ for a special honor, and /a/ is not other-

wise special for example, /a/ is not a single low vowel, neutral as to front and
back, set off from the other vowels which come in pairs including a front member
and a back member, but is rather, specifically, the low back vowel, contrasting
with /e/ just as /o u/ contrast with /e i/. But between the other two interpreta-
tions there seems little choice. One sensible way to describe this situation is
simply to say that there are three satellite elements, only two of which con-
trast.
In one variety of standard German we find an interesting system which can
be diagrammed as follows (the peaks covered by the chart and description are
those in stressed syllables):
simple complex
i iy
e ey eh
a ay aw ah
ii iiy
6 oy
u uw
oy ow
The vowel phoneme /a/, which is low central unrounded but not extremely low
when it constitutes a peak by itself (as in Mann 'man'), occurs in three con-
trasting complex peaks: with a glide towards high-front-unrounded position in
mein 'my', with a glide towards high-back-rounded position in Haus 'house',
and with a lowering lengthener in man 'one'. This means that the system in-
volves at least three contrasting satellites in complex peaks. Since there is no
other case of greater than two-way contrast with a given vowel as nucleus, and
some cases of no contrast at all, there is leeway for various interpretations. The
interpretation indicated by the chart is as follows. We set up a satellite /y/
80 A MANUAL OF PHONOLOGY

which has two allophones. After /a/ the only allophone is that already described;
likewise after /o/. After /i e ii 6/ what one hears sometimes seems to be a glide

towards high-front position (rounded or unrounded depending on whether the

vowel is /i e/ or /ii 6/), but more often seems to be simply a raising lengthener:
/iy/ usually begins and ends higher than /[/ as whole peak, and so on for the
others. The ambiguity of hearing between these two allophones makes no
structural difference under the interpretation we have chosen. Since /y/ occurs
also as an onset, it is a semiconsonant. We set up a satellite /w/ which likewise
has two allophones: after /a/ only that already described, after /o u/ either this
or a raising lengthener, with the same ambiguity as for satellite /y/ after /i e 6 ii/.
This unit /w/ occurs only as satellite in complex nuclei, and so it is a covowel.
Finally, we set up a satellite /h/ which appears after vowels only as a lowering
lengthener; this unit, however, occurs as onset in forms like Hand, Haus 'hand,
house', and so, like /y/, is a semiconsonant.
The system of any one personal variety of English is apt to show asymmetries
reminiscent of those found in Cree, Menomini, and German. The writer's English
could be charted as follows:

simple complex
/i/ sit /iy/ seat /ih/ museum
/e/ set /ey/ sate /eh/ can (aux.)
/ae/ sat /aey/ ash
/i/ just (adv.)
/a/ cup /gh/ the (stressed)
/a/ cop /ay/ sigh /aw/ cow /ah/ spa
/u/ book /uw/ new
/of wash /oy/ boy /ow/ no
/of calm /oh/ law

All three satellites are semiconsonants : cf. their occurrence as onsets in yes,
hue, well, when, hot (/y hy w hw h/). The allophones of /h/ as satellite in com-
plex nucleus include scalar lengthener, lowering lengthener, and centering glide.
2432. Satellites Precede or Follow Nucleus. In Winnebago there are five vowels
which occur as simple peaks: /ieaou/. There are five homogeneous long vocoids
among the complex peaks; for reasons which we shall see in a moment, we in-
terpret them as geminate clusters /ii ee aa oo uu/. The other complex nuclei
are all considerably rarer, and a few may occur in addition to those of which
we are sure enough to list. The nucleus precedes the satellite in /ae ai oe oi ei ui/:
in these, /e/ and /i/ are glides not only towards, but to, the region of peak-nucleus
/e/ and /i/, but are less prominent than the preceding vowel. The satellite
precedes the nucleus in /oa ua ea ia uo eo io ie/: the second element is more
prominent. As our transcription implies, all of these complex nuclei are inter-
preted as sequences of two vowel phonemes. Very occasionally more than two
vowel phonemes occur in sequence: the point of division between the successive
syllables is always statable. Thus in the postconsonantal sequence /ioi/, the
first /i/ is a simple peak, the /oi/ a complex peak of the variety listed above.

Instead of classifying the vowels to account for their behavior as nuclei or as

 A MANUAL OF PHONOLOGY 81

satellites in clusters, we need only list the five of them in order of decreasing
"strength": /a o u e i/. If two different vowels are present in a peak, then the
"stronger" one is the peak, the other the satellite.
This picture is complicated by nasalization, which manifests itself in a peculiar
way. We shall represent nasalization by a superscript / n/ before the first vowel-
letter of a peak: the nasalization effects not only the peak but, in some cases,
the preceding onset or part of n
it. If the onset is /m/, then / / is necessarily

present. If the onset is /r/, then / n / may be present or not: the consonant which
we are representing as /r/ is [n] when within the domain of /"/, but [r] (a tap
or very brief trill) otherwise. The peak influenced by / n/ is nasalized as a whole,
whether it includes one vowel or two, except that /e/ and /o/, occupying any
position within a peak, remain oral. Thus there is no contrast between, say,
/ke/ and */k n e/: only the former occurs. Nor is there any contrast between
*/me/ and /m n e/: only the latter occurs, phonetically [me] (with nonnasal
n n
[e]). But /re/ and /r e/ contrast, phonetically [re] and [ne]; and /ka/ and /k a/,
n
or /kea/ and /k ea/, contrast, the former pair phonetically [ka] and [kaj, the
latter [kea] and [keaj. Possibly the contrast between nasal and nonnasal should
be interpreted as accentual in character.
It may be that Spanish should be classed here; the problem of cutting between
peak and onset, coda, or interlude is difficult (or meaningless?), but it is possible
that such sequences as /u6/ in bueno or /i6/ in viene, /au/ in ndutico, /u&/
in buey, should be interpreted as complex peaks. If so, then the satellite in a
complex peak is unstressed /i/ or /u/, occurring either before or after the nuclear
vowel, or both.
2433. Coordinate Construction in Complex Peaks. In a number of languages,
complex peaks seem best interpreted as involving coordinate constructions of
two units, usually members of the same structural class. In Taos the vowel
phonemes which compose simple peaks are as follows:
oral nasal
i u i v
e a e 9
a o Sk

These have been presented tabularly to show the lack of parallelism between the
oral and the nasal set. Complex peaks are all oral or all nasal, and consist of two
vowels in sequence: /ie ia uo ao [/ (rarely /ue/, perhaps only in unassimilated
Spanish loans). We are told that usually there is little difference in prominence
between the two vowels of a complex peak, and that a syllable containing one
takes only about as long to pronounce as one with a simple peak in the same stress
and pitch circumstances.
In Apachean (Chiricahua), as we saw in 2221, a syllable peak may consist of
a single vowel phoneme /i e a o/, oral or nasal, or of /n/ or /m/; a complex peak
contains a geminate cluster of any of these, or a non-geminate cluster /ea ai ei
oi ao eo/, oral or nasal as a whole. All the complex peaks are perhaps best inter-
preted as two phonemes in a coordinate construction.
Just as it is not always easy to decide, in the analysis of a hierarchically or-
82 A MANUAL OF PHONOLOGY

ganized system, where to cut between obvious ICs, so, when the cut has been
made, it is by no means always obvious whether the construction in which the
immediate constituents stand is coordinate or subordinate. In the preceding
section and the present one we have chosen Winnebago as the most likely candi-
date for vowel clusters within peaks, standing in a subordinate construction, and
Taos as a likely candidate for vowel clusters within peaks, standing in a co-
ordinate construction. The difference lies largely in the fact that in Winnebago
there is reputedly a clear difference in prominence between one and the other
part of the complex peak, whereas in Taos the report states, to the contrary,
that normally the two constituents are about equally prominent. These may be
very tenuous grounds for the differentiation; it is possible that structurally no
such distinction should be made.
The problem presents itself in way in Mazateco. Setting aside
a more complex
the tonal constituent of a peak, and then from the remainder the contrast
lifting
between oral and nasal, which applies to the peak as a whole, we are left with
vocalic constituents as follows: simple /i e a o/; more complex /ai ao ie ia io oi
oe oa/; most complex /iai iao oia oai oao ioa ioi/. The variation in relative
prominence among the constituents of these complex nuclei is extremely com-
plicated, seeming to depend partly on which position is occupied by a particular
vowel and partly on which vowel occupies the position. If the choice between
subordinate and coordinate must be made, it seems best to regard all the complex
nuclei
even those of three vowels as coordinate.
244. Vowel Systems. In a language which has only simple peaks, there is a
set of phonemes, either just vowels or vowels and semivowels, which occur as
peaks. In a language which has both simple and complex peaks, we find after
analysis some set of phonemes which occur as simple peaks, or as the nuclei of
complex peaks, or as members of complex peaks where the construction is co-
ordinate: these phonemes, again, may be just vowels, or may include semivowels,
demivowels, or omnipotents. In any case, the set of phonemes which occur as
peaks, or as nuclei in complex peaks, or as coordinate members of complex peaks,
form one or more systems in terms of the ultimate phonological constituents
which compose them.
To determine whether the phonemes in question constitute a single system or
more than one system it is necessary to survey their privileges of occurrence, as
or in peaks, under varying accentual and other conditions. In Yuma there are
five vowels which occur as peaks or peak nuclei in stressed syllables all of these
;

occur also in unstressed syllables, but in addition there is a sixth vowel, found
only in unstressed syllables, in contrast with all of the five and with zero. Here we
can segregate the set of five which occur in stressed and unstressed syllables and
treat them as a single system; the sixth vowel constitutes a special system in its
own right. Quite similarly, the French schwa ("e muet," /&/) seems not to pat-
tern as one of the ordinary vowels, but as an element apart from all the rest,
even though in this case it is not stress differences which are involved, but posi-
tions of occurrence relative to surrounding segmental phonemes. Where nasaliza-
tion applies in some way to whole peaks or whole syllables (as in Apachean or
Winnebago), it is necessary in the first instance to examine the system which
 A MANUAL OF PHONOLOGY 83

occurs without nasalization and the system which occurs with it; one may find

parallelism, in which case one can speak of a single system, or lack of parallelism,
in which case it is often necessary to speak of two systems. The latter is the case
in Taos, where oral nuclei involve six vowels in one pattern, whereas nasal nuclei
involve only five vowels in quite a different pattern.
Nasal continuants (/m n/ etc) which are semivowels in some languages, seem
,

regularly to be marked as constituting separate subsystems in their occurrences

as nuclei. Thus in Chiricahua Apache nasal continuants occur as syllable peaks
only in syllables with no distinctive onsets or codas, and in complex peaks only
in geminate sequences, not in mixed sequences with a vowel. In most of the cases
where laterals or trills are semivowels or demivowels, they are similarly ear-
marked as special; this is not true, however, of the retroflex high omnipotent
/r/ of Mandarin, which forms part of the regular peak system.
By a "vowel system," then, for the purposes of the present survey, we shall
mean the system formed by the phonemes which occur as peaks, or as peak-
nuclei, in any given language, or any one such system when there are two or
more. We shall set aside subsystems represented by nasal continuants, laterals,
or trills, when these occur as semivowels or demivowels, since there seems to be
very little relation between these and the systems formed by the other relevant
phonemes. But the term "vowel" will be used here in the sense just described,
larger than the limited sense of 242.
Vowel systems can be classed in terms of the dimensions of contrast involved
in them and the number of contrasting points in each dimension. The one
dimension that is always present is that of tongue-height, and attested vowel
systems involve two, three, or four contrasting heights. We shall therefore use
this as the primary criterion for our classification. A second dimension in many
cases consists of the two-way contrast between front-unrounded and back-
rounded. Sometimes the contrast is only front-back, with rounding secondary;
occasionally the second contrast is fundamentally rounded-unrounded, with
frontness and backness secondary. When three points are involved in the second
dimension, there are two cases: front-unrounded versus front-rounded versus
back-rounded, involving both the factors of lip-rounding and tongue frontness
and backness but only in three combinations; and front-unrounded, back-un-
rounded, and back- rounded, involving the same factors but in a different set
of three combinations. Sometimes central occurs rather than front in the first
type, or instead of back in the second.
Such a second dimension of contrast may apply at all the tongue-heights, or
only at some of them. With rare exceptions, the number of contrasts along the
second dimension never greater for a lower tongue-height than for a higher one.
is

If more than three contrasts are involved at any tongue-height, then one
has not a single added dimension, but two: tongue frontness and backness and
liprounding or spreading working independently. To this generalization, how-
ever, Mandarin is probably an exception.
By no means all the systems which are theoretically possible with the raw
materials just described are actually attested. In our survey below we have
assigned no system to the type with three high and three low vowels. Some
 .

84 A MANUAL OF PHONOLOGY

systems, with /i u u e a o/ or /i i u e a o/, might be so classed, but the /a/ in

any such case is typically lower than the /e o/, and we have been led to choose
an interpretation with three heights. Yet such a minor difference in height is not
always decisive: we class Fox as a two-by-two system despite the fact that the
vowel classed as low back, /a/, is typically lower than that classed as low front,
/e/. It is perhaps arbitrary to have insisted on a three-height interpretation for
the systems first mentioned when allowing a two-height interpretation for lan-
guages like Fox. In the survey that follows there are doubtless various cases of
arbitrary decision, and perhaps some of indisputably wrong classification.
2441. Systems with Two Heights. A 2+1 system (this type of coding will
explain itself as we proceed) of the shape
i u
a
is reasonably common; it is reported for Arunta, Cree, Eskimo (most dialects,
perhaps not all), some Arabic Coeur
dialects (including Iraqi), Salishan (except
d'Alene;some of the languages may have a neutral /a/ as a separate system),
Muskogean (except Creek-Seminole), Ojibwa, Kechua before the introduction of
Spanish loans, Totonac, Lak, and Wishram. In Wishram, /i u/ are apparently
semivowels, so that only /a/ is a "pure" vowel in the sense of the definitions

of 242. The subsystem unaccented vowels in Russian and Tunica, or of

of
nasalizable vowels in Winnebago, Iowa-Oto, Osage, and Temoayan Otomi, is
also of this type.
A 2X2 type is almost equally widespread:
i o
e a
In Rutul the high back vowel sometimes rounded, sometimes not, depending
is

on environment. In the other attested cases it is always rounded: in Fox and

Shawnee the low back vowel is usually unrounded, but rounded in certain en-
vironments. This system is reported, in addition to the cases already mentioned,
for Apachean, Campa, Chatino nasal vowels, many dialects of Nahuatl (not
Tetelcingo), Creek-Seminole (/e/ rare and limited to occurrence before /y/),
and Wichita; it was probably the system of Proto-Germanic, and almost certainly
that of Proto-Central Algonquian.
A 3+1 system is reported for Amahuaca:
i i u
a
though the /i/ may be lower than /i u/, placing Amahuaca rather with Ilocano
and others (2442).
3+2 systems occur in two varieties:
i u u
e a
in Tabassaran and Kyuri; but with /i/ instead of /u/ in Huichol.
Turkish has a 2X2X2 system:
i ii i u
e 6 a o
 A MANUAL OF PHONOLOGY 85

with tongue-height, frontness-backness, and rounding-unrounding all working

independently. We might describe the Apachean system as 2X2X2, the third
dimension being tone:

e" a e a
but the pitch components are not confined to occurrence within this system of
eight units, since the languages also have syllabic nasal continuants with one
or the other of the tones. Therefore we extract the tones and regard them as
separate elements, leaving merely a 2X2 system of vowels.

2442. Systems with Three Heights. A 1+1+1 (or 1X3) scheme is reported for
Adyge and possibly for Abkhaz and Ubykh
i

9
a .

This is the only known type in which only tongue-height is involved. All of the
vowels in the Adyge version of this system have variants which are relatively
more front or back, and relatively more or less rounded, but such variations are
dependent on the environment, not a part of the vowel itself.
Ilocano, Dibabaon, and perhaps some other Filipino languages have a 2+1 + 1
pattern:
i u
9
a .

The mid vowel /a/ has fronted variants, and also higher variants (central or
back without rounding). Spanish loans have rendered the situation more compli-
cated for some speakers.
Potawatomi has a 2+1+2 pattern:
i o
9
e a
The mid central vowel /a/ varies more widely from one environment to another
than the other four, but there is no reason to set it off as constituting a special
subsystem by itself. Nevertheless, the system resembles a 2X2 pattern with an
added neutral vowel in the middle.
Probably the commonest pattern of all certainly the most familiar is the
2+2 + 1 scheme:
i u
e o
a .

In Spanish, and many other cases, the second dimension for the mid and high
vowels is front-unrounded versus back-rounded. For Russian, however, the
second dimension is rounded versus unrounded: /i/ has central and front al-

lophones, and /u/ back and central (or even front) allophones, and there is

similar,though less marked, variation for /e o/. In Keres, likewise, the second
dimension is rounded-unrounded: /i/ and /e/ are central or back, and front
allophones are rare if they occur at all.
 . .

86 A MANUAL OF PHONOLOGY

A 2+2+1 scheme, of one or another of these varieties, is reported for Awar,

Badaga (with an added contrast between retrofiex, partly retroflex, and non-
retroflex), Chatino oral vowels, Chitimacha, Chontal, Coeur d'Alene, Czech,
Erzya, Fijian, Greek (modern and early Classical, but not late Classical),
Georgian, Hausa, Huasteco, Iowa-Oto oral vowels, Japanese, Keres, Kota, Latin,
Miwok, Nootka, Osage oral vowels, Polish, Polynesian, Kechua (with Spanish
loans), Russian stressed vowels, Spanish, Serbo-Croatian, Saho, Tagalog (the
/i/:/e/ and /u/:/o/ contrasts relatively weak), Tojolabal, Tonka wa stressed
vowels, most dialects of Yokuts (including Yawelmani), and Zuni. In Malay
and Yuma such a system accompanied by a neutral vowel /q/, found in Yuma
is

only in unstressed syllables; this was probably the pattern in Chaucerian Middle
English. In Taos, Terena, Bariba, Supide, and Senadi the vowels of nasalized
peaks have this pattern, those of nonnasalized peaks some larger pattern.
Cebuano Bisayan, and perhaps some dialects of Tagalog, have a system which
can only be realistically set up as follows:
[i] u
[e] o
a .

[i] and [e] are both heard as allophones of a single vowel, while their parallel
back vowels are phonemically distinct. Any rearrangement to show just the four
vowel phonemes seems highly unrealistic from the phonetic point of view.
The 2X3 pattern is rarer:
i u
e o
e a
Chipewyan, Dargwa, Menomini, Persian, Ukrainian, and Yuchi have this system.
Taos vowels in non-nasal peaks have a pattern differing only in phonetic details,
mainly the fact that the mid back vowel is unrounded.
A 2+3+1 pattern appears in Oneida:
i u
e 8 o
a .

/q/ and /u/ are nasal, the other four oral, but there seems to be no better way
of setting them up. No distributional characteristics set /a/ and /u/ up as over
against the other four.
A 2+3+2 pattern appears in Lifu:
i u
e a o
ae a
The 3+2+1 pattern is attested in two versions, depending on the combination
of front-back and rounded-unrounded found in the "central" high vowel:
i ii u
e o
a
 A MANUAL OF PHONOLOGY 87

in Taki-Taki, late Classical and early Medieval Greek; but

i i u
e o
a
in Cashibo, Bulgarian, Mazahua Otomi nasal vowels, Mixteco, Sierra Popoluca,
and Zoque.
3+2+2 pattern occurs in some Bulgarian dialects and perhaps in Votyak:
i i u
e o
e a
The other version of this (with /u/ instead of /i/) is not attested.
The 3+3+1 pattern with /ii/ is fairly common:
i ii u
e 6 o
a
in German, Dutch (some dialects, and overall pattern for standard Dutch),
Hungarian (some dialects), French (some dialects, particularly in the south),
Zyryan. In Terena the vowels of oral syllables have the other version, recurrent
in Maidu and Romanian:
i i u
e 9 o
a .

Votyak may belong here rather than in the 3+2+2

type. In Terena the seven-
way contrast appears only before /h/ ; environments (as for the vowels
in all other
of nasal syllables) /i/ and /a/ are lacking, and the system reduces to the 2+2+1
type.
Both versions of the 3+3+2 pattern are attested: Finnish and some Hun-
garian dialects have
i ii u
e 6 o

while Cuitlateco and perhaps some dialects of Korean have

i i u
e 9 o
a o
If Korean dialects have this pattern, the low vowels are front /se/ and back /a/,
rather than unrounded /a/ and rounded /o/ as in Cuitlateco.
The symmetrical 3X3 pattern is apparently also attested in two versions. In
Trukese, Thai, Temoayan and Mazahua Otomi, and English:
i i u
e a o

but in Esthonian:
1 u u
e 6 o
e
 : . . :

88 A MANUAL OF PHONOLOGY

In Mandarin there appears a 4+1 +

1 system in which front-back and round-

ing-unrounding do not function separately:

i u r u
a
a .

/r/ a high strongly retroflexed vocoid, which, like the other three high vowels,
is

occurs both as a peak and in margins. The mid and low vowels vary from front
(unrounded) to back (with or without rounding) depending on environment.
Koibal, Karagin, and possibly some dialects of Korean have a system describ-
able as 2X (2+2+1):
i ii i u
e 6 3
se a
Standard Korean is skew in lacking the /ii/, though the remaining vowels fall

obviously into this pattern. (Cf. the skewness of Cebuano Bisayan, above.) In
some dialects the /ii/ may occur as a separate phoneme; in others it is known
that both /ii/ and /6/ are lacking, which places those dialects in the 3+3+2
pattern.
There is some evidence suggesting that certain dialects of Old English had
the following scheme
i ii i u
e 6 9
ae a o
As displayed, this would be describable as (2+2+l) + (2X3). Naturally, details
are obscure: in particular, it is not clear whether the vowel listed as /o/ was
rounded or merely back.
2443. Systems with Four Heights. A 2+1 + 1 1 system is reconstructed by +
Fairbanks for the Old Slavic of the Zographensis manuscript (oral vowels only)
i u

a .

The higher-mid, lower-mid, and low vowels all have variants differing as to
front-back or rounding or both, dependent on environment. Thus /a/ after a
palatal consonant was (roughly) [e], but after a plain consonant was [o]. This
system not attested for any language known through direct observation, but
is

it is sufficiently reminiscent of the 1X3 system of Adyge, the 2+1 1 system +

of Ilocano, and the 4+1 + 1 system of Mandarin to qualify easily as a "realistic"
reconstruction.
The 2+2+2+1 system is reasonably frequent:
i u
e o
e o
a
in Italian, late Latin (?), some dialects of Portuguese (oral vowels), Tunica
stressed vowels, Loma, Kiowa, Tetelcingo Nahuatl, and the oral syllables of
 :

A MANUAL OF PHONOLOGY 89

Bariba, Senadi, and Supide. In Loma the second dimension of contrast is un-
rounded and rounded: /i e e/ have both front and central (or possibly back)
allophones, whereas /u o o/ are always rounded and back. In the other cases
/i e e/ are front unrounded.
Different Polish dialects are said to represent both versions of the 2X4
scheme
1 u i u
e e
and
e e
se a
a D
in the first, the fundamental second-dimension contrast is between front and
back, while in the second it is between rounded and unrounded.
Some Portuguese dialects have a 2+2+3+1 pattern in oral syllables:
1 U
e
e A
a .

With nasalization, /e a o/ do not occur, and the system simplifies to 2+2+1.

In other dialects the contrast between /a/ and /a/ is lacking and one has a
2+2+2+1 pattern in nonnasal syllables.
The Lyster dialect of Norwegian is reported to have a 3+3+3+1 system:
1 u u
e 6
e 03
a .

Some conservative French dialects, and perhaps some dialects of Danish and
Norwegian, have a 3+3+3+2 pattern resembling that just displayed except
that the bottom row includes /ae/ and /a/. Still other Danish dialects seem to

have a symmetric 3X4 system, the bottom row /a? a a/. The crosswise contrasts
for the lowest vowels in this case is not the same as those for the higher vowels:
/&/ is front, /a/ central, and /a/ back but perhaps not rounded.
2444. Skew Systems. Uncertainties Votyak, and
of classification, such as for
holes in otherwise neat patterns, such as the absence of /u/ in varieties of stand-
ard Korean which have an /6/, are normal enough. But there are also some re-
ports which apparently describe completely skew vowel systems which will
fit into no classification of the sorts worked out above. We must suspend judg-

ment in these cases: perhaps the investigator has erred, but perhaps he is quite
accurate and really skew systems indeed exist.
One example is Hopi. Whorf sets up
six vowel phonemes. Two, /i/ and /a/,
are fairly clear. There an /e/ which varies between mid and low front un-
is

rounded. And there are three vowels which vary between mid and high: central
unrounded /a/, front rounded /6/, and back rounded /o/.
Another example is Chahar Mongol, for which the report gives: lower high
/i/, /i/, /u/; higher mid /6/; mid /o/; higher low /a/, /o/, /o/; low /a/; inter-
secting this, front unrounded /i/; unrounded central to back /i/, /a/, /a/;
rounded front /o/; rounded central /6/; rounded back /u/, /o/, and /o/.
90 A MANUAL OF PHONOLOGY

2445. Double and Multiple Systems. In 2441-3 we separated subsystems,

such as that in stressed and that in unstressed syllables, and treated each inde-
pendently. Here we summarize the types of multiple systems which are attested.
The contrast of nasal and oral exactly doubles the number of vowels in Apach-
ean (2X2, all either oral or nasal) and in Yuchi (2X3, all either oral or nasal).
In no instance are there more nasal than oral contrasts. In Winnebago, Iowa-Oto,
and Osage, the oral vowels form a 2+ 2 + 1 system and the nasalized (or nasaliza-
ble) vowels a 2+1
system. In Bariba, Supide, and Senadi the oral vowels form a
2+2-1-2+1 system and the nasal (or nasalizable) vowels a 2+2+1 system.
The same holds for some dialects of Portuguese; others have a 2+2+3 + 1
system for oral vowels. In Terena, oral vowels before /h/ form a 3+3 + 1 system,
but elsewhere, and with nasalization, the back unrounded vowels are lacking
and the system is 2+2+1. In French the ratio of oral to nasal vowels is quite
high (the number of the former depending on dialect); some dialects have a 3 + 1
nasal system //, /?/, /o/> /aV, others a 2+1 system without the /qe/. French
is quite unique in that the nasal system involves only relatively low vowels; in

most languages where nasal vowels occur, at least the highest and lowest of the
oral vowels are matched fairly closely by nasal vowels.
Taos, Chatino, and Mazahua Otorni stand apart in that the nasal vowels
cannot in any sensible way be paired off with oral vowels, but must be taken
as constituting a totally different pattern. Thus Taos has a 2X3 oral system,
but a 2+2+1 nasal system; Chatino has a 2+2 + 1 oral system but a 2X2 nasal
system; Mazahua Otomi, finally, has a 3X3 oral system but only a 3+2 + 1
nasal system.
Badaga is the only language for which contrasts of retroflexion are attested
as playing a major role; the three-way contrast exactly intersects the 2+2+1
basic vowel system.
It is rather common for fewer vowels to occur in unstressed syllables than in
stressed: Russian has 2+2+1 in stressed syllables, but only 2+1 in unstressed;
Tunica 2+2+2+1 in stressed, but only 2+1 in unstressed. But
happens
it also
that a language may have more contrasts in unstressed position. has a Yuma
2+2+1 set of vowels stressed or unstressed, and a /o/ only unstressed; Middle
English was perhaps the same.
French and Malay both have a /&/ in addition to their principal vowel systems.
This is not separated from the principal vowel system by occurrence relative to

stress, because neither language has any accentual system, but in both cases

certain special phonetic and distributional features mark the /a/ off.
244(5. Hierarchical Structure in Vowels. In any simple vowel system, or any
simple subsystem of a double or multiple system, there seems no reason to regard
the construction in which the ICs stand as other than coordinate. That is, in a
2X3 system, for example, we regard the frontness and the height of /i/ as of
coequal importance, rather than faking either as a satellite to the other,
On the other hand, in a double or multiple system there is often reason for
regarding the members of all subsystems but one as involving an attributive
construction. Thus in Yuchi, where there is a 2X3 system of oral vowels and an
identical system of nasal vowels, we would interpret /[/ as nucleus /i/ and
 A MANUAL OF PHONOLOGY 91

simultaneous satellite /./; in Badaga we would regard any half-retroflexed or

fully retroflexed vowel as consisting of a non-retroflex vowel as nucleus and the
degree of retroflexion as satellite.
This also applies in any case where the phonetically more complex subsystem
forms a neat selection of some of the possibilities of the phonetically less complex.
Thus the fact that Winnebago /e o/ are not subject to nasalization does not
prevent us from regarding f\ q, 11/ as consisting of /i a u/ under the influence of
a nasalization factor (2432). However, where the two subsystems are incom-
mensurate, as in Taos, this particular type of componential analysis is not pos-
sible.
Following this clue, we might decide to treat Russian stresses in the same
fashion in which we decided, in 2311, to treat English stresses: set up a phoneme
of absence of stress, for example on the second vowel of /gorat/ 'city', rather
than a phoneme of stress on the first vowel (/gorat/). Doing this does not render
unstressed vowels phonetically more complex, but it renders them by fiat

more complex, and paves the way for saying that only /i a u/ can
structurally
be modified by the simultaneous satellite /*/, just as in Winnebago only /i a u/
can be modified by nasalization. A consistent following of this procedure would
then require us to describe Yuma with a phoneme of stress, since of the six
vowels which occur unstressed, only five (all but /a/) occur stressed.
Simultaneous satellites in vowels then come to occupy a status structurally
comparable (in part) to sequential satellites in complex peaks (243). This is

particularly desirable because in some instances what is phonetically a simultane-

ous satellite turns out to be in complementation with something else, just as the
directional glide in English complex peaks /ay ey oy/ is in complementation
with the onset glide vocoid in yes. Some French dialects, particularly in the south,
have no contrast between [ej and [en]; the nearest approach to the latter is [enaj,
and it is advisable to treat the former as structurally /en/, the latter as /ena/
(similarly, of course, for the other nasalized vowels of the dialect). This treat-
ment does not destroy the distinction between (phonetic) simultaneity and suc-
cessivity, but it capitalizes on a complementation and portrays the actual
network of contrasts more efficiently; /en/ remains a complex peak, not a simple
peak followed by a coda.
25. Syllable Margin Systems. By the two classifications of phonemes in 242,
the phonemes which occur as margins (as or in onsets, codas, or interludes) in a
language which has only simple syllable peaks are either consonants or semi-
vowels; and, in a language which has both simple and complex peaks, either
consonants, semiconsonants, demivowels, or omnipotents. We wish now to
consider the systems into which these phonemes, in any one language, enter, in
their marginal roles, regardless of whether they also occupy other roles in other
environments. For this purpose we shall generalize the term "consonant" to
subsume any phoneme which occurs in or as a margin, when so occurring with
one exception: glide vocoids which are allophones of phonemes which occur as
syllable peaks or as nuclei in complex syllable peaks (Chinese /i u ii r/) are
excluded. These have already been covered in our survey of vowel systems, by
the special definition of "vowel" used for that purpose. It will be noted that
92 A MANUAL OF PHONOLOGY

co vowels were not covered in that survey and are also not covered here: they
occupy a special sort of status, where they occur at all, dealt with adequately
in 2431.
The consonants our generalized sense) of any one language fall into various
(in
distributional classes depending on their privileges of occurrence relative to each
other in simple and complex onsets, codas, and interludes. The determination of
these distributional classes goes hand in hand with the examination of the
variety of structure found in margins. The same consonants fall into certain
constitutional classes depending on the ultimate phonologic constituents of which
they are composed. In analytic work these two matters have to be investigated
together, since otherwise one cannot tell, for example, whether an affricate [c]
is a unit phoneme or a cluster /ts/ or /ty/, or whether a glottalized consonant,

say [k'], which occurs as an onset, is to be interpreted as simple (one phoneme)

and /?/) However, the methodology will concern
or as complex (a cluster of /k/
us later; need only be pointed out here that questions of this sort are not
it

always equally important though they must always be asked, for one cannot
even discover that they are unimportant without asking them.
251. Distributional Classification. No general typology for distributional classes
of consonants can be offered. In some languages this matter is extremely simple,
in others extremely complex, and the complexities vary from one case to another
in the most whimsical way. We shall, however, give several examples (beyond
the hints in 221).
251 1 The Simplest Situations. In Fijian, all onsets consist of a single consonant,
.

and consonants occur only as whole onsets. Therefore the only distributional
classification of consonants is into a class which includes them all. This is the
case, or nearly so in a good many other languages, including some which have
codas or interludes as well as onsets. In Yokuts, for example, any consonant
may occur as an onset or as a coda; the only limitation is that the glottalized
sonorants /m' n' 1' w' y'/ occur only directly after a vowel. This sets these aside
as a special distributional subclass; otherwise all are in the same class.
In Fox the occurrent onsets and interludes can be charted as follows:
p t 6 k s h m n
pw tw kw sw sw hw mw nw w
py ky hy my ny y
hP ht he hk
hpw htw hkw
hpy hky
sk
skw
sky
All of these occur as interludes; those in the first three rows, except /h hw hy/,
occur also as onsets. It seems reasonable to call the single-phoneme units simple
margins; to call the two-phoneme units in which /w/ or /y/ is the second complex
margins, with the /w/ what precedes; and to call the units
or /y/ a satellite to
in which /h/ or /s/ precedes a stop (whether a /w/ or /y/ follows or not) hyper-
complex margins, with the /h/ or // satellite to the simple or complex unit that
follows it. The labels ("simple, complex, hypercomplex") are of no importance^
 A MANUAL OF PHONOLOGY 93

but the assertion that the clusters are of the subordinating type rather than co-
ordinate, and that the direction of satellite-nucleus relationship is of one kind
rather than another, is roughly in keeping with our general habits of usage of
these terms in the analysis of hierarchic structures. We break a three-termed
cluster such as /hpy/ first into /h/ and /py/ because of the greater freedom of
occurrence of the latter macrosegment as well as medially.
initially in
A first Fox consonants would then be: (1)
distributional classification of
/p t 6 k/, which occur after a satellite; (2) /s m n/, which occur before but not
after a satellite; (3) /s h/, which occur before but not after a satellite, but also
as preposed satellites; (4) /w y/, which occur as postposed satellites. A finer-
grained distributional classification would ultimately set up a separate class
almost for each consonant, since no two (save only /m n/) have exactly identical
privileges of occurrence.
Mazateco has only onsets, but these attain a fair degree of complexity. Simple
onsets include all the single consonants, a few of which occur only in this role
(e.g., /p b/, only in Spanish loans). The consonants which occur both as simple
onsets and in complex ones are /tkc6vymnn y ssh?/. Complex onsets
are /nt nk nc nc n6 sk st sk sn/. Since there is greater variety in the second
member of these than in the first, we may perhaps say that the first is satellite

to the second for /sn/ this

; is dubious, and perhaps it should be taken as coordi-
nate. Hypercomplex onsets consist of a simple or complex unit preceded or fol-
lowed by /h/ or /?/: /ht hk he he" he hv hy hm hn hn y hnt hnk hnc hng/, /th
kh ch ch h vh mh nh sh sh nth nkh nch n6h nh/, /?v ?y ?m ?n ?n y ?nt ?nk
?nc ?n6 ?nc/, /t? k? c? 6? c? v? y? m? n? n y ? s? s? nt? nk? nc? n? nc? sk?
st? k?/. Here we interpret /h/ or /?/ as satellite. There are also two onsets,
/he? h<5?/, which we interpret as nuclear /c 6/ with a satellite on either side.
Rougher or finer distributional classifications can be deduced from these data.
2512. More Complicated Cases. All of the above examples are fairly simple.
In a language which has, in addition to onsets, fairly complex codas or interludes
or both, the situation can become extremely complex, even if the total number
of individual consonant phonemes is not very large. This is true, for example, in
English. Often, despite the complexity, certain generalizations can be made as
to what occurs and what does not. In many languages which have a contrast
such as that between voiced and voiceless for some consonants, it will happen
that only voiced consonants, or only voiceless consonants of those that pair
off will occur in almost true in English:
a single onset, coda, or interlude. This is

it holds for onsets, but there are codas /dd/ (width) and /dst/ (midst) in which

it does not hold, and many such interludes. Some speakers pronounce these

words with /td/ and /tst/, and for them the generalization holds for codas. In
standard Polish and in Portuguese the statement holds throughout. In Russian
it holds save for the pairs /f v/ and /f y v y/ the voiced : members of these pairs
occur both after voiced phonemes and after voiceless phonemes (of those that
pair).
Another commonly found limitation applies in clusters in which the first
member is a nasal continuant and the next is a stop: only such clusters occur

as have continuant and stop in the same position of articulation. This applies in
Spanish and in many other languages.
94 A MANUAL OF PHONOLOGY

In any such case often convenient to set up one's phonologic units differ-
it is

ently. Since voicingand voicelessness apply not individually to the paired conso-
nants, but to any sequence of one or more flanked by other phonemes (unpaired
consonants, vowels, junctures), one may halve the number of consonant pho-
nemes and recognize voicing, or voicelessness, as an additional phonemic unit,
having as its domain that which we have just described. If we do this in Polish,
for example, we set up "voicing" as a sort of accentual phoneme; writing this
with a diacritic / v / after the string of consonants affected, we write /. . . asa
.../,/... as v a ./ (instead of /.
. . aza .../);/... atfa .../,/... atf w a
. . . . ./

(instead of /. adva .../); and so on. There is the additional advantage that
. .

/*/ is never written before microjuncture, and this automatically covers the fact
that voiced consonants, of those that pair off as voiced and voiceless, never occur
before microjuncture. There are, of course, other ways to state this structural
fact, but this way seems to subsume as many different relevant structural con-
siderations as possible at once.

2513. Correlation between Constitution and Distribution. We must note also that
the distributional classification of consonants often correlates, at least in part,
with the constitutional classification. In Fox even the finest-grained distributional
classification leaves /m/ and /n/ together: constitutionally, this is the class of
nasal continuants. A looser classification leaves /w/ and /y/ together, but iso-
lated from all /w/ and /y/ form the constitutional class
the other consonants:
of glide vocoids in Fox. k/ form a distributional class in that /h/ can
/p t c"

precede any of them as satellite: /p t 6 k/ is the constitutional class of stops and

affricates. In Mazateco, /h/ and /?/ are members of a single distributional class,
and they go together constitutionally as "glottalic" consonants.
It is only when there is some correlation, of this kind, between distributional
and constitutional classes that a change of phonemicization, of the sort proposed
above for paired voiced and voiceless consonants in Polish, can be undertaken.
And often there is some lack of correlation. English has a constitutional
at least
class of nasal continuants, /m n n/: the former two go together fairly well
distributionally, but /n/ stands quite apart, since /m n/ occur as onsets and in
complex onsets, while /n/ does not.
Let us see what the correlation is in Mandarin. Distributional class (1),

/p p' m 1 1'/, includes those consonants which occur alone or with following /i/
or /u/ in onsets. Class (2), /n 1/, go alone or with following /i u ii/. Class (3),
/c go alone or with following /i u ii r ru/. Class (4), /f k k' x/, go alone or
c* s/,

with following /u/. Class (5), /n/, goes only alone. Now (1) includes labial stops
and nasal, and also apical stops, but not the apical nasal. The two members of
(2) are both apical continuants, and there are no others. (3) contains all the
sibilants : two affricates and one and
spirant. Class four includes the dorsal stops
spirant but also the labial spirant. Thus
some correlation, better in
there is

some spots than in others, but far from complete and exact.
The correlation is worst
or best when no distributional classification is
possible at all; that is, when all the consonants occur in all possible positions,
as in Fijian or Yawelmani. It is at its worst in the sense that there are no dis-
tributional considerations to aid the analyst in deciding just how to distinguish
between simple and complex margins (e.g., in the case of a [c] which may be a
 A MANUAL OF PHONOLOGY 95

unit /c/ or a cluster /ts/). It is at its best in the sense that there are no dis-
tributional considerations to interfere with a decomposition of the consonants
into ultimate phonologic constituents and a classification of the consonants on
the basis of those constituents: the constitutional classification is the only one
that counts, and some possible questions about "unit or cluster" lose their
meaning. We will illustrate this with Fijian.
Fijian onsets include /t k m b n d ng <5 m n n s r
n
r 1 w y/. Four of these,
m n n n
sound They
/ b d g r/, like clusters to speakers of English. could be so inter-
preted: thefirst three let us say as phonemically /mp nt nk/, setting up /p t k/

as phonemes which are everywhere stops but are voiced or voiceless depending
on environment, or else as /m/3 n<5 07/, setting up //? (5 7/ as phonemes which are
everywhere voiced but are stops or spirants depending on environment. The
first of these requires adding a /p/, which does not occur otherwise in Fijian

(save in the speech of some, in recent loans). The second requires adding a /y/,
which does not occur at all (save as a free variant of /k/). Either of these interpre-
tations, or the one originally indicated which takes the prenasalized consonants
as units, does no violence to the system: the choice has to be based purely on
considerations of esthetics or "elegance," which reside in the analyst, not in the
data, or else has to be suspended. The only reason that we do not like to suspend
the choice is that our current terminology and techniques render it difficult to
describe a system which is determinate just to the point to which Fijian is, but
in which further questions of the "phoneme or cluster" type are irrelevant.
But this apparent indeterminacy is encountered only because all the Fijian
m n
d n g/ occurred only
consonants belong to a single distributional class. If / b
medially, not initially and medially, then they would form a separate distribu-
tional class of consonantal entities, and the cluster interpretation would have
more to support it even if /m n rj/, like all other consonantal units, did not
occur otherwise save before a vowel.
252. Constitutional Classification. Consonantal systems are highly varied and,
in some not seem feasible to handle them as wholes
cases, quite complex. It does
for constitutional classification; seems better to develop some manner of
it

breaking them up into subsystems. Of course, the particular way in which we

shall do this is in no sense the sole "correct" way, and can no doubt be im-
proved on.
In order to illustrate our approach, let us tabulate the consonants of Ossetic:

p t c 5 k
<
6'
p t c k'
b d 3 3 g

f s s X
V z z 7

m n
w y r 1
96 A MANUAL OF PHONOLOGY

In the upper left-hand box we have

included all consonants which are stops,
some environments. This set constitutes the
affricates, or spirants in at least
obstruent system of the language. In the lower left-hand box we have included
nasals, semivowels, laterals, and trills
the sonorant system of the language.
The upper right-hand box includes all consonants produced in pharynx or glottis
in this case, just one consonant. We shall use the already physiologically
meaningless term laryngeal as a cover term for these last.

A generalization of this yields our organization for most consonantal systems.

We include among the obstruents not only all stops and affricates and most
spirants, but also, in some cases, a [y]-like, [w]-like, or [l]-like consonant if it

fits neatly into the scheme, and if distributional facts do not militate against
such a treatment. On the other hand, a [v]-like spirant sometimes seems to pair
off distributionally with a and so may belong to the sonorants.
[y]-like glide,

In a few cases a simple two-way partition of all consonants (apart from laryn-

geal, if any) does not fit the facts, and a more complicated set-up is required.
Sometimes one or more of the laryngeals count as manner consonants, in the
sense that they match one or another of the styles of delivery found for obstru-
ents. For example, in Nootka /?/ matches the obstruent set /p' t' c' c' X' k' k' w
q' q' w /, while /h/ matches the obstruents /s 1 x x w x x w /- Occasionally a man-

ner consonant is not a laryngeal: for example, the /y/ of Russian, which matches
the whole set of palatal consonants as over against the plain consonants.
In a few cases all the consonants, apart from manner consonants (if any),
must be otherwise dichotomized in the first instance, the classification into
obstruent and sonorant coming second. To illustrate this we tabulate the Lifu
system

p t t c k
f s s X
h
M N J? s
W L

b d d 5 g
V 3 z z 7
m n n D
w 1

Here the contrast between and voiced is operative throughout the

voiceless
system except for /h/; the manner consonant and pairs off with the
latter is a
whole set of voiceless obstruents and sonorants. Distributional classification
supports the constitutional grouping indicated above: only voiced consonants
(and all of them except /w rj g 7 v/) occur finally. (An alternative, worthy of
serious consideration, is to regard the voiceless set as clusters of the voiced set
and /h/; one could then say that /h/ occurs finally neither after a vowel nor
 A MANUAL OF PHONOLOGY 97

after a consonant.) Syllable structure is almost as in Fijian: vowel, or consonant

plus vowel, except for the sprinkling of final consonants.
We shall begin our survey of the constitutional classification of consonant
systems with obstruents; then we shall deal with sonorants; then with manner
consonants and laryngeals.
2521. Obstruent Systems. The usual classification of obstruents is according to
manner and position. Thus, in the table of Ossetic obstruents in 2520, all those
in a single row (say /p t c c k/) share a manner (voiceless unaspirated, some-
times glottalized), while those in a single column (say /p p b f v/) share
all

roughly the same position (labial). Similarly, we could tabulate the obstruents
of either Lifu half -system as follows:
p t \ 6 k
f s s x .

A simple way would be to tabulate each system

to classify obstruent systems
in this way, with rows and columns marked. There would then be several relevant
number of columns; (2) the position or range
typologic considerations: (1) the
column; (3) the number of rows; (4) the manner or range
of positions for each
of manners for each row; (5) holes
as, in Ossetic, the lack of any spirants to
match the apical stops.
We more complicated procedure, though it is
are going to follow a slightly
not certain that the information gained by it is either clearer or better. As one

point of reference we shall be concerned with the number and variety of positions
of articulation for stops; positions for spirants will be judged against this. But
before dealing with manners we shall first perform another operation.
This preliminary operation consists in deleting from the tabular array of
obstruents those which will leave a rectangular array with all pigeonholes filled,

and listing the deleted obstruents separately. In each case, we shall do this in
whatever way yields the largest rectangular array. In Ossetic, for example, we
could obtain a rectangular array of three rows of frv3 entries each by deleting
the eight spirants; but we can obtain an array of five rows of four columns each
by deleting the three apical stops. The latter is a larger array than the former
(20 pigeonholes in the 5X4, only 15 in the 3X5), so we choose the latter rather
than the former. If the consonants deleted in this initial operation allow of a
second comparable operation, we perform the operation again. In Ossetic the
leftovers do not allow any reapplication of the procedure, but in some cases
they do.
Any rectangular array obtained in this way is a symmetric set. A number of
obstruent systems include no symmetric set at all; at least one consists of a
single symmetric set. The number of non-intersecting symmetric sets included
in an obstruent system, and the size of the largest, relative to the whole number
of obstruent phonemes, are a rough measure of the symmetry of the system.
In surveying manners, we shall be concerned only with the nature of the
manner-contrasts within a symmetric set. This yields results somewhat at vari-
ance with what might otherwise be obtained. In Yawelmani, for example, it is
possible to find a four-way manner-contrast recurring in the set /c c' c' s/, in
98 A MANUAL OF PHONOLOGY

/c c' c' s/, and in /k k' k' x/. However, the largest symmetric
which can be set
excised from the obstruents of Yawelmani does not include /s which are,s x/,

instead, the leftovers. Therefore we are concerned only with a three-way manner-
contrast in the language.
In a few cases symmetric sets of equal size can be drawn from the whole
obstruent system in either of two ways. This is the case for Latin, Takelma,
Sierra Popoluca, Hungarian, Supide, and Georgian. Each of these will be dealt
with in both of the possible ways, since our criteria, as stated above, afford no
grounds for choice.
Systems like Lifu, where some single contrast in effect doubles the number of
consonants, will be treated here in terms of "half-systems" that is, the obstru-
ents of whichever half includes more. But systems in which it is necessary to
make a three-way primary classification of non-laryngeal consonants, in place
of the two-way grouping into obstruents and sonorants, will be left until after
our discussion of sonorants.
25211. Stop-Position Systems. All known languages have at least some conso-
nants which are stops at least in some environments. In the survey of stop-
positions which follows, we shall use the symbols usually associated with voiceless
stops ("p," "t," "k," and so forth) to represent positions; it is not to be deduced
from this that the languages in question necessarily have voiceless stops, or even
that they necessarily have stops at all listed positions for all manners.
We shall take up, first, systems in which there occur only "pure" stops. A
great many languages have phonemes which are phonetically affricates but which
obviously go with the "pure" stops as additional "positions." Systems of this
sort we shall consider second. Some languages have affricates which are single
phonemes rather than but which count as manners rather than as
clusters,
additional positions; these, of course, will be dealt with in conjunction with our
discussion of manner systems.
Apart from affricates functioning as "positions," some languages have what
we shall call color-modified stops (labialized velars and the like). When these
are units rather than clusters, they invariably function as additional "positions."
Finally, some languages have coarticulated stops, which likewise seem always
to function as "positions" rather than as manners.
252111. Pure Stops Only. There is but one known language which has only
two positions: Hawaiian, with bilabial /p/ and lingual /k/. There is evidence
that some decades ago the latter was pronounced freely either as [t] or as [k],

and this is why we merely as "lingual." Today, the dorsal closure

characterize it

is apparently regularly used, which is why we choose the symbol "k." We might
also describe the two stops as labial and non-labial. So described, the contrast
recurs in the nasals, /m/ and /n/, though the latter is regularly apical. There
isno manner of articulation (for obstruents or sonorants) for which more than
two positions are in contrast. We may therefore safely chart the stop-position
system thus:
p k.
In Samoan (barring recent loans), only the stops /p/ and /t/ occur, and in
 A MANUAL OF PHONOLOGY 99

Fijian (with the same proviso) only /t/ and /k/. However, the situation in
these languages not like that Hawaiian, since a three-way position contrast
is

appears elsewhere in each system: Samoan /m n n/, Fijian /m n n m b n d n g/.

Our chart of stop positions for Samoan should then be
p t -
and for Fijian,
t k,
where the dashes indicate a position which occurs distinctively somewhere in
the system, though not for stops.
A three-position system of the pattern
p t k
is of course extremely widespread ; it is attested for Breton, Cheyenne, Dibabaon,
Dutch, Finnish, French, German (standard and the dialect of Brienze), Ilocano,
Latin, Maidu(?), Marshallese, Portuguese, Tagalog, Takelma, Taki-Taki,
Terena, Isthmus Zapotec.
Four-position systems are not so common. South Greenlandic Eskimo has
P t k q ;

Keres and Campa have

p t t* k ;

and Bororo has

P t t k,
where the second position is apico-labial. The recent report of this is the only
reliable evidence to show that such a position of articulation functions in speech
anywhere.
Five-position systems are extremely rare. Kota has
P t t t k
(// interdental) but ; it also has a pair of phonemes which are sometimes spirants
and sometimes affricates: [s]~[c] and [z]~[5]. We may not be justified in setting
these aside; perhaps properly we should treat Kota in the next subsection.

252112. Systems with Affricates as Positions. We classify these with a pair of

numbers, the first specifying the number of pure stop positions and the second
that of affricate positions.
A 2-1 system is found in Oneida, Seneca, and Cherokee:
t k
c .

In Seneca and Cherokee /c/ varies between apico-alveolar (like ft/) and lamino
Oneida it is always [6]. These languages lack all labial phonemes, and
alveolar; in
the 2-1 scheme looks like part of the far more widespread 3-1 scheme which
adds a bilabial stop. Four main varieties of 3-1 system can be distinguished.
Supide has
p t k
f,
(i.e., voiced and voiceless) as unit
the only clearly attested labial affricates

phonemes and there is residual doubt even here. The other three systems all
have a tongue-front affricate of some sort, but we can distinguish between
 :

100 A MANUAL OF PHONOLOGY

apico-alveolar, lamino-alveolar, and a type which varies (as in Seneca). Thus

we have
p t k
c
in Kiowa;
p t k

in Fox and English;

p t k
[oWl
in Menomini and Cree.
These last three cannot always be distinguished from our reports. Languages
which have one or another of the three include Alabama, Amahuaca, Arapaho, :

Arunta, Burmese, Chickasaw-Choctaw, Cree, Creek-Seminole, Crow, Dakota,

Delaware, English, Fox, Hidatsa, Hitchiti-Mikasuki, Iowa-Oto, Kaingang,
Karok, Kiowa, Koasati, Korean, Kraho, Mandarin, Menomini, Ojibwa, Osage,
Potawatomi, Shawnee, Spanish, Thai, Tunica, Winnebago, Chukchansi Yokuts.
One 3-2 system is quite common
p t k
c 6
in Armenian (East and West), Chitimacha, Georgian, Italian, Japanese, Ossetic,
Mesquital Otomi, Polish, Romanian, Shipibo, Tarascan, Tojolabal, Tsotsil,
Tubatulabal, Tzeltal; and perhaps in Russian (this depends on how we break
the whole Russian consonant system into half -systems; see 25224).
A second 3-2 system is reported only for Motilone:
p t k

Several 3-3 systems are attested. Mazateco has

p t k
c 5 ;

Serbo-Croatian may have the same, though the descriptions would lead us to
list the affricates rather as /c c y 6/. Chiricahua Apache, and perhaps Yuchi, have
p t k
c 6 X ;

and Tillamook, with no labials, has

t k q
c X.
6
4-1 systems are of six types. Unaaliq Eskimo and Kutenai have
p t k q
c ;

Cuzco Quechua has rather

p t k q

Turkish has
p t k k
6;
 A MANUAL OF PHONOLOGY 101

Zoque has
p t t* k

Sanskrit has, and Lifu (half -system), Badaga, and Chawchila-Choynimni-

Gashowu Yokuts have
p t t k

finally, Sierra Miwok has

p t t k

The Sanskrit /&/ may, of course, have been a pure stop [t y].
4-2 systems are also of several types. Sui, Totonac, and perhaps some Kechua
(Callejon de Huaylas) have

Hungarian has
t k
c c ;

Czech and Sierra Popoluca have

p t ty k
c 6
(though in Sierra Popoluca affrication can alternatively be interpreted as a
manner rather than as two added positions); Western Miwok, Pashto (?),
and Wikchamni Yokuts have
p t t k
c 6 ;

and Yawelmani Yokuts has

p t t k
c Q.
The evidence about Yuma is not entirely clear, but a 6-1 pattern seems to fit

p t ty t k q
c .

Our report lists /k y k w q w /, but a distributional analysis suggests that these

also
three might better be taken as clusters. A consonant of the [t y type can also ]

easily be a cluster in many languages, but the distributional facts for Yuma are
not the same for this as for the velars and it seems best to regard it as a unit /t y /.
In the above survey we have distinguished between some systems which might
well be classed together: e.g., the Sui situation with /p t k q/ and the Hungarian
situation with /p t k k/. In one structural sense the phonetic difference between
these two may be irrelevant. There are other contexts, however, in which it is
not, so that it has seemed preferable to keep them apart in the first instance;
regrouping can be undertaken by anyone who so wishes.
252113. Systems with Color-Modified Stops as Positions. All attested systems
with color-modified stop "positions" also have affricate "positions." Further-
more, with just one exception, the basic position to which the coloring is added
is dorsovelar (or two contrasting dorso-velar positions, one fronted and one
 :

102 A MANUAL OF PHONOLOGY

back), and with no exceptions at all the color-modification is produced by lip-

rounding. This last point is due largely, or perhaps wholly, to the way we have
organized matters for our discussion: palatal modification sometimes appears
for dorsals, or for apicals or laminals, but this seems to result in a somewhat
different basic position of articulation, so that the resultshave been treated in
preceding sections. When palatal modification runs through a whole system
(sonorants as well as obstruents), as in Russian or Marshallese, then we have the
sort of phenomenon discussed earlier for Lifu, and the results will be treated
in 2523.
We classify with three numbers: the first for the number of "pure" stops, the
second for the number of color-modified positions, and the third for the number
of affricate positions.
2-2-1 systems are rare. But Wichita has
t k
kw
c,
and Cuicateco has
t k
kw
c.
It has been suggested that in such cases, there being no labials, the labialized
dorso- velars be assigned the structural position of an ordinary bilabial, placing
the system therefore in the same category as the three-one type dealt with in
252112.
3-1-1 systems: Comanche, Isleta, Sandia, Taos, Picuris, and Tonka wa have
p t k
kw
c ;

Cuitlateco, Mixteco, and Villa Alta Zapotec have rather

p t k
kw
6.
Trukese has our one instance of a color-modified stop which is not dorsal
p t k
w
P
c .

There is only one attested 3-1-2 system, found in Bannock, Cashibo, Hopi,
Huasteco, Huichol, Olmeca and Sierra Nahuatl, Mazahua and Temoayan Otomi,
Sierra Zapotec, and Zuni:
p t k
kw
c 6.
 A MANUAL OF PHONOLOGY 103

There is found in Navaho and in Nahuatl

also only one attested 3-1-3 system,
dialects other than the two mentioned just above:
p t k

3-1-4: Chipewyan:

3-2-2: Coeurd'Alene:

P t - q

c 6.
3-2-3: Kalispel:

P t - q

c 5 X ;

and Tlingit:
t k q
kw q*
c c X.
4-1-2: Santa Clara Tewa:
p t t* k
kw
c 5.
4-2-2: Bella Coola and Kwakiutl:
p t k q
kw qw
c X.
4-2-3: Nootka, Duwamish, and Snoqualmie:
p t k q
kw qw
c 6 X.
252114. Systems with Coarticulation. One other
device by which a position-
system is sometimes filled out is by the use of coarticulated stops. In Bariba, for
example, there are /p/, /t/, and /k/, and also a type of stop produced by simul-
taneous closure of the lips and of dorsum and velum. This last is not a cluster
/pk/ or /kp/, for much the same reasons that a /c/ or a /k w/ is often not a
cluster: it patterns like a unit, and occurs in contrasting voiceless and voiced
manners as do the three simple stops. It is a little puzzling to know where to
place the symbol representing such a phenomenon in charts like those we have
been presenting. If one uses a single fine, then where in sequence should it be
listed? If one uses a separate fine, should it be entered under /k/ or under /p/?
We are told that in Bariba the coarticulated stops are more apt to be mistaken
for velars than for labials, so that perhaps we can regard them as types of dorso-
 :

104 A MANUAL OF PHONOLOGY

velar stops with labial modification, rather than the reverse. It is further worthy
of note that systems with a coarti ciliated /kp/ "position" do not have, in any
known case, a contrasting /k w /> an d that all known coarticulated stops involve
bilabial and dorso-velar closure, never any other combination. So we can chart
the Bariba system thus:
p t k
kp ;

this recurs in Loma, Kpelle, Mende, Yoruba, Mano, and Jukun. Gourma and
Senadi have a 4-1 system:
p t f k
kp ;

and one dialect of Jukun has a 3-1-1 system:

p t k
kp
c .

25212. Manner Systems. Here we summarize the types of contrast of manner

which are attested, naming in each case one or two languages where the type is
found, but without listing the obstruents in question; in 25213, where we
tabulate whole obstruent systems in terms of symmetric sets, positions, and
manners, fuller examples are given.
Except when especially listed, affricates count in the following as stops.
Two-way manner contrasts are of the following types
(1) (voiceless) stop versus (voiceless) spirant: Samoan, Fox.
(2) (voiceless) stop versus spirant which is voiced or voiceless depending on
environment: South Greenlandic Eskimo.
(3) (voiceless) stop versus voiced spirant: Yuma. (In a system of any of the
above three kinds, the stops may be always voiceless, or may be voiced in some
environments.)
(4) voiceless versus voiced stops: Tagalog.
(5) nonglottalized stop (sometimes slightly voiced or slightly aspirated, de-
pending on the language) versus glottalized stop: Huasteco.
(6) unaspirated stop versus aspirated stop: Tarascan, Chitimacha. In the
latter, the unaspirated stops are sometimes glottalized.

(7) voiceless versus voiced spirant Hungarian.:

In each of the above we have named first the less positively characterized or
"unmarked" manner, except that in a few cases there is no particular reason
for selecting either manner for this special status. In more complicated manner
systems there usually is good reason for such a selection.
Three-way manner contrasts are of the following types:
(1) stop, affricate, and spirant (all voiceless): Keres.
(2) unaspirated stop (sometimes voiced), aspirated stop, and voiceless spirant:
Mandarin.
(3) voiceless and voiced stops, voiceless spirants: standard German.
(4) voiceless stops; weak obstruents which vary between voiced stop, voiced
non-fricative spirant, and voiceless non-fricative spirant; and strong voiceless
spirants: Spanish.
 A MANUAL OF PHONOLOGY 105

(5) nonglottalized stop (sometimes aspirated), glottalized stop, voiceless

spirant: Tillamook.
(6) nonglottalized stop (sometimes voiced), glottalized stop, and obstruents
which are aspirated stops in some positions, voiceless spirants in others: aborigi-
nal Cuzco Quechua.
(7) voiceless stop (sometimes aspirated), glottalized stop, and voiced stop:
Western Mi wok.
(8) voiceless unaspirated stop, aspirated stop, and voiced stop: Thai.
(9) voiceless unaspirated stop (sometimes glottalized), aspirated stop, and
voiced stop: Georgian.
(10) unaspirated stop (sometimes slightly voiced), aspirated stop, and glot-
talized stop: Yawelmani.
Four- way manner contrasts are of the following types:
(1) stop versus spirant and voiceless versus voiced (or, in some cases, fortis
versus lenis) French, Dutch. This
: is and away the commonest type.
far
(2) voiceless stop, voiced stop, voiceless affricate, and voiceless spirant: Sierra
Popoluca.
(3) voiceless nonglottalized unaspirated stop, glottalized stop, aspirated stop,
and voiced stop: Sandia.
(4) voiceless stop (sometimes aspirated), glottalized stop, voiced stop, and
voiceless spirant: Kwakiutl.
Five-way manner contrasts are of two types:
(1) voiceless stop (unaspirated, sometimes glottalized), aspirated stop, voiced
stop, voiceless spirant, and voiced spirant: Georgian, Ossetic.
(2) unaspirated stop (sometimes voiced), aspirated stop, glottalized stop,
voiceless spirant, and voiced spirant: Chiricabua Apache.
Many languages, of course, have no contrasts of manner at all: Hawaiian,
Korean, Cree, Terena, Huichol. This does not imply that such a language cannot
have both stops and spirants, or even two different manners of stops: under our
criteria, as expounded in 25210, it means only that the obstruent system in-
cludes no symmetric set.
Many languages are reported as having more manner contrasts than are here
recognized. Apparent manner contrasts are sometimes due to allophonic vari-
ation, as in Ojibwa, where /p t 6 k s s/ are voiced in certain surroundings, other-
wise voiceless. It is not difficult to eliminate cases of this sort. It is somewhat
harder to be sure of eliminating all cases in which apparent contrasts of manner
are better taken as the difference between single obstruents and clusters of one
kind or another. We shall discuss here several instances in which we have pre-
ferred a cluster interpretation (and so obtained fewer genuine manner contrasts),
plus a number where we have not chosen the cluster interpretation though per-
haps one should do so.

A rather common situation is for there to be a medial, or otherwise limited,

contrast between short and long obstruent. In Finnish the contrast is found only
medially, and the obvious interpretation is that the longs are geminate clusters
/pp tt kk ss/ comparable to non-geminate clusters /pt tk ps ts ks sp st sk/ and
to geminate /mm nn rr 11/. Ojibwa has both medial and final longs, but the
106 A MANUAL OF PHONOLOGY

geminate interpretation is again fairly obvious: /pp tt 66 kk ss ss/, comparable

in distribution to /mp nt n6 nk/. We choose a similar interpretation in most
other cases.
In Potawatomi the long fortis stops and spirants cannot be interpreted as
geminates, since such geminates as /pp tt 6c kk/ occur, often with each consonant
released separately, but in any case lenis rather than fortis. However, it seems
best to extract a feature of fortisness from all the single fortis obstruents or
fortis clusters of obstruents, as something which occurs or does not occur as a
property of the whole obstruent sequence (of one or more). One then has the
contrast between /nteptan/ 'I obtain it' and /ntept-an/ 'I hear it', where the
fortis feature // spreads through the whole cluster /pt/. This is comparable to
the extraction of voicing proposed in 2512 for Polish and Portuguese.
In Delaware, also, the geminate interpretation does not at first seem possible.
One finds medial sequences such as [k'k], in which the first stop is released with
an aspiration just as in a non-geminate cluster [k't]. However, it turns out that
this aspirate release can be set up as one allophone of a special vowel, for which
we shall use the symbol /i/, and which also has murmur-release as an allophone
in certain environments. Having done this, the two "clusters" listed above can
be treated as /kik/ and /kit/, so that the medial long stops can be taken as
genuine geminates /pp tt 66 kk/. These four are then the only geminates which
occur; all other apparent longs involve /i/.
Korean has an apparently somewhat complicated situation. Initially one finds
voiceless and k s], voiceless and strongly aspirated (with
slightly aspirated [p t 6
pharyngeal constriction) and voiceless fortis long [p' t' 6' k' s'] pro-
[p' t' 6' k'],

nounced with glottal constriction. Medially one finds short and often voiced
[b d 3 g], short voiceless aspirated [s], voiceless and strongly aspirated [p' t' 6' k'],
and long voiceless (not fortis and without glottal constriction) [p- t- <r k- s*].
Finally one finds only short voiceless unaspirated (often unreleased) [p t k].
The simplest interpretation is in terms of five obstruents /p t 6 k s/ and an /h/;
all of which occur initially and medially and three of which occur finally; the
strongly aspirated initials and medials are clusters /ph th 6h kh/ the initial and
;

medial long stops and spirant are /pp tt 66 kk ss/.

In Tonkawa there occur glottalized stops which can best be interpreted as
clusters of the simple stops and /?/: /p ? t ? c? k ? /. This is supported by the
occurrence of /k w V, where the stop is released before the glottal catch is formed,
and by none of these clusters with /?/ occur finally.
distribution:
The lasttwo examples show that we are especially led to suspect a cluster
interpretation when there is some key environment (in both of the examples,
final position) in which only one manner of obstruents, or of stops, occur. This
does not mean that a cluster interpretation is always possible under these dis-
tributional circumstances, or that, if possible, it always represents a gain in

clarity over an interpretationterms of contrasting manners. But in some cases

in
one or both of these latter points do seem to hold.
In Winnebago, for example, only aspirated stops [p' t' 6' k'] occur finally.
These also occur initially and medially. In initial and medial clusters of two
consonants, with a few exceptions, the second member of the cluster is voiced:
 A MANUAL OF PHONOLOGY 107

we have, thus, medial [p'g] and [k'b]. We therefore interpret the aspirated stops
as single phonemes /p t 6 k/; the two listed clusters are /pk/ and /kp/, where
the voicing (and lack of aspiration) of the second member is allophonic. What
appear to be initial and medial unaspirated and slightly voiced stops are clusters
of /p t 6 k/ and a phoneme /h/, which is voiced in this position according to the
general rule. What appear to be initial and medial glottalized stops are clusters
of /p t 6 k/ and /?/
which, contrary to the rule (and not surprisingly, con-
sidering the physiology of production of a glottal catch) is not voiced.
In Takelma we find initial and medial contrast between voiceless, voiced, and
glottalized stops; but finally only the voiceless ones occur, plus clusters of /?/
and a stop in which the releases are separate. We can interpret initial and medial
[p' t' k' c'] as clusters / ? p ? t ?k ?s/; all four of these clusters then also occur

finally. A similar reinterpretation of initial and medial postvocalic [b d g] as

/hp ht hk/ finds some support in the other habits of clustering, but seems a bit
far-fetched. The reduction thus leaves us with two stop manners instead of

three but not with just one, as for Winnebago.
Sanskrit, and certain modern languages such as Hindustani, are often said to
have four types of stops: voiceless and voiced, intersecting unaspirated and
aspirated. both named cases the aspiration (be it voiceless or voiced) is
But in
rather patently simply the phoneme /h/, which recurs elsewhere: this leaves just
a two-way manner contrast.
Sui presents a complicated system of syllable-onsets which need not simply be
taken as separate unit phonemes, though cluster interpretations do not eliminate
the fairly great amount of skewness shown by the system. The onsets include
voiceless unaspirated stops [p t c 6 k q], voiceless aspirated stops \p' t' c' ' k' q'],

voiced stops [b d], preglottalized voiced stops [>b 'd], voiceless spirants [0 s x],
voiced spirants and nonspirantal glides [w z y y y], and preglottalized voiced
spirants or glides ['w 'y onset system also includes /h/ and /?/, and we
'7]. The
can reanalyze the obstruent onsets as follows: /p t c c k q/, /ph th ch 6h kh qh/,
/b d/, /?b ?d/, /hw hz hy h 7 /, /w z y 7 7/, / ? w ?y V-
Similar reductions could be undertaken in other cases: say in Mandarin,
Yawelmani, or Chiricahua. In Mandarin all that we could do would be to regard
the aspirated stops as clusters of the unaspirated stops and /x/; this last varies
between dorso-velar and mere aspiration, but seems to pattern distributionally
(other than its participation in the clusters just mentioned, if they be recognized)
as a dorso-velar obstruent rather than as a laryngeal. For this reason we here
reject the reinterpretation.
In Yawelmani we cannot reinterpret the aspirated and glottalized stops as
clusters of the plain stops and /h/ and /?/ unless we also recognize syllable
juncture, changing our classification of Yawelmani as to syllable-type. If we did
not do the latter, then the notation /. VkhV ./ would be ambiguous for . . . .

[. .Vk'V . .] and [. . VkhV

. .], where in the former the [k']
. . .
begins a new
.

syllable, in the latter the [k] ends one syllable and the [h] begins another. But
if we write syllable- juncture with a hyphen, then we can distinguish
the two as

/. . . V-khV ./ and /.
. . Vk-hV ./, and so on in the many other similar
. . . .

cases. The choice is trivial: the language allows of either treatment, and forces
108 A MANUAL OF PHONOLOGY

neither.We can choose a cluster interpretation, with syllable juncture; or a

componential interpretation of the more complex consonants, without syllable
juncture. More or less arbitrarily, we reject the reanalysis.
The not so clear for Chiricahua (and Navaho). Apparently we are
case is

already forced to recognize syllable juncture, and only one stop-type occurs
terminally in syllables, so that one does not have the sort of balance between
onset and coda that one finds in Yawelmani. One could reinterpret syllable-
initial [c 6 X k] as clusters with /h/, and [c' 6' X' k'] as clusters with /?/; this
would still leave two manners of spirants, which seem to be related to the voice-
less unaspirated stops just as closely as are the aspirated and the glottalized
stops,but which cannot be reinterpreted as clusters of the "basic" stops and
anything to be found elsewhere in the system. Here, again, we have rejected the
cluster reinterpretation.

25213. Tabulation of Whole Obstruent Systems. Having now discussed, sepa-

rately, positions and manners, we are ready to classify and tabulate whole
obstruent systems. We shall begin with the simplest and procede to the most
complex, using the number of manner contrasts as principle criterion, the kinds
of manner contrasts as subordinate criterion.

252131. No No Manner Contrasts. Hawaiian has only two

Symmetrical Sets:
obstruents: /p k/. Kapauku and Marshallese (half -system) have only three:
/p t k/. Arunta has four: /p t & k/. These are the only languages known to have
no spirants at all.

Many Polynesian dialects (not Hawaiian or Samoan), and many Finnish

dialects (not formal standard Finnish) have three stops and one spirant: /p t k s/.
Oneida has the same ratio, but the stop positions are different: /t c k s/; and in
Cherokee, /t c k s/.
Korean has four stops and one spirant: /p t 6 k s/. Cree and Menomini agree
except that the affricate /c/ and spirant /s/ vary between apical and laminal.
Wichita also agrees, but with different positions: /t c k k w s/.
Standard Finnish has /p t k d s/.
Terena has three stops /p t k/, two spirants /s &/ which are voiceless or voiced
depending on environment, and one spirant /v/ which is always voiced.
Comanche has five stops and one spirant: /p t c k k w s/.
Cuicateco has four stops /t 6" k k w /, one always voiceless spirant /s/, and two
always voiced spirants //3 <5/.
Huichol has six stops and one spirant: /p t c k k w s/. e"

252132. Two-Way Manner Contrast. We shall subdivide on the basis of the

type of contrast of manner found in the largest symmetrical set.

2521321. Stop Versus Spirant. Several languages have a two-by-two set

P t

f s
with various leftovers: Samoan /v/; Trukese /p w c k/; Creek-Seminole and
Hitchiti-Mikasuki /6 k 1/; Alabama and Koasati / k b !/
 .

A MANUAL OF PHONOLOGY 109

Many have
t 6
s

with leftovers: Fox, Ojibwa, Potawatomi, Tunica /p k/; Seneca /k/; Motilone
/p S V.
Shawnee has leftover /p k/ and
t 6
6 s.
Sierra Miwok has leftover /p \ k/ and
t ?
s .
Cheyenne has leftovers /p / and
t k
s x ;

Amahuaca has the same with leftovers /p c k 6/ or, perhaps, /d/ and /s/ should
be interchanged. Kutenai has the same plus /p c q !/

Hidatsa has leftovers /p t/ and

6 k
s x
A number of languages have
c 6
s s

with various leftovers: Hopi and Olmeca Nahuatl /p t k k w /; Totonac /p t k q 1/;

Sierra Nahuatl /p t k kw g/; other Nahuatl dialects /p t a k k w/; Cashibo
w
/p t k k / (and a //3/ which may belong to the obstruent system) aboriginal ;

Mazateco /t k/. <?

Choctaw-Chickasaw has a two-by-three set

p t 6
f s

with leftover /k b !/

Crow, Winnebago, and Delaware have leftover /p/ and a different two-by-
three set
t 6 k
S S X .

The Swiss-German dialect of Brienze has still a different two-by-three, with

leftover /s v/:

p t k
f s x .

Tonkawa has leftover /p t/ and

c k k*
s X xw .

Yuma has a two-by-three set

c t^ k
s x,
4P

110 A MANUAL OF PHONOLOGY

y and four which a subsidiary symmet-

plus four random leftovers /t q 1 l /, fit

ric set:

P t
v o\
Here, also, the contrast is stop versus spirant, but the two spirants are always
voiced, whereas the three of the major symmetric set are voiceless. This differ-
ence in allophonic behavior leads us to keep the two sets separate. A similar
reasoning leads to the following set for South Greenlandic Eskimo:
p k q
7 7
with leftovers /t s s/ (instead of including /t/ and /s/ in the symmetric set).

Lifu and Karok have two-by-four sets:

p t 6 k
f e x,
Lifu with leftover /\ s/, Karok only with /s/. In both cases, possibly the assign-
ments of /s/ and /&/ should be reversed.
In the two-by-five set of Sui the contrast is between voiceless stop and voice-
less spirant or nonspirantal glide; the leftovers are /b t d/:
p c 6 k q
w z y 7 7 .

2521322. Voiced Versus Voiceless Stop. Several languages have a two-by-

three set
p t k
b d g
with various leftovers: Tagalog, Ilocano, Dibabaon /s/; Latin (alternative
treatment) /f s/; Kiowa /c s z/; Takelma (alternative) /s 1 x/.
The set
p t 6 k
bg d 5
isfound in Burmese (leftovers /s z z/) and Japanese. The latter has leftovers
/c x/ and four which form a smaller symmetric set,
'
s

z 2.
Sierra Popoluca (one alternative) has
p t ty k
b d dy g
and symmetrical leftovers

.
Supide (one alternative) has
p f t k
b V d g
and symmetrical leftovers
f V
s z .
 A MANUAL OF PHONOLOGY 111

Bariba has asymmetrical leftovers /f s z/ and

p t k kp
b d b
g g .

Mazahua and Temoayan Otomi have

p t k kw
b d g w
g
plus a smaller set of leftovers
c 6
s g
z z.
Sanskrit and Badaga show two-by-five symmetrical sets of the shape
p t t k
b d d J g,
the former with leftovers /s s &/, the latter with /s/ and perhaps //
Tubatulabal and Itahan have
p t c 6 k
b d 3 3 g,
the former with leftover //, the latter with /f v s /.
Hispanized Zoque has
p t t* & k
b d d* 3 g,
and adds /f s /> plus / c / which varies between voiced and voiceless.
Cuitlateco has leftovers /s / and
p t 6 k kw
b d 3
w
g g .

The Kota system is probably best set up as a symmetrical two-by five

p t t t k
b "d d d g
with leftovers /s z s/.
Hungarian (one treatment) has a two-by six

p t c c" k k
b d 3 3 g g
plus symmetrical leftovers
f s

V z 2 .

Bannock also has a two-by-six, but with one leftover, /s/:

p t c 6 k kw
b d w
3 3 g g .

2521323. Plain Versus Glottalized Stop. Tojolabal, Tsotsil, and Tzeltal have a
two-by-five
p t c 6 k
p' t' c' 6' k\
Tojolabal leftovers are /s S/; Tsotsil /s S/ and a pecuhar bilabial phoneme
(analysis uncertain) ; Tzeltal /bds /.
112 A MANUAL OF PHONOLOGY

Huasteco has leftovers /p 8/ and /b/ ([b~/3]), and a different two-by-five:

p t c k kw
p' t' c' k' k' w .

Zuni has leftovers /s 1/ and a two-by-six

p t c c" k kw
p' t' c' 6' k' k' w .

2521324. Piom Fersws Aspirated Stop. This is attested only for Tarascan and
Chitimacha, with the same symmetrical set:
p t c c" k
p' t' c' 6' k< .

Chitimacha leftovers are /s s/; Tarascan /b d g s s/. In Chitimacha the plain

stops are sometimes glottalized.
252133. Three-Way Manner Contrast. Here, again, we subclassify on the basis
of the types of manners involved.
2521331. Stop, Affricate, and Spirant. This is attested in Keres and Campa,
with leftover /p k/
t t^
c 6
s s .

2521332. Two Stop Manners, One Spirant Manner. Mandarin has unaspirated
and aspirated stops, and voiceless spirants:

p c k
p' c' k'
f s x
with leftover /t t'/.

A good number of languages have voiceless and voiced stops, plus voiceless
spirants. Takelma (alternative) has leftover /p b 1/ and
t k
d g
s x .

Latin (alternative) had leftovers /k g/ and

P t

b d
f s.
All other cases have three-by-three sets. Taki-Taki and German:
p t k
b d g
f s x
the former with leftover /v z/, the latter with these and /if. Mesquital Otomi
has rather
p t k
b d g
f x.
 : :

A MANUAL OF PHONOLOGY 113

with an odd /z/ and a secondary symmetrical set

c 6
s

Dagur Mongol has

t c k
d 3 g
s x
with /p b/. Finally, Breton has
p t k
b d g
s x
with an uncertain set of leftovers: /< v z 2/, of which all but /z/ perhaps form
a subsidiary symmetrical set (or else the same with /</ and /</ inter-
result
changed).
Spanish belongs marginally to the type of German and Taki-Taki listed above,
/c" 3/ (some dialects only //, others also /0/). But Spanish /b d g/,
with leftovers
and 1%/ when present, have weak spirantal allophones as well as stop allophones,
which can even be voiceless in some environments.
The Loma system deviates, with voiced and voiceless stops but with voiced
spirants:
p t k
b d g
v z 7,
plus /k p g b b' f s/ and a lenis voiced labio-dental spirant in contrast to /v/.
A number of languages have plain stops (sometimes aspirated), glottalized
stops, and voiceless spirants. Tillamook
c 6 X k q
c' 6' X' k' q'

s 1 X ?!
with leftover /t t'/. Bella1 Coola:
kw
1

c k q q
c' k' k >w q' q
s X xw X X'

with leftover /V 1/ and a two-by-two set

P t
p' t'.
Coeur d'Alene
c 6 kw q w
q
c' 6' k' w q' q' w
S Xw X xw
and w three-by-two
leftover l\ g 1/ plus a
p t
p' t'

b d.
114 A MANUAL OF PHONOLOGY

Kalispel:
c c X kw q w
q
c' 6' X' k' w q' q' w

s 1 xw x xw
with a leftover two-by-two like that of Bella Coola above. Duwamish has the
same three-by-six, with a left-over two-by-three like that of Coeur d'Alene.
Nootka and Snoqualmie have a three-by-seven:
c 6 X k kw q q"

c' 6' X' k' k w q' q

s s I X Xw X X'

to which Nootka adds the two-by-two of Bella Coola and Kalispel, while Sno-
qualmie adds the three-by-two of Coeur d'Alene and asymmetric /3 3 g w /-
Similar to the above are several systems in which plain stops and glottalized
stops contrast with a series for which aspiration and spirantization must be
regarded as structurally equivalent. Aboriginal Cuzco Kechua, for example, had
an odd /s/ and the following three-by-five:
p t 6 k q
p' t' 6' k' q'
p' t' & k' q',
where the members of the last row were aspirated stops in certain environments,
voiceless spirants in others. Picuris and Taos are similar, except that the aspirated
stops fill out the pattern for certain positions, the spirants for others:
tp c k kw
p' t' c' k' k ,w
p' t' S X xw .

Picuris adds /!/, Taos /I b d g/.

2521333. Three Stop Manners. The Yokuts languages (or dialects?) contrast
plain stops (sometimes slightly voiced, more in some of the languages than in
others) with aspirated stops and with glottalized stops. Chukchansi has
b d 3 g
p t 6 k
p' t' 6' k'
plus /s s x/. Chawchila-Choynimni-Gashowu has
b d d I g
P t t 6 k
6'
P' t' V k'
plus /s s x/. Wikchamni has
b d d 3 5 g
P t t c IS k
t' t' c' 6' k'
P'
and /s s x/. Yawelmani has
b d d 3 3 g
P t t c k
p' t' t' c' k'
c'
and /s s x/.
 A MANUAL OF PHONOLOGY 115

Georgian (alternative) contrasts plain stops (voiceless, unaspirated, sometimes

glottalized) with aspirated stops and voiced stops:
p t c c" k
p' t' c' 6< k'
b d 3 3 g,
with a voicing-variable /v/ and a secondary two-by-three system
s x
Z 2 7 .

Where the Western Miwok system fits in is not clear. Ostensibly there are
voiceless, glottalized, and voiced stops:
t t
t' t'

dd,
plus odd /p b 6 6' / and a secondary two-by-two
c k
c' k'.
/c' k'/ show an environmentally conditioned variation between glottalized stop
and voiceless spirant [s x]. In the case of the major symmetrical set, it is not clear
which of the three manners is the "unmarked" type.
Thai, however, obviously contrasts voiceless unaspirated, voiceless aspirated,
and voiced like Georgian, except that the first set do not tend to be non-
distinctively glottalized:
p t k
p' t' k'

b d g
plus /I 6' f s/.

252134. Four-way Manner Contrast.

2521341. Voiced and Voiceless, Stop and Spirant. French and Brazilian Portu-
guese have
P t

b d
f s

V z

plus /k g s &/. Continental Portuguese is the same except that /b d g/ have

weak spirantal allophones.
Supide (alternative) has
f t
V d
f s

V z

with a leftover two-by-two

p k
b g-
 :

116 A MANUAL OF PHONOLOGY

Dutch has a four-by-three with no leftovers:

P t k
b d g
f s X
v z t ;

/g/ is quite rare (except in loanwords).

English, Turkish, Czech, Serbo-Croatian, and Romanian have the same main
symmetrical set
P t c
b d 3
f s s

V z 2,
to which English adds /k g <5/ (and one i night interchange the assignment of

/s z/ and /0 (5/) Romanian adds /k g c/. Turkish adds a second set

;

k k
g g >

Czech adds /c x/ and a second set

ty k
dy g j

Serbo-Croatian adds /c/ and a second set

cy k
y
3 g
Polish and Hungarian (alternate) have

P c c
b 3 3
f s

V z 2,
to which Polish adds /x/ and a two-by-two
t k
d g
while Hungarian adds a two-by-three
t k k
d g g-
Senadi has
P t ty

b d dy
f s sy

v z zy
and a subsidiary set
k kp
b
g g
The Russian half-system, finally, has /c/ and
p t 6 k
b d 3 g
f s s X
v z i 7
 :

A MANUAL OF PHONOLOGY 117

2521342. Other Four-Way Contrasts. Sierra Popoluca (alternative) contrasts

voiceless and voiced stops, voiceless affricates, and voiceless spirants:
t V
d d*
c 6
s s
with a two-by-two second set
p k
b g.
Saho, Santa Clara Tewa, and Kwakiutl contrast voiceless, voiced, and glot-
talized stops and voiceless spirants. Saho:
t k
d g
t' k'
s X
plus /&' z 2/ and a two-by-two
- b 3
f s .

Santa Clara Tewa:

P t 6 k
b d 3 g
t' 6' k'
P'
f e s X
plus /t y / and a three-by- two
c kw
c' k' w
s xw .

Kwakiutl
c X k kw q q
w
3 X g g ? r
c' V k' k w q' q
s \ X xw X X
with a three-by-two subsidiary set

P t

b d
P' t\
Sandia and Isleta have four manners for stops: voiceless unaspirated, glot-
talized, aspirated, and voiced:
p t k
p' t' k'
p' t' k'
b d g.
Isleta adds /c c' kw s s 1/; Sandia adds /s s 1/ and a two-by-two
c kw
c' k ,w .

252135. Five-Way Manner Contrast. In Georgian (one alternative treatment)

 : : .

118 A MANUAL OF PHONOLOGY

Ossetic, and East Armenian, the five-way contrast is between plain stop ("un-
marked": voiceless, sometimes globalized), aspirated stop, voiced stop, voiceless
spirant, and voiced spirant. Georgian:
c 6 k
c' V k' P t

3 5 g + p' t< + v,
s s X b d
z 7 z

where /v/ is varies as to voicing. The voicing contrast which is not

weak and
phonemic for Georgian /v/ establishes separate phonemes in Ossetic and East
Armenian, so that the system is
p c 6 k
p' c' 6' k'
b 3 3 g + t t' d.
f S S X
v z 2 y
In Athabascan (Chiricahua, Navaho, Chipewyan), on the other hand, the
five-way contrast is between plain stop ("unmarked": unaspirated and unglot-
talized, but sometimes slightly voiced), aspirated stop, glottalized stop, voiceless
spirant, and voiced spirant. Chiricahua has the simplest system of the three
languages
3 3 X g
c 6 A k
c' ' r k' + bd 1 1'

s \

z 2 l 7
Navaho adds to this the leftovers /k w x w
y /; the
w last is often simply [w]. These
additions do not yield any subsidiary symmetric set. Chipewyan has the same
four leftovers as Chiricahua, but a larger main set
X w
3 3 g g
c 6 X k kw
0' c' V r k' k' w
s s 1 x xw
S W
y 7 z 7 1 .

As Navaho, /y/ is often [w]; note that the /y/ and /l/ of Chipewyan,
in like
the /l/ of Navaho and Chiricahua, belong to the obstruent system.
2522. Sonorant Systems. The sonorant system of a language is denned nega-
tively: a pheneme a sonorant if it is not (1) part of the obstruent system (like
is

the /y 1/ of Chipewyan), (2) part of the vowel system (like the /i/ of Mandarin
or Italian, which has an allophone [y]), or (3) a laryngeal or a manner consonant
(English /h/, Menomini /h ?/> Russian /y/). Sonorants uniformly turn out to
be nasal continuants, laterals, trills, or glide vocoids, but, of all these, only
sounds of the first sort seem always to be sonorants. For that reason, we shall
deal first with nasal systems (as subsystems of sonorant systems), then with
nonnasal sonorants, and then with a handful of sonorant systems where the
nasal-nonnasal contrast is lacking or requires special handling.
 A MANUAL OF PHONOLOGY 119

As in dealing with obstruent systems, we shall handle "half -systems" in cases

like Lifu, Marshallese, and Russian, where an overriding contrast breaks almost
all the consonants into two subsets each of which includes obstruents and
sonorants.
Some languages have two manners which resembles one
for sonorants, one of
of the manners found any neat breakdown of
for obstruents, without yielding
all the non-pharyngeal and non-glottal consonants into two sets. Nootka, for

example, has /m m' n n' w w' y y'/, where the glottalization recurs (in slightly
different form phonetically) in one manner of obstruent; but there is no exact
pairing between glottalized and non-glottalized, because of the spirants in the
obstruent system. Plain versus glottalized sonorants are found not only in
Nootka, but also in Kwakiutl, many of the Salishan languages (Coeur d'Alene,
Bella Coola, Kalispel and others, but not Duwamish, Snoqualmie), Yokuts, and
Navaho. In these cases, the following sections will deal with the "half -system"
obtained by deleting the contrast between plain and glottalized. There seems to
be no other contrast which works in just this way.
25221. Nasal Systems. The numberof nasal phonemes (differentiated solely by
position of articulation) which appear in various languages ranges from none
at all to four. Two generalizations can be made, neither of them perfect. The
number of positions of articulation for nasals exceeds the number for stops only
very rarely: Samoan is a case, with /m n
but only /p t/. The only languages
rj/

with but a single nasal have an /n/, and usually have no labial consonants at all;
for example, in Iroquoian; but Winnebago and Senadi are exceptions of a peculiar
sort (252212 below).
252211. No Nasals. This is reliably reported for Quileute, Duwamish, and
Snoqualmie; probably also in a few other southern Coast Salishan dialects. In
all of these cases, it is known that the languages at an earlier period in their
histories had nasals, which have become voiced stops.

252212. One Nasal. Tillamook, and most or all of the Iroquoian languages
(Seneca, Oneida, Cherokee, Mohawk), have no and only one
labial consonants,

nasal, /n/. Arapaho also has only an /n/, though its stops include /b t c k/.
Winnebago has only one sonorant that is always a nasal: /m/. [n] and [r]
contrast only in a way which is tied up with oral and nasal vowels; we extract a
nasalization feature (2432), which leaves /r/ as a sonorant undefined as to
nasality.
There are some languages where c*i first one is led to recognize two or more
nasals, but where on further examination a different solution presents itself.
Senadi, for example, has nasal continuants [m nn g]. It also has both oral and
y

nasal vowels: in vowel sequences, either all are nasal or all are oral. After a nasal
continuant onset, only nasal vowels occur. After other onsets both nasal and
oral vowels occur. There are syllabic nasals, with a tone, but these do not con-
trast as to position of articulation, since each occurs only before a homorganic
stop or spirant. All of this can be handled with a single nasal phoneme, /n/.
This unit occurs, with a tone, as a syllable, but only directly before a consonant-
initial syllable, and the position of articulation of the /n/ is that of the following
consonant. Thus we have /nba/, /nda/, /nd y a/, and /nga/, which are phoneti-
120 A MANUAL OF PHONOLOGY

cally [mba], [nda], [n y d y a], and [rjga]. /n/ occurs, secondly, without a tone, but
immediately followed by /b d d y / or /g/ and one or two vowels in this case the :

/n/ nasalizes both the stop and the vowel or vowels. Thus /nba nda nd y a nga/
are [m4 n4 n y 4 o41- Finally, /n/ occurs directly before a vowel or two vowels,
in which case it is represented only by the nasalization of the vowel or vowels:
/bna dna d y na gna tna/ are [b4 d4 d y 4 g4 t^J. This reanalysis is not a trick: it is
an attempt to extract in the most economic way those factors which are max-
imally independent of each other in their occurrence, non-occurrence, and co-
occurrence. Some of the other languages which we shall deal with below in other
ways would perhaps allow of some such reinterpretation, but many of them
do not.
252213. Two Nasals. All systems with two or more nasals have an /n/, and
all with more than two have both /m/ and /n/. Thus systems with just two
nasals have /m n/.
This is reported for: Alabama, all Central Algonquian, Amahuaca, Arunta;
Bariba, Bella Coola; Chipewyan, Chiricahua, Chitimacha, Choctaw-Chickasaw,
Coeur d'Alene, Crow, Cuicateco, Cuitlateco; Dakota, Delaware; German (of
Brienze), Georgian; Hawaiian, Hitchiti-Mikasuki, Hopi, Huasteco, Huichol;
Kalispel, Karok, Kiowa, Koasati, Kutenai, Kwakiutl; Latin; Western Miwok,
Motilone; Nahuatl, Navaho, Nootka; Ossetic; Polish; Romanian, Russian; Saho;
Takelma, Terena, Tiwa, Tojolabal, Tonkawa, Totonac, Tsotsil, Tunica, Turkish,
Tzeltal; Yokuts (except Wikchamni); Sierra Zapotec, Zuni.

252214. Three Nasals. There are three patterns for three nasals: /m n n/,
/m n n y /, and (quite rare) /m n n/.
The first, with /n/ more limited in distribution that /m n/, is reported for:
Creek-Seminole, Dutch, English, Finnish, German, Japanese, Korean, Taki-
Taki. There seems to be no distributional limitation in: Bannock, Burmese;
Cantonese; Dibabaon; Ilocano; Kraho, Loma; Mandarin, Sierra Miwok; Sa-
moan, Supide; Tagalog, Tarascan, Thai, Tubatulabal; Wikchamni.
The second: Campa, Cashibo, Czech; French; Hungarian; Italian; Keres;
Mazateco; Otomi; Portuguese; Santa Clara Tewa, Serbo-Croatian, Spanish;
Yuma.
The third is reported for Breton and for Badaga, though the latter may have
also a phonemically separate /n/.

252215. Four Nasals. All languages with four nasals include /m n n/.
The fourth is /n y / in: Lifu (half -system), Senadi (but see reanalysis in
252212), Cuzco Kechua, Sui, and Zoque.
The fourth is /m w / in Trukese, matching the stop /p w /-
The fourth is /n/ in Kota (and possibly Badaga has these four).
South Greenlandic Eskimo distinguishes between fronted /n/ and backed /g/,
as for its stops.

25222. Non-Nasal Sonorants. A

few systems have only glide vocoids among
their non-nasal sonorants, glide vocoids at all. We deal with
and a few have no
these first, then with the commoner situation in which system involves both
glide vocoids and other types of non-nasal sonorants.
 A MANUAL OF PHONOLOGY 121

252221. Glide Vocoids Only. Chiricahua Apache and Navaho have only a
/y/.
Bannock has only a /w/. /w y/ occur in Arapaho, Chitimacha, Plains Cree, Fox,
Kraho, Menomini, Motilone, Nootka, Ojibwa, Potawatomi, Seneca, and Wik-
chamni. Winnebago has /w y/, plus the nasalization-indifferent /r/.
252222. No Glide Vocoids. /r/ only: Chipewyan, Crow.
/I/ only: Cantonese, Mandarin, Samoan, Sui, Kiowa.
/I r/ : Georgian.
/I 1 r/: Breton.
/I l
y r/ : Italian.
/I r r/ some Spanish
: dialects.
/I r r/ (the last a retroflex flap): aboriginal Tarascan.
/I l
y r r/: Castilian Spanish, Portuguese.

/I r r r/: Hispanized Tarascan.

252223. /w/ and (Mers. /w 1/: Hawaiian, Tetelcingo Nahuatl, Lifu (half-
system).
Cashibo has //3/, probably to be classed as a sonorant, and /r/.
/w r/: some Spanish (e.g., southern Mexican).
1 r

/w 1 r f/: Sierra Zapotec (the second trill is not described; this may be like
southern Mexican Spanish).
/w w
7 7 / Coeur d'Alene (the last two are dorso-velar voiced spirants
1 r :

which seem to belong to the sonorant system).

252224. /y/ and Others, /y 1/: South Greenlandic Eskimo.
/lry/: Sierra Popoluca, some French; standard German (in which /w/ is a
covowel and thus not a part of the marginal system) Russian. ;

/I r r y y/: Czech (/ry/ is an apical trill accompanied by a laminal spirant).

/I rr y/: Mazateco.
/I ly r y/: Turkish, Serbo-Croatian.
/I J r r y/: Kota (plus /v/ which perhaps belongs to the sonorant system).
252225. The commonest scheme is /w y 1/, found in Ala-
/w y/ and Others.
bama; Bella Coola, Burmese; Cherokee, Choctaw-Chickasaw, aboriginal Cuit-
lateco, some Cree dialects, Creek-Seminole; Duwamish, Dakota, Delaware;
Hitchiti-Mikasuki, Hopi, Huasteco; Koasati, Kutenai, Kalispel, Kwakiutl;
Loma; Miwok; Nahuatl (except Telelcingo); Oneida; Snoqualmie, Shawnee,
Supide; Tillamook, Takelma, Tonkawa, Totonac, Tubatulabal; Yokuts (except
Wikchamni); Zuni.
/w y r/: Amahuaca, some Cree, Isleta, Japanese, Keres, Sandia, Santa Clara
Tewa, Trukese.
Karok and Campa have //Syr/.
Korean has /w y/ and a retroflex /l/ which is sometimes lateral and some-
times not.
/w y 1 r/: Aranta, Bariba, Dibabaon, Dutch, English, German (Brienze),
Huichol, Ilocano, Hungarian, Ossetic, Otomi, Picuris, Romanian, Taos, Terena,
Tojolabal, Tunica, Tzeltal, Tagalog, Taki-Taki, Thai. Finnish and Tsotsil have
/v/ or /ft/ instead of /w/.
/wylrr/: Hispanized Cuitlateco, Hispanized Zoque.
 : ..

122 A MANUAL OF PHONOLOGY

/wylrr/: Saho (the /r/ is a retroflex flap, sometimes sounding like [d]).
/w y 1 r/: Badaga.
1

/w y y r/ Yuma, Cuzco Kechua.

1 l :

252226. More tfmn Two GZide Focotds. Some dialects of French have both /w/
and /u/ phonemically distinct from /u ii/, yielding a system /w u y r/. 1

25223. Symmetric Sets of Sonorants. The exact doubling of sonorants through

the contrast of nonglottalized and glottalized (2522) of course places them all
in a symmetric set. Apart from this, symmetric sets of sonorants are rather rare,
but the following cases appear:
Italian, Portuguese, Castilian Spanish, Serbo-Croatian, Yuma, and Cuzco
Kechua have
n ny
1 \y

Czech has rather

n ny
r ry .

Breton and Badaga have

n n
1 I.
Kota has a larger set:
n n
1 I

r r

25224. Unitary sonorant Systems. A few languages have well-defined sonorant

systems which cannot be divided into a nasal system and a nonnasal system.
Winnebago, which we treated in 252212, might also be classed in this way. A
closely related language, Hidatsa, must be: there are two sonorants, /w r/, but
these are oral [w r] in some environments, nasal [m n] in others.
Wichita also has just /w r/; the /r/ varies as does that of Hidatsa, but the
/w/ is apparently always oral.
The situation in Fijian is similar, but we defer its treatment to 2523 because
it does not have a well-defined sonorant system.
2523. Systems Not Marked by Obstruent-Sonorant Dichotomy. Here we shall
discuss two types by Lifu, where the whole
of situation: (1) that represented
set of consonants, apart from laryngeals, is more or less neatly halved by some
overriding contrast; (2) the situation in which a neat dichotomy of non-laryngeal
consonants into obstruents and sonorants is impossible.
Isthmus Zapotec is reminiscent of Lifu
p t k b d g
s s z &
N
The consonants
(m)
W L
(m)
wiry. n

in the left-hand set are voiceless fortis; those in the right-hand

set voiced and lenis. It will be noted that the matching is not exact there are no
:


A MANUAL OF PHONOLOGY 123

fortis opposite-numbers for /r y/, and /m/ is fortis or lenis depending on environ-
ment which reason we have entered the same symbol on both sides). There
(for
are, in addition, a long apical trill /r/ which occurs in a handful of native words
as well as in Spanish loans, and a bilabial trill which occurs in just one word.
Marshallese affords another example, but the basis of overriding contrast is
plain and palatal:
p t k py ty ky
m n n m y ny ny
1 P
r ry .

Other details are not clear. It is not certain whether there is a manner consonant
/y/ going with the palatals, but if there is it is apparently not feasible to treat
the palatals as clusters.
The last comment applies to Russian. In the following table we list what may
be taken as the basic "half -system" of Russian, apart from the manner consonant
/y/; a hyphen before the symbol means that the consonant in question occurs
and a hyphen after it means that it occurs in the palatal
in the plain variety,
variety. Contrastis thus certain only for those which are both preceded and fol-

lowed by hyphens. Some of the contrasts are very rare; in any case, individual
speakers vary, and it is possible that we have provided too much or too little
or both, for different speakers:
-p- -t- -c c- -k-
-b- -d- 5- -g-

-f- -s- -- -x-

-
-V- -z- -l- 7 (-)
-m- -n-
-1-

-r- .

Many under the impression that most (or all) of the Slavic
Slavicists operate
languages have the same sort of fundamental set-up of largely paired plain and
palatal consonants. But in many of the languages (not Russian, and possibly not
Polish) it is perfectly feasible to regard most of the "palatal" units as clusters
with /y/. Our option of this latter treatment for Serbo-Croatian and Czech may
be no less arbitrary than a Russian-like treatment, but it is also certainly no
more arbitrary.
Iraqi Arabic has a system in which the contrast of plain and pharyngealized
cuts across the obstruent-sonorant dichotomy, but only in a very irregular and
partial way. The largest symmetrical set which can be drawn from the obstru-
ents is

p t 6 k
b d 3 g
f s s X .

To must be added the non-pharyngealized obstruents /0 z/> and the non-

this ft"

pharyngealized sonorants /m n r 1/. Four of the obstruents are matched by

pharyngealized ones: /% k $/; there is also a /?/ which is unmatched by any
 : :

124 A MANUAL OF PHONOLOGY

plain obstruent. Only one of the sonorants is paired: /]/. The manner-consonant
going with the pharyngealized consonants is /?/> a voiced pharyngeal spirant
which is sometimes a stop. There are, in addition, /? h/ and a pharyngeal /h/
which can be regarded as a pharyngealized partner to /h/. /w y/ may be separate
phonemes, or may be only allophones of vowels.
Next we consider Fijian and Chatino, where no simple obstruent-sonorant
dichotomy is feasible. A trichotomy is possible in Fijian:
Pure Obstruents: t

s
n
Mixed obstr-sonor: *d g
Pure Sonorants:
Nasals m
Mixed nasal-nonnasal
Nonnasals

w
A similar treatment is possible in Chatino:

p t k
b d g ?h
ss

mb n
d n
g

m n
wy r 1

In neither of these cases does there seem to be any justification for forcing
the prenasalized voiceless stops into the classification of "pure obstruents" or of
"pure sonorants." However, it is possible in Chatino (not in Fijian) that they
should be treated as clusters. Mixteco resembles Chatino in general outline.
2524. Manner Consonants. Often, in analyzing a language, one hesitates for a
long time between interpreting certain consonantal elements as units or as
clusters. This is of course true of affricates, but it also happens for consonants of
contrasting manners: one may seem to have the choice between recognizing two
or more manners, or just one manner, the consonants of the other manners being
clusters.
The cluster interpretation is impossible if there is no appropriate element in
the system which can function as the "modifying" element of each cluster. Thus
a [p'] can be a cluster providing there is also a [b] or [p] and an independently

occurring /h/, but if there is no independent /h/ then an interpretation as /ph/

is precluded.
However, even if this first condition is met, the weight of evidence may be
against the cluster interpretation, or, indeed, such an interpretation may be
 A MANUAL OF PHONOLOGY 125

downright impossible. For example, a [p ] cannot be /ph/

if there is a con-
trasting cluster /ph/ in the systemunless, of course,
should turn out that
it
the latter is efficiently interpretable as /p-h/, with an intervening juncture.
In those cases in which the first condition is met, but in which the cluster
interpretation is nevertheless rejected, we propose to speak of manner consonants.
If [p'] is a unit phoneme /p'/ (paralleled by similar units at other positions of
articulation), and there is an /h/ in the system, then (other things being suit-
able)we say that /h/ is a manner consonant, going with the aspirated stops.
/h/ would seem to be a manner consonant, under this definition, in Chiti-
macha, Thai, Kechua, Lifu, and perhaps others. English and German /h/ is a
different matter: its syllable-margin allophone [h] might be said to parallel the
voiceless aspirated stops (though the latter are not aspirated in all positions),
but its postvocalic allophone (in complex peaks: scalar or lowering lengthener
or centering glide) is voiced and does not fit. In Kechua, the obstruent series

paralleling /h/ are sometimes voiceless aspirated stops, sometimes voiceless

spirants. In Lifu, /h/ parallels the whole voiceless half-system.
/?/ would seem to be a manner consonant in Huasteco, Western Miwok,
Tojolabal, Tsotsil, Tzeltal, and Zuni, where the paralleling obstruents are glot-
talized stops. The same applies in Coeur d'Alene and Snoqualmie, where we

would be led to expect also a manner consonant /h/ but /h/ does not occur.
In Czech there is a voiced /h/ which perhaps can be taken as a manner conso-
nant going with the voiced stops and spirants.
In Iraqi Arabic the pharyngeal /ty can be taken as a manner consonant going
with the various pharyngealized consonants.
In Russian, /y/ is certainly a manner consonant going with all the palatalized
consonants. The same holds in Marshallese
the language has a /y/.
if

Both /?/ and /h/ are manner consonants in Bella Coola, Duwamish, Kalispel,
Tillamook, Nootka, and Kwakiutl (perhaps others), where the former goes with
the glottalized stops and sonorants, the latter with the spirants. In Chipewyan,
Chiricahua, Yokuts, Navaho, and Santa Clara Tewa, /?/ goes with the glot-
talized s^ops (in Navaho also with the glottalized sonorants), while /h/ goes
with aspirated stops. In Tiwa, /?/ goes with the glottalized stops, /h/ with both
the aspirated stops and with voiceless spirants, there being no minimal contrasts
(both manners at the same position) for the latter.

It will be noticed that the only non-laryngeal manner consonant is the Russian

/y/. This is partly due to our method of analysis. Since one hesitates between the
unit and cluster interpretation in such cases as [k w q w x w x w ], it may be that, if
a system includes a /w/ but analysis of labialized stops and spirants as cl"sters
with /w/ is impossible, /w/ should be called a manner consonant. However,

never does one find such extensive parallel sets with and without [w]-type
modification.
2525. Laryngeals. A good number of languages have no laryngeals at all:

Arunta, South Greenlandic Eskimo, Fijian, most dialects of French, Italian,

Kota, Latin (early Latin an /h/), Loma, Mandarin, Marshallese (?), Polish,
Portuguese, Spanish, and Trukese.
126 A MANUAL OF PHONOLOGY

Many have only an /h/: Badaga, Bariba, Breton (?), Burmese, Cherokee,
Cree, Crow, Delaware, Dutch, Finnish, Fox, some dialects of French, German,
the German of Brienze, Hungarian, Korean, Lifu, Muskogean Nahuatl (Matlapa,
Sierra Nahuat, Tetelcingo), Kechua, Romanian, Sanskrit, Serbo-Croatian,
Shipibo (?), Taki-Taki, Tarascan.
Many have only an /ty: Cashibo, Chontal, Coeur d'Alene, Dibabaon, Hopi,
Ilocano, Motilone, Nahuatl (Milpa Alta), Ojibwa, Potawatomi (/h/ in inter-
jections), Samoan, Senadi, Snoqualmie, Supide, Tagalog, Terena, Totonac,
Yuma, Sierra Zapotec and Villa Alta Zapotec.
Shawnee has a single glottalic phoneme which is [h] before vowels, [ ? before ]

consonants.
Czech has only its voiced /fi/.
Georgian and Ossetic have a pharyngealized glottal catch /.'/, which appar-
ently sometimes has a very far back dorso- velar closure.
A great many languages have /h ?/: Amahuaca, Bannock, Bella Coola,
Campa, Chatino, Cheyenne, Chipewyan, Chiricahua, Chitimacha, Comanche,
Cuicateco, Cuitlateco, Dakota, Duwamish, Hawaiian, Hidatsa, Huasteco,
Huichol, Iowa-Oto, Japanese (/?/ very limited and perhaps tied up with intona-
tion), Kalispel, Karok, Keres, Kiowa, Kraho, Kutenai, Kwakiutl, Maidu,
Mazateco, Menomini, Sierra Miwok, Western Miwok, Mixteco, Nahuatl
(Olmeca, Ixcatepec), Oneida, Osage(?), Otomi, Picuris, Seneca, Sierra Popoluca,
Sui, Takelma, Taos, Thai, Tillamook, Tojolabal, Tonkawa, Tsotsil, Tubatulabal,
Tunica, Turkish, Tzeltal, Wichita, Winnebago, all Yokuts, Isthmus Zapotec,
Zoque, Zuni.
Tangsic has /h/ and a voiced /h/; possibly (but probably not) a glottal
catch /?/
Navaho, Santa Clara Tewa, Sandia, and Isleta have /? h h w /-
Saho has /h h ?/, the second a voiceless pharyngeal spirant, the third a voiced
pharyngeal spirant.
Nootka has /? h : h/, the latter two pharyngealized.
Iraqi has /? h h V-
26. Ultimate Phonologic Constituents. Some macrosegments (in some languages)
consist of a single syllable. Some syllables consist of a lone peak, or of a lone
onset; some peaks, onsets, codas, or interludes consist of single phonemes. In
just thissame way, some phonemes consist of a single ultimate phonologic con-
stituent (or component or feature), while most of them, in any single language,
consist of bundles of several ultimate components.
The elements an intonational system are quite apt to be impervious to
of
further analysis. The
three PLs of English intonation are at once intonation
elements ("suprasegmental phonemes" in American jargon) and ultimate constit-
uents. Of the three TCs, it may be that / T / can be regarded as consisting of two
components, /{/ and "rise," and / j / similarly of /|/ and "dying-away fall";
this leaves /|/ itself as consisting of a single ultimate constituent.
The elements of a linear accentual system, likewise, are usually not further
divisible: medial and soft stress in English are both ultimate phonologic constit-
uents. In non-linear accentual systems a further breakdown is often pos^ble.
 A MANUAL OF PHONOLOGY 127

It is among (segmental) phonemes, however, that one-constituent units are

rarestand multiple-constituent bundles commonest. The degree of indeter-
minacy which has characterized our statements all the way through our large-to-
small survey is not destined to disappear as we approach the lowest hierarchic

level. In surveying peak and margin systems we often found it necessary to

indicate alternative classifications of phonemes. This automatically implies an
indeterminacy as to the ultimate phonologic components found in the phonemes
so classified. For, in one sense, the constitutional classification of some set of
phonemes, and the breakdown of the same phonemes into their ultimate phono-
logic constituents, are simply two ways of doing the same thing. If we describe a
five-vowel system

by saying that two of the vowels are high, two mid, and one low, and that the
high and mid vowels are respectively front unrounded and back rounded, this is
tantamount to saying that /i/ consists of the ultimate components "high" and
"front-unrounded," and so on. The difference is largely in the grammar of our
statements: (1) "/i/ is high and front-unrounded" describes /[/ with adjectives
"high" and "front-unrounded," while (2) "/if consists of the features high and
front-unrounded" analyzes /i/ in terms of things named by nouns "high" and
"front-unrounded." Obviously, then, any indeterminacy as to the constitutional
classification ofphonemes is at the same time an indeterminacy in the analysis
of phonemes into ultimate constituents.
Yet this does not imply that the task of extracting and listing ultimate phono-
logic components can be completed by a simple mechanical transformation of our
statements about constitutional classification. The discussion and examples which
follow will demonstrate this.
We shall do two things in this section. First we shall take certain phonetically
defined features, and show how in different languages they can be of relevance
in any of many different
ways or even not at all; we shall be interested particu-
larly in showing how the hierarchic level at which a given feature makes its

appearance can vary. Then we shall give several examples of componential

analysis of specific languages, to show the degree of indeterminacy which still
characterizes our joint professional efforts along this line.
261. Sample Features. We shall consider first nasality and then tone.

2611. Nasalization. In a very few languages, as we have seen (e.g., Quileute),

nasalization plays no role at all. In most it plays some role, but this is highly
variable: we can Fox (or English), Oneida,
distinguish at least its function in
Taos, Cashibo, Chiricahua, Winnebago, Hidatsa, and Senadi.
In Fox (or English, or any of many other languages), nasalization functions
as an ultimate phonologic component for consonants only Fox /m n/ are nasal :

(and voiced), /p t/ oral (and usually voiceless); in English only nasalization

distinguished /m n rj/ and /b d g/.
In Oneida, one consonant is nasal (/n/) in the single vowel system, two of the
;

six vowels have nasality as a secondary characteristic.

128 A MANUAL OF PHONOLOGY

In Taos, again, nasalization functions for consonants, but also, and quite
independently, for vowels. The situation for consonants is much as in Fox or
English. For vowels, nasalization a component which participates on a par
is

with tongue and lip position: /[/ a coordinate construction of simultaneous

is

high, front unrounded, and velic open. There is only one other possibility: to
take the nasalization as nucleus, and the tongue and lip position as satellite.
This is suggested by the fact that in the few one-syllable sequences of two vowels,
it is the whole sequence which is nasal or oral. It will be remembered that the

nasal vowels cannot be regarded as some subset of the oral vowels with added
nasalization (2446); this is why the nasalization cannot be regarded as a simul-
taneous satellite to the remainder.
In Cashibo, nasalization functions for the consonant system as in Taos, Fox,
or English, but quite differently in the vowel system. All the vowels may be
modified by adding nasalization, and this satellite feature applies separately to
each vowel-occurrence: sequences such as /ai/, /aj/, /aj/, and /[/ all contrast.
In Chiricahua, consonantal nasalization is once again about the same; the
additional factis that a nasal sonorant can occur as satellite to a tone, yielding

a mora, one or two of which constitute a syllable (/m/, /mm/). In syllables with
vowels as peaks, nasalization occurs as satellite to the whole remaining peak,
be it one vowel or two. When the nasalization, if present, is removed, the re-
mainder consists of one vowel or two, each with a tone, and the tongue and lip
positions are satellite to the tone. In Mazateco, likewise, nasalization makes its
appearance at the level of the whole syllable peak, simple or complex, not for
the individual vowels.
In Winnebago we find the first trace of interdependance between nasalization
for consonants and for vowels (2432). In Hidatsa we find a further step in this
direction: nasalized vowels occur, in contrast to oral vowels, after all conso-
nants, but nasal consonants are allophones of oral sonorants in position before
nasalized vowels : /wa wa. ra ra7 are [wa mq, ra naj.
In Senadi, a "long component" which stretches either
finally, nasalization is

through one or two vowels (/k na/, /k naa/, b n a/) or through a stop consonant
/b d d y g/ and the vowel or vowels that follow it (/ n ba n baa/ = [m4 m44D
(252212).
This survey shows, at least, that it is trivial merely to ask of a language
"does it have nasal vowels?" The possible answers are not just "yes" and "no,"
but manifold.
2612. Tone. Setting aside the use of pitch in intonation, we still find a variety
of ways which tone can function in a phonologic system.
in
In Chiricahua, a tone-occurrence forms the basic unit of duration: anything
which is simultaneous with the tone-occurrence is satellite to it; a syllable (de-
fined by syllable juncture) includes one or two tone-occurrences.
is more complicated: a vowel, or a nasal
In Senadi the basic unit of duration
continuant with a tone. A vowel
accompanied by a simultaneous satellite con-
is

sisting of one tone or of two in sequence, but the single vowel-occurrence lasts
about the same length of time whether it carries one tone-occurrence or two.
 A MANUAL OF PHONOLOGY 129

On this basis, we may say that a syllabic nasal has the nasality as nucleus and
the tone (always just one) as satellite.
In Bariba the tone-occurrence not only the basic unit of duration, but
is

actually defines the syllable there

many syllables in an utterance as
are as
there are tone-occurrences. The vowel color or nasal sonorant which accompanies
a tone-occurrence is therefore satellite to it, as in Chiricahua.
In Mazateco a sequence of one, two, or three tone-occurrences stands in a co-
ordinate construction with a sequence of one, two, or three vowels to form a
peak, to which, in turn, nasality may be added as a satellite; the number of
tone-occurrences does not correlate with the number of vowel-occurrences.
Mandarin, Cantonese, Burmese, Thai, and the like, the domain of a
Finally, in
tone is the microsegment in most of these languages the microsegment is simply
:

a syllable (denned by syllable juncture), but in Mandarin it may include a simple

interlude.
Phonologically, then, it is just as unrigorous to speak of "tone languages" as
it would be to speak of "nasalized vowel languages."
Examples of Componential Analysis. Our first example will be Fox. The
262.
phonemes which we shall assume are /pt6ksshmnwyioea /. It may be -

that /w y/ are in complementation with /o i/, but for the sake of simplicity
we shall assume that this is not the case, so that these constitute four phonemes,
not two.
/p t c k s h m
n w y/ occur only marginally, whereas /i o e a / occur only
in or as peaks. We
might at first seek to find some feature present in all of the
first group and in none of the second, or else vice versa, which correlates with

this: say a feature nonsyllabic versus a feature syllabic. But in the case of /p t 6
k s s h m n/ the other features which are bundled together imply necessarily
(for Fox) the presence of the so-called nonsyllabic feature; in the case of /e a /

the other features which are bundled together imply with the same necessity
the presence of the syllabic feature. It is only in the case of /w y/ versus /o i/
that the difference might be said to lie in the presence of nonsyllabic versus
syllabic. It will be more economic if we can detect some other articulatory char-
acteristic in /w y/ versus /o i/, and set this up as the primary determinant of
the contrast, regarding the matter of nonsyllabic and syllabic as a resultant of it.

Now the factor which seems primarily to render /w y/ nonsyllabic, as over

against the similar /oi/, is a factor of timing: /w y/ do not last as long as /o i/;

they are glide-vocoids rather than peak-vocoids (165). They are also usually a
bit higher, and perhaps normally pronounced with less force than /o i/. All of
this we will summarize (since the various factors are not separable) by saying
that /w y/ contain a feature shortness which is not present in /o i/.
/w y o i/ all contain a feature high (referring to tongue-height), while /e a/
contain a contrasting feature low. /w o a/ contain a feature back, while /y i e/
contain front. These features obviously are paired: each of the set /w y i o e a/
contains high or low, and also front or back, while other phonemes do not contain
any of the four. Similarly, while contain shortness, and /o i/ distinctively
/w y/
lack it we may say they contain nonshortness no other phoneme (not even
130 A MANUAL OF PHONOLOGY

/e a/) contains either shortness or nonshortness; this pair of features is operative

only for bundles of features which include high.
any given occurrence will contain high or low, and front or back, but this
// in
will depend on environment (high and front when // follows /i/, and so on).
We can interpret /-/in either of two ways. One is as a phoneme which is not
composed of articulatory features, but purely of a timing factor: // means
"more of the same." By this interpretation, // is like punctation in music: the
note itself represents all the "features" (pitch and so on) while the dot has to
do entirely with timing. But we do find one feature constantly present in //,
and so a second interpretation is possible: // is always voiced, and is the only
phoneme of which this is true. We will choose this second interpretation, set up a
feature voiced, and say that // contains this feature and only this one.
For /m n w y i o e a/ this last feature, voiced, is present or not depending on
environment, and so neither voiced nor its opposite, voiceless, is a distinctive
component. But /p t 6 k s h/ are apparently always voiceless, so that we can
recognize voiceless as one feature in each of those bundles.
t 6 k s s/ are obstruents, and /m n w y/ are sonorants. However, we cannot
/p
setup a feature obstruent or sonorant; doing so would imply that all obstruents
have some clear articulatory characteristic in common which differentiates them
from sonorants and other nonobstruents. When the terms "obstruent" and
"sonorant" were introduced (2520) it was necessary to give first a rough cross-
language definition, which defined "obstruent" as stop or spirant; and then it
was necessary to indicate that in one or another language some spirants function
not as part of the obstruent system but with the sonorants (e.g., the /v/ of
Finnish, working like a /w/), while some apparent sonorants function as part
of an obstruent system (e.g., the /y/ of Chipewyan, which pairs with voiceless
// as /z/ pairs with /s/). And in Fox it is simply not the case that /p t 6 k s /
share some articulatory feature absent from all other phonemes. We will be able
to define the class of Fox obstruents, and the class of Fox sonorants, in terms of
the features we do extract, but not vice versa.
/p t 6 k m n/ all involve some complete oral closure: we will call this feature
stop (which in this context obviously does not imply concomitant velic closure),
/m n/ contain also the feature nasal, defined as velic open. Since nasal is present
only when stop is, we regard nonnasal as distinctively present in /p t 6 k/, but
outside of these six phonemes the contrast between nasal and nonnasal is ir-

relevant.
/p m/ contain labial, a "position of articulation" feature, /t s n/ contain
apical; /& s/ contain laminal, and /k/ contains dorsal. /&/ has spirantal offglide,
and perhaps in /h/, but we do not have to set this up as a
this recurs in /s /,
separate feature, since we can predict when it will occur. For /s s/ also contain a
feature spirant; we can say that any bundle containing laminal, or spirant, or
both, automatically has spirantal offglide.
The features stop and nasal apply only when a "position of articulation"
feature is present in a bundle, and nasal, moreover, only in a bundle containing
 A MANUAL OF PHONOLOGY 131

stop. However, a bundle containing a "position of articulation" feature need not

contain stop; it may contain spirant instead. We can interpret stop and spirant
as opposites, as are nasal and nonnasal, or high and low.
However, one possible interpretation for /h/ is that it contains only spirant,
not also any "position of articulation" feature. In medial clusters like /hp ht
he" hk/, there is sometimes some local friction in the mouth for the /h/, at the

position of the following consonant; this is no constant feature of /h/, which

occurs also in a good many other forms: as voiceless glottal spirant (in which
form a macrosegment-final voiceless or whispered /. &hwa/ differs from . .

/. . &wa/, where the italicized portion indicates the portion pronounced voice-
.

lessly), and as a voiceless vocoid in the position of a following voiced vowel.

Spirant (liberally interpreted), without position of articulation, seems a better
common denominator for this range of allophones than a possible alternative:
that of taking /h/ as containing voiceless as its only constantly present positive
feature.
One other feature which occurs quite regularly need not be regarded as pri-
mary: rounding, /w/ and /o/ are rounded, and no other phonemes are. We can
say, however, that any bundle which contains high and back also contains, as a
secondary implied concomitant, the feature of rounding.
We have now itemized the following sixteen features, in six pairs and one set
of four mutually exclusive ones
shortness : nonshortness
high : low
back : front
voiced : voiceless

stop : spirant
nasal : nonnasal
labial : apical : laminal : dorsal .

We began with sixteen phonemes, and have discovered sixteen primary ultimate
phonologic components, which means that there has been no achievement of any
greater "economy" of any kind. However, despite some inclination to think in
this direction among current linguists, this is not the point of the analysis of
phonemes into ultimate phonologic components. The point is to find out, in as

determinate a way as possible, what actually distinct types of articulatory events

do in fact comprise the stock of ultimate phonologic components of a phonologic

system, and the various interrelationships in which they stand. Above we have
merely a list. The ways in which these sixteen components participate in the
bundles we have called the Fox phonemes can be shown by a table. In the table,
the first line is for the four "position of articulation" components, the relevant
presence of which is indicated by the abbreviations la ap lm do. The remaining
lines are for the six pairs of mutually exclusive components: a plus indicates the
presence of the one listed at the left above, and a minus the presence of its
opposite.
132 A MANUAL OP PHONOLOGY

/P t 6 ks hmn w y o i a e 7
pos.
stop
la

+ + + +
ap lm do
--- + +
ap lm la ap

nasal - - - + +
shortness + +
high + + + +
back + - + - + -
voiced +
If we had accepted the other proposed analysis for // (as a timing feature), then
that phoneme would be omitted from the table and the last line would be left
out, since voiced and voiceless would be completely determined by the other
features in a bundle plus the position of that bundle in the macrosegment and
relative to //.
By the analysis presented in the table, there a single phoneme (//) which is

consists of a single distinctive feature; three, /h a e/, which consist of two;

eight, /s m n w y o i/, which consist of three; and four, /p t 6 k/, which con-
sist of four. By the alternative analysis of //, one phoneme consists of a single

feature, /h/; four of two, /she/; and ten of three, /p t 6 k m n w y o i/.

We transcribe two brief utterances, /ihkwe-wa/ 'woman' and /newa-pama"wa/
'I see him, her', each said as a macrosegment with the most usual intonation

pattern. In each row we put plusses and minusses in parentheses when that
feature is present for that particular occurrence of the phoneme because of its
environment.

A h k w e w a/

pos. do
stop - +
nasal (-) (-) - (-) (-) (-) (-) (-)
shortness + +
high + -4-
1
- (-) + -
back - + - (-) + +
voiced (+) ( + ) ( + ) + (-) (-)

The // in the transcription of /ihkwe-wa/ is low and front because it accom-

panies a vowel which is; that vowel, /e/, is voiced because it precedes //. The
terminal /wa/ is voiceless as part of the normal intonation. '

/n e w a pa m a
w a/

pos. ap la la
stop + + +
nasal + () (-) (-) (-) - () + () () (-) (")
shortness + +
high - + - (-) (-) + -
back - + + (+) + + (+) + +
voiced (+) ( + ) (+) (+) + -(+) (+) (+) + (-) (-)
 : :

A MANUAL OF PHONOLOGY 133

The plus-or-minus sign in the nasal row in the transcription of /newa-pama--

wa/ represents optional partial nasalization of a vowel adjacent to a nasal
continuant.
We give as an additional example an analysis of Nootka phonemes into
ultimate phonologic constituents, but without preliminary discussion. For
Nootka we will interpret // as a "timing" feature, not composed of any con-
catenation of articulatory positions or motions, and will therefore omit it from
consideration. The distinctive features are:
labial, apical, laminal, lateral, front dorsal, back dorsal (the "position of
articulation" features);
spirant, glottal, nasal, spirantal release;
pharyngeal constriction;
rounded, front;
shortness;
high, mid, low.
Shortness applies only in the presence of high, as in Fox. In conjunction with
high or mid, front and rounded pair off, front implying unrounded and rounded
implying back. Pharyngeal constriction applies only in bundles containing spirant
or glottal and not containing a "position of articulation" feature. A "position of
articulation," unless accompanied by spirant, implies stop, which can thus be
regarded as a secondary feature. Spirant implies spirantal but not vice
release,
versa, and minimum contrasts require the recognition of the latter as a separate
primary feature. Voicing always present when the bundle contains high, mid,
is

or low, or when it contains nasal, and never present otherwise, so it is not a

primary feature.
There are so many phonemes in Nootka that we present them in a succession
of tables.
First the set involving high, mid, or low (hi, mi, lo on the table; "ro" = rounded,
"fr" = front);

/w w' u y y' i e a/

height: hi hi hi hi hi hi mi mi lo
ro : fr ro ro ro fr fr fr ro fr

shortness + + + +
glottal - + +
Next those involving labial or apical (la and ap on the table)

/p P' m m' t t' n n' c c' s/

position la la la la ap ap ap ap ap ap ap
spirant: +
spir. rel. - - - + + +
glottal + - + - + - + - +
nasal + + - + +
 : :: : :

134 A MANUAL OF PHONOLOGY

Next the set involving laminal (lm) or lateral (It)

/ 6> V 1/

position lm lm lm It It It

spirant - + - - +
glottal + - +
spirant rel. + + + + + +
nasal

Next the set involving front dorsal or back dorsal (fd, bd)

/k k' x kw k' w xw q q' x qw q' w xw/

position
spirant -- + -- + -- bd+ bd--bd
fd fd fd fd fd fd bd bd bd
+
glottal - + -- + -- + -- + -
spir. rel. -- + -- + -- + -- +
nasal
rounded --- + + + --- + + +
Finally, the laryngeals:

n h >
h/

spirant + +
glottal + +
phar. constr. + +
Notice that the two consonants /?/ and /h/ have been treated as containing a
and spirant; that
single (positive) distinctive feature each, respectively glottal
these are mutually exclusive (we could have plotted them in a single line in the
tables) and that each functions as a differential for a whole series of consonants.
;

That many consonants which do not include glottal are paired by consonants
is,

identical save that they do; and similarly for spirant. /?/ and /h/ are, of course,
manner consonants in Nootka (2524).
We transcribe a short Nootka utterance ('hunting wolves'?):

K a- q' a n a V a ? r V
position bd bd ap It

sp :gl gl gl gl gl sp
spir rel. - + +
nasal +
phar constr + + ( ) +
height lo lo lo lo hi
ro : fr fr
shortness -
 A MANUAL OF PHONOLOGY 135

It is obvious thatwhole utterances or texts could be transcribed as we have

transcribed a few words of Fox and one of Nootka, with a considerable range of
possible conventions of symbolization. Such a transcription is comparable to the
full score of a piece for orchestra, or even a piano piece written out on a grand

staff; in contrast, our usual phonemic notation, consisting of an essentially linear

sequence of symbols, with some diacritics, is comparable to a figured bass. There

was a period in the history of music when the use of figured bass notation was
quite adequate: performers were adept at "realizing" a figured bass, at a key-
board instrument, in a way which might vary somewhat from one performance
to another, but such that all variation fell within the range of what was "non-
distinctive" relative to the musical tradition of the time. This was true in large
part because the total number of permitted simultaneous bundles of notes was
relatively small, and the sequences in which they were permitted to follow each
other were likewise highly restricted. For more complicated music a figured bass
is not adequate. That we are able to use essentially linear transcriptions for
speech is due to precisely the factors that rendered figured bass a reasonable and

usable notation earlier in the history of music the total variety of combinations
:

of articulatory motion, and the sequences in which the various combinations

occur, are in every language relatively limited and small in number.
The comparison of "full score" componential transcription with a full orches-
tral score in music breaks down at one point, if we use for the former such
componential analyses as those which have been presented above. In an orches-
tral score there is a line (a staff) for each instrument, and on it are placed the
marks indicating at each moment what that instrument should be doing. Now
the "instruments," in the case of speech, are certainly the various articulators
in the mouth and the movable parts in the throat and at the back entrance to
the nose. But we have not provided, in the "full scores" given above, a separate
"staff" for each "instrument" in this sense: rather, since in both Fox and Nootka
oral articulators function almost exclusively one at a time, we have specified,
along one single "staff" of the "score," which articulator is to function, and
along other "staves" what function it is to perform.
It is not too difficult to reanalyze the segmental phonemes of Fox or Nootka
in order to achieve a "full score" notation more directly comparable to that in
music. We shall do so for Nootka, which offers a bit more challenge in this
connection.
must have a staff for each of the following "instruments": the lower lip;
We
the apex of the tongue; the blade of the tongue; the tongue as a whole; the front
part of the dorsum; the back part of the dorsum; the velic; the pharynx; and
the glottis.
For the glottis we must have two symbols for different functions: one for
closure, and one for voicing. We shall use k for the former, v for the latter.
Indication of voicing could be omitted, since, as we have already shown, its
occurrence and non-occurrence are predictable in terms of what other features
are present in a bundle, but it seems advantageous for our present purposes to
work from the allophonic rather than the strict phonemic level. Where neither of
these symbols appears on the glottis staff, the glottis is quiescent.
136 A MANUAL OF PHONOLOGY

For the pharynx we need only one symbol: c for constriction. The absence of
this mark means that the pharynx is relaxed and thus open. For the velic, also,
a single symbol will do: o for open. We choose this rather than a positive symboli-
zation for "closed" since it will save work: the velic is usually closed.
For the back and front dorsum we need two symbols: k for closed (stop clo-
sure), and s for spirantal approximation producing turbulence in the passing air.
When neither of these is written, as usual, the articulator in question is doing
nothing relevant.
For the blade of the tongue we have to distinguish between the following
functions: k for closure against the alveolum and adjacent regions; s for spirantal
approximation thereto; ks for closure followed by spirantal release; 1 for closure
of the central part against the alveolum and adjacent regions but with spirantal
approximation at the sides; kl for closure followed by this type of spirantal re-
lease. The symbol k will not occur along the blade staff save in these two combi-
nations.
For the tip of the tongue the symbols k, s, and ks will suffice.
For the tongue as a whole we have to distinguish between high, mid, and low,
and for the first two of those also between front and back: we will use i, u, e, o,
and a for the five combinations.
For the lips we need two symbols, k for closure and w for rounding.
All these "instruments" are like percussion instruments in an orchestra in
that they are individually capable of very few distinct roles. In an orchestral
score an instrument which is confined, or virtually confined, to either making a

sound or not say a bass drum or a pair of cymbals is scored not on a full staff
of five lines but on a staff of a single line; if there are some few differences of
function, different symbols on that single line are used. In our articulatory score
we will use spaces between lines, instead of lines, as our separate staves, just as
we did for the first analysis.
Two features of timing have to be symbolized. One is the "shortness" feature
by the first analysis: we willput a breve over the symbol on the tongue-as-a-
whole staff. The other is //, which we will indicate as in linear transcription with
a postposed dot, also on the tongue-as-a-whole staff.

With these conventions, the "orchestral score" for the word transcribed earlier
in another way looks like this:

/q' a- q' a n a r a ? V h/
lips

tongue a- a a a V
tip k
blade kl
fd
bd k k
velic

pharynx c c

glottis k V k V V V k v k V
 A MANUAL OF PHONOLOGY 137

We show three more short utterances: 'thus much', 'three', and 'seven':

w
/q a m' a*/ /q a 6 c' a/ n a X p u/

lips w k k w
tongue a a -
a a a u
tip ks
blade ks kl

fd
bd k k
velic

pharynx
glottis v kv v v k v k V V

The essential difference between the two methods of componential analysis

which have been demonstrated lies in this: in one, positions and manners of
articulation are alike regarded as features all possible features are interpreted
as constituting a homogeneous class, except insofar as privileges of occurrence
with respect to each other serves to divide them up. In the other, positions of
articulation (or, rather, actively functioning articulators) are regarded as items
individually capable of two or more differentiated functions, some of those

functions being the same for two or more organs, some of them
There is no
not.
question as to "right" and "wrong" here: the two sorts of analysis produce
mutually convertible results, and the fact that neither forces itself on us to the
exclusion of the other is but another aspect of the relativism or indeterminacy
of phonologic pattern.
Quite apart from the use of componential analysis in discovering the stock of
ultimate phonologic constituents of a language (which can often be accomplished
without resort to "articulatory score" transcription), the second kind of analysis
is heuristically useful. Ifbegin with a large set of articulatory scores for a
we
particular language, which include indication of all occurrences of any feature

which may in some cases be phonologically relevant including, say, the round-
ing of Fox /w/ and /of, or the aspiration of Nootka plain stops not followeu
by a vowel, or the difference between the spirant and stop allophones of Spanish
/b d g/ then careful examination of the combinations in which all these features
occur, not only in simultaneous bundles but a o in sequences, can be of material
assistance in seeing the relative status of different features: that the presence of a
particular feature implies, or is mutually exclusive with, the presence of another;
that the occurrences of some particular feature are completely predictable in
terms of others; that a particular feature is predictable in certain environments,
but not in others; and so on. One is not likely to reach the point of such an
examination without first having achieved at least a tentative phonemicization
of the language and a tentative linear notation, but the examination can reveal
interrelationships that are otherwise harder to detect, or which may otherwise
even remain hidden altogether.
138 A MANUAL OF PHONOLOGY

27. Balance and Symmetry. Problems of balance and symmetry are related,
but distinguishable. Considerations of both kinds have come up in the course of
our typologic survey, but some summarizing statements are needed, and some
aspects have not come up at all.
271. Balance. We have found using reasonably homogeneous analytic tech-
niques from one language to another vowel systems with as few as three mem-
bers and with as many as twelve, and consonant systems with as few as eight
(Hawaiian) and as many as forty-two (Kwakiutl). Some languages in the
Caucasus are reported to have an even larger number of consonants, but we
cannot be sure that the same principles were used, by those making the report,
in deciding between single-phoneme and cluster interpretations. Now, speaking
roughly in terms of "consonant" and "vowel," what are the ranges of possible
balance between the number of consonants and the number of vowels in a single
phonologic system? Can a language have eight consonants and three vowels?
Can one have eight consonants and twelve vowels? Forty- two consonants and
three vowels? Forty-two consonants and twelve vowels?
Obviously we cannot hope to answer such questions definitively, partly because
there are too many chances for alternative interpretations of phonologic systems,
but mainly because there are too vast a number of languages for which we have
no reliable data. This is no reason not to give tentative and partial answers;
perhaps we can at least discern some general tendencies. What we have done is
to compute the ratio of the number of vowel phonemes to the total number of
segmental phonemes for some sixty-odd languages on which our information is
reasonably accurate. In this operation, covowels, such as the // of Fox or the
raising and lowering lengtheners of Cree, were counted with the vowel phonemes,
but distinctions made by other than tongue and lip position, say nasalization,
were omitted. The graph in Figure 12 shows the results. A vertical line appears
at each point on the abscissa for which at least one language was found: that is,
none of the languages examined had fourteen phonemes, or thirty-five. The
lower curve shows the lowest ratios of vowel phonemes to all phonemes found in
the different systems, while the upper curve shows the highest ratios.
In interpreting the graph it must of course be remembered that there were
relatively fewer instances of systems with very few or very many phonemes than
there were with an intermediate number. Despite this, we can probably safely
read the following generalizations: The range of possible ratios of vowels to all

phonemes is smallest in systems with very few or very many phonemes. The
actual ratios tend to be lower in systems with very many phonemes than in those
with very few. The greatest range of possible ratios is found in those systems with
a moderate total number of phonemes.
Of the four theoretical extremes asked about earlier, we can with some safety
conclude that only one is apt to occur
that with very many consonants and
very few vowels. With as few consonants as eight or ten, we will not find the
smallest number of vowels, nor the greatest number. And with very many
consonants we will not find very large numbers of vowels.
Other questions can be asked about balance. What are the highest and lowest
140 A MANUAL OF PHONOLOGY

ratios of number of obstruents to number of sonorants? Some of the ratios are:

Hawaiian .5, Marshallese .6, Samoan .8, Trukese 1.0, Hidatsa 3.0, Nootka 3.375,
Chipewyan 11.33. Since fuller consonant systems are almost always built via a
proliferation of obstruents, we can expect that the highest ratios will be found
in the languages with the largest number of consonant phonemes; this is con-
firmed by the above examples since Hawaiian has 8 consonants, Marshallese 16,
Samoan 10, Trukese 14, 10, Nootka 37, Chipewyan 39.
Hidatsa
What are the highestand lowest ratios of stops and affricates to spirants?
Of stops and affricates to nasals? Of nasals to non-nasal sonorants? A good many
questions of this kind were asked, and tentatively answered, in the course of our
survey of margin systems, and we need not repeat the data here.
272. Symmetry. The matter of symmetry is different. A set of four things is
symmetric if in some sense a is to b as c is to d. There can be no question of
symmetry where a and c are consonants and 6 and d are vowels, save in the limit-
ing case where a and c are consonants of the type /w y/. Fewer than four things
cannot manifest symmetry in this sense. It makes sense to say that English /t/
is to /k/ as /d/ is to /g/ it does not make sense to seek an x such that English
;

/t/ is to /k/ as /p/ is to x.

Every phonologic system seems to have at least some symmetry, and at least
some lack of symmetry. What are the symmetries of Fox phonemes? The four
vowels constitute a symmetric set (/i/ is to /o/ as /e/ is to /a/, or, with a trans-
formation, /i/ is to /e/ as /o/ is to /a/). The set /w y o i/ form another; by
virtue of this, the set /w y a e/ do also. The four consonants /p t m n/ are sym-
metric, and the four consonants /t 6 s s/. Two consonant phonemes, /k/ and /h/,
do not fall into any of these symmetries; likewise the covowel // does not.
Since there are sixteen phonemes in all, perhaps the ratio of three to sixteen can
serve as a rough index of the asymmetry of the whole system. In Yawelmani
Yokuts, with thirty-nine phonemes, only /a/ and // stand unrelated; the ratio
is about .05. Nootka, with forty-three phonemes, also shows just /a/ and //

standing outside. In Hawaiian, with thirteen phonemes, three, /a ? h/, partici-

pate in no symmetric set; this assumes that we can define /k/, /n/, and /l/ as
lingual and say that /p/ is to /k/ as /m/ is to /n/ as /w/ is to /l/. This last
example shows that it is not always entirely obvious whether a particular sym-
metry is to be recognized in a system or not.
Any "square," "rectangular," or "cubical" vowel system shows complete sym-
metry: the two-by-two system of Fox, the three-by-three system of English,
the two-by-two-by-two system of Turkish. Any other vowel system involves lack
of symmetry. Perhaps a three-plus-two-plus-one system, like the /i i u, e o, a/
of Sierra Popoluca, shows as little symmetry as one can find.
We can achieve a measure of the symmetry of obstruent systems by taking
the ratio of the obstruents which belong to a symmetric set (2521) to the total
number. The raw-materials for this computation are at hand, for a number of
languages, in 25213. Here are the results, for some of these languages, arranged
in order of decreasing symmetry:
1.00 Kwakiutl, Sierra Popoluca, Hungarian, Senadi, Turkish, Dutch, Supide,
Nootka, Duwamish, Isthmus Zapotec.
 A MANUAL OF PHONOLOGY 141

.95 Georgian (counting /v/ as an obstruent).

.94 Russian (half-system), Polish, Picuris, Serbo-Croatian.
.93 Aboriginal Cuzco Kechua.
.90 Snoqualmie, Bella Coola.
.89 Czech.
.88 Chipewyan, Coeur d'Alene, Tillamook.
.87 Ossetic.
.86 Yawelmani, Japanese, Crow, Winnebago, Delaware, Tagalog, Ilocano.
.84 Sandia.
.83 Chiricahua, Chitimacha, Cuitlateco.
.82 Taki-Taki, Spanish, Mandarin.
.80 Zuni, Lifu (half-system), Samoan.
.79 Taos.
.77 Sui.
.76 Kota.
.75 English, German, Keres, Campa, Tonkawa, the German dialect of
Brienze.
.74 Navaho.
.73 Bariba.
.71 Huasteco, Hispanized Zoque, Italian, Yuma.
.69 Thai.
.67 Isleta, French, Portuguese, Western Miwok, Takelma, Tarascan, Bur-
mese, South Greenlandic Eskimo, Choctaw-Chickasaw, Cheyenne,
Hidatsa, Fox, Ojibwa, Potawatomi, Tunica, Shawnee.
.64 Breton.
.60 Loma.
.57 Sierra Miwok, Trukese, Creek-Seminole, Hitchiti-Mikasuki.
.53 Fausa.
.50 Mazateco, Hopi, Olmeca Nahuatl, Alabama, Koasati, Saho.
.44 Totonac, Sierra Nahuatl, most other Nahuatl dialects.
.00 Hawaiian, Kapauku, Marshallese (half -system), Arui.ta, most Poly-
nesian, Finnish, Oneida, Korean, Cree, Menomini, Wichita, Terena,
Huichol.
The jump from .00 to .44 is not surprising: an obstruent system can have no
symmetry at all unless it includes at least four obstruents, and almost all systems
with more than six or seven obstruents have four of them in symmetry. There
is a rough tendency for systems with the largest number of units to have a greater

degree of symmetry, but the range of variation is pretty wide.

Of especial interest is a "hole" in a system where one would expect an element
because of predominating symmetric patterns. One such is in the vowel system of
Korean. The vowels that do occur fit into a 2X (2+2+1) pattern, symmetrical
for front and back, and, except for low, for unrounded and rounded. But this
would imply ten vowels; the "high front rounded" spot is actually, for most
speakers, simply not filled. Another example occurs among the obstruents of
Bella Coola. Plain and glottalized stops and affricates pair off very neatly at
various positions of articulation, and the dorsals and the tongue-front affricates
 :

142 A MANUAL OF PHONOLOGY

are matched by spirants. However, among the laterals one finds only /!/ and the
glottalized stop /X'/; there simply is no plain stop /X/.
In the two instances just cited we are quite clearly dealing with a symmetric
pattern where one position implied by the pattern is simply not occupied. This

isa very different matter from an asymmetric pattern with all positions occupied,
though it is not always easy to decide which situation confronts one in a given
case. However, it is perfectly obvious that a three- vowel system /i a u/ is no
example of a four- vowel symmetric system with one position not occupied;
similarly an obstruent set-up with /p t k/ and /s/ is not a symmetric six-point
pattern of three positions and two manners, with two holes. Either an asymmetric
pattern or a symmetric pattern with a hole represents a degree of asymmetry,
but of two somewhat different kinds.
273. Frequency Aspects. There is another point of view from which both the
problem of balance and that of symmetry have to be examined. Up to this point
we have been investigating these matters relative simply to the stock of phonemes
(or other phonologic units) which occur in a language. But there is also the mat-
ter of relative frequency of the various units, or various kinds of units, in the
actual stream of speech (text frequency).
It takes much more work to reach quantified conclusions along this line, for
none can be obtained without extensive counting in actual texts. The percentage
of vowels in the stock of phonemes of a language is one matter. The relative text
frequency of vowels is a very different matter. In Hawaiian, where there are
more different consonant phonemes than vowel phonemes, the latter occur some-

what more frequently than do the former so much requires no counting, since a
syllable always contains a vowel but only sometimes begins with a consonant.
In the same way it can be proved without a count that, in Yawelmani, conso-
nants must be more frequent than vowels, since every syllable contains at least
one of each (as onset and peak) but some syllables also have a coda consonant;
in Bella Coola the ratio must be in the same direction and even larger.
Relative frequency bears on the problem of symmetry in an interesting way.
As a point of departure, let us consider the set of English spirants, /f v 6 <5 s z s &/,
which manifest symmetry in that /{/ is to /v/ as /&/ is to /<5/, and so on. The
relative text-frequencies of these eight consonants are as follows
/s/ .279 III .113
// .210 // .050
/z/ .182 /$/ .023
hi .140 III .003.
That a phoneme chosen at random from a long text is a spirant, then it
is, if

will be /s/ with probability .279, and so on. It will be noticed, first of all, that
while /s/ has a higher relative frequency than /z/, /{/ than /v/, and // than
hi, for the remaining pair the more frequent is the voiced member. Thus, al-
though without regard to frequency we can assert that /s/ is to /z/ and /0/ is

to /o"/, if frequency is considered this is not the case.

Another interesting point stems from the extremely low relative frequency of
hi. One could make a count in a very long English text and stand a reasonable
 :

A MANUAL OF PHONOLOGY 143

chance of finding no occurrence of /&/ at all. The difference between relative

frequencies of .003 and of .000
is very slight. Yet just this is the difference be-

tween a symmetric pattern with one very low-frequency member and a sym-
metric pattern with a "hole." Now suppose that we had frequency figures for
Taos obstruents, as indicated below (the figures given below are actually com-
pletely faked).We could describe the Taos system by listing twenty units, as
follows
b .005 d .013 .000 .018 w .000
3 g g
p .055 t .032 c .068 k .096 kw .236
p' .177 t' .060 c' .024 k' .039 k' w .002
p' .001 t' .008 s .076 x .047 xw .043
The two units to which we have
assigned frequency .000 are the two which, so
far as our reports show, do not occur in Taos at all. Yet this description perhaps
portrays the Taos obstruent system quite as accurately as would a description
which omitted /$/ and /g w / from the table altogether.
Working purely from the stock of phonemes, then, a hole in a pattern appears
as an asymmetry. Taking relative frequency into consideration, such a hole may
be far an asymmetry than is some extreme deviation of the set of relative
less of
frequencies from some "most normal" distribution. For example, we would
probably say that a situation in which one of a set of eight contrasting units
accounted for eighty percent of the occurrence of any of the eight, while the
remaining seven accounted about equally for the remaining twenty percent, was
far more "skew" than the English spirants actually are, or than the Taos obstru-
ents including a zero-frequency /$/ and /g w / would be.

3. The Principles of Phonologic Analysis. In the taxonomic survey of 2

factual statements on this or that language were presented with a minimum of
theoretic or methodologic comment. What justifies the claim that English has
nine vowels? That Nootka has thirty-seven consonants? That glottalization is a
positive "mark" in Nootka, but a negative device (equivalent to absence of the
positive mark of aspiration) in Chitimacha? How does the analyst find out?
Why, in some cases, should an analyst hesitate between alternative interpreta-
tions rather than making a forthright assertion? These are the sorts of questions
which shall concern us here.
The process of phonologic analysis has two aspects: gathering and collation.
We call these "aspects," rather than "steps," because it is not possible to do

pure gathering first, pure collation afterwards the two cannot altogether be
separated, but they can be distinguished.
Analysis is undertaken under a wide variety of circumstances. At one extreme,
an investigator does fieldwork of the observation-participation type in a com-
munity where a language is spoken. At the other extreme, there are no speakers,
or they are unavailable, and one is forced to work via secondary sources, mainly
written records. For linguistics, any sort of written record is a secondary source;
the philologic and historical classification of some documents as "primary" and
others as "secondary" is achieved from a different slant.
144 A MANUAL OF PHONOLOGY

The fundamental assumptions are unchanged in these various circumstances,

but the details differ a great deal. When one works from secondary sources be
they inscriptions several thousand years old or field reports from a contemporary

colleague some of the collation has been done, and cannot always be undone
in order to try a different approach. Of course all the essential steps of gathering
have been done in these cases, and although internal inconsistencies may lead
the new investigator to question the accuracy of the gathering, he cannot do it
over again unless he himself can work directly with native speakers. From the
poorest of field-reports or surviving documentary evidence, sometimes virtually
nothing can be deduced with any certainty. From the best, one can get some-
thing close to the whole story.
Usually the situation lies between these extremes: one can see fairly well the
limits within which accuracy of reinterpretation is possible, though these limits
may not be as narrow as one would wish. For example, from Boas's reports on
Kwakiutl we can be sure that there is either a three- vowel system /i a u/ or a
five-vowel system /ieao
u/, with contrast of length not, say, some other sort
of three- or four-vowel system or a system with eight or nine vowels. But one
cannot with certainty choose between the three- vowel and the five- vowel system,
and one cannot be absolutely sure whether there is, or is not, also an indeter-
minate /a/. Sufficiently detailed examination of Boas's material can lead one to
favor one or another of the alternatives, but only independent field work would
settle the issue.
We shall now discuss gathering and collation in as separate a fashion as is
realistically possible. In our discussion of gathering we shall assume that the
circumstances are those first mentioned above: an investigator working directly
with an informant. Most of the principles which apply under these circumstances
are also as operative as conditions will allow when, at the opposite extreme, one's
data are written records of forms of speech no longer current. Under the latter
conditions there are also many special techniques, such as the various methods
for dating manuscripts, the study of the development of writing-systems,and
the use of information about earlier or more recent stages of the same language,
on which we shall offer no comment. We refrain not because the special problems
are not challenging and important, but because they would take us too far afield.
In our discussion of collation we shall assume that the analyst has complete and
accurate data something which in the general caseis true only after a great

percentage of the process of collation has already been tentatively performed.

31. Gathering. The fundamental assumption in phonology is that, despite the

wide variation of actual articulatory motion from one speech-event to another in
a community, the speakers of a language themselves hear some speech-events as
the same, some as different. To put this in a slightly different way: some speech
events, under ideal hearing conditions (no background noise), sound the same to
the native speaker, regardless of any variations in actual articulatory motion
which the analyst may be able to observe, while others sound different. If two
forms regularly sound the same (barring distortion by background noise) to the

native speaker say meet and meat in English then when these forms occur as
 A MANUAL OF PHONOLOGY 145

whole utterances a speaker cannot tell, from the acoustic evidence alone, which

of the two is being said. It

only rarely that one has to distinguish between two
is

forms purely on the basis of the acoustic evidence; usually there is context. Thus,
if A asks What B answers meat, A knows that B has
did you have for dinner? and
said meat rather than meet, not from the acoustic evidence, but because one does
not have meet for dinner. At the same time, A knows that what B said was meat,
rather than potatoes or pork or company, from the acoustic evidence and nothing
else (and therefore, under poor hearing conditions, he may not be able to make
the discrimination). Phonologic analysis, then, is concerned with the way in

which utterances, under ideal hearing conditions, are kept apart by virtue of
acoustic evidence and it alone; any other observations which an analyst makes are
relevant only insofar as they contribute to this end.
To the field situation, the informant brings native control of his language,
including the whole system of what "sounds same" and what "sounds different"
to him. The investigator brings several things: (1) his understanding of the funda-
mental assumption, including an understanding of what it means for items to
"sound same" or "sound different" to a native speaker; (2) some training in
general phonetics, by which he will be able to tell, at least in part, what articu-
latory motions are probably responsible for any particular bit of speech-sound
he elicits from his informant, and in terms of which he can put down in a written
notation something which represents more or less accurately each elicited speech-
event; (3) most important, his ability, as a human being, to be enculturated, to
learn a new language, to empathize.
The basic problem in phonologic field work is to bring it about that there
a single nervous system, both the native or native-like active control
reside, in
of the phonologic system, and the clear understanding of what is meant by
"sounding same" and "sounding different." For, when this has been accom-
plished, becomes possible to tabulate the various articulatory motions which
it

differentiate utterances that sound different, ignore differences of articulation

which do not make any such difference, and thus achieve a description of the
phonologic system. Until the basic problem is solved, this is quite impossible:
the investigator cannot be sure whether two items that sound the same to him
actually sound the same to the informant, and cannot be sure that clear differ-
ences to his own ears have any relevance for the informant.
Obviously there are two ways, in theory, in which the basic problem can be
solved: make the informant into an analyst, or make the analyst learn some of
the language. It is a happy but rare event when the first of these alternatives is

possible; usually the informant's concern with his own language is too purely
instrumental and too subordinate to other interests. What would happen, for

example, if one simply asked an English-speaking informant whether meat and

meet sound the same or not? Experience shows that the answer is often wrong,
not through any dishonesty or inability on the part of the informant, but because
he does not understand the question. Meat and meet have different meanings; they
occur in different contexts of other words; they are spelled differently (a consider-
ation which is of course relevant only for some languages). It is very easy for an
146 A MANUAL OF PHONOLOGY

informant to say, and perhaps honestly to believe, that the two sound different
because of these various other differences. I have seen in print the assertion that
grey and gray "ought" to be pronounced differently.
This does not mean that one cannot address such questions to an informant,
only that the responses require careful interpretation. The analyst is apt to have
his own tentative opinion in any such matter he thinks he hears a difference, or
:

thinks that he does not. If the informant's answers in general agree with the
analyst's opinion, then one can be fairly sure that the informant does under-
stand the question as it is meant and is answering correctly. And in case of dis-
agreement there is a simple test which can be performed, providing the psycho-

logic factors in the field situation do not render it inadvisable.

For the test one needs two informants, A and B. A has claimed, let us say,
that meat and meet sound different. It is produce ten or
arranged that B will

twelve utterances, each one of them meat or meet, in a random but prearranged
order unknown to A. A is then asked to tell in each case which form he hears.
If he can indeed tell in a significantly large proportion of the cases, then the
investigator was wrong; he cannot, then the investigator was right. Unless the
if

informant's pride is by outcomes contrary to his claim, two or three tests

injured
of this kind can render him more effective as an informant, by showing him more
clearly what the analyst means by "sounding same" and "sounding different."
And the analyst, who of course has no pride to be injured, is aided because he
discovers either that there is some subtle distinction which he has missed and
which he must master, or else that he was indeed on the right track.
The converse of the meet-meat situation will also be encountered for example, :

some speakers of English pronounce the word root sometimes with /uw/, some-
times with /u/. The analyst might not know, when he first noticed this fluctua-
tion, whether it was a case of free variation within the bounds
phono- of a single
logic unit, or a case of free alternation, for the particular form, between two
distinct phonologic shapes. Perhaps ultimately the meanings of the two forms,
root with /uw/ and root with /u/, are slightly different, but the degree of pre-
cision about meanings that can be achieved early in field work is not great enough
to catch any such subtle difference in connotation if indeed there is so
any
that there would be no help from semantics. However, in due time the investi-
gator would discover that /wuwd/ and /wud/ (wooed, wood-would), /kuwd/ and
/kud/ and a number of other pairs of obviously different forms,
(cooed, could),
are kept apart precisely by the difference, /uw/ versus /u/, which appears in the
two different pronunciations of root; and thereupon the answer would be clear:
way we call free alternation, not subphonemically.
root varies in the
The English examples which we have used above seem trivial, but this is
because, knowing English, we know the answers. It must be remembered that
problems of these kinds are encountered in the field when one is not yet thor-
oughly at home with the phonologic system which is being studied, in which
circumstances examples of the kinds given are by no means trivial.
Whatever help the analyst is by questioning his informant or using
able to get
tests on him, sooner or later it is almost always necessary that the analyst acquire
 A MANUAL OF PHONOLOGY 147

the ability to make

for himself judgments of "sounding same" and "sounding
different," not on any basis of general phonetics but in a way which matches
closely the identifications and non-identifications which the native informant
in
fact makes. That is, the analyst must empathize, must to some extent learn the
language with which he is working. We know of no set of procedures by which a
Martian, or a machine, could analyze a phonologic systeman entity, that is,
to which even the basic biologic
and cultural common denominator of humanness
would be alien and would require specification. The only procedures which can
be described are rules for a human and depend essentially on his
investigator,
ability to empathize. To refuse to do on avoiding it as much as
this, to insist
possible in the name of a spurious "objectivity"as, for example, Franz Boas

did is to place oneself in much the same position as that of a bacteriologist who
would refuse to stain his slides. Obviously there are numerous pitfalls in the path
of the investigator who seeks to enculturate to the phonology of a new language,
but equally obviously we have reason to believe that the task can be accom-
plished. Any newcomer to a community, say a newborn child, in due time learns
the language of the community, including its phonologic system: learns, that is,
to identify and differentiate acoustic stimuli in much the same way as do those
already in the community. The matching may not be exact, but apparently ap-
proximate matching is good enough, since the members of a speech community
do in fact communicate via speech. All that we can ask of an investigator is that

he do as good and precise a job along this line as a child does but that he keep a
record as he does it, so that we will have the materials to which to apply the
other aspect of analysis, collation.

The fundamental assumption of phonology is really just that an assumption,
not capable of any kind of proof save the way in which it serves to coordinate
and put sense into a wide variety of observations. Its logical status is much like
that of, say, the particle theory in physics: particles are not observable in any
direct way, but assuming them ties together a large number of matters which are
directly observable. No one has yet proposed any alternative assumption which
gives us a basis on which to work in phonology: without the fundamental hypoth-
esis as we have stated it, the analyst is completely lost. No linguist ever ap-
proaches a new language with the intention of finding out whether that language
has a phonologic pattern or not: its existence is necessarily taken for granted,
and the analyst's aim is entirety one of discovering the specific details.
There is a second assumption involved in phonology which is closely related
to the first but not deducible from it. This is the assumption that every utterance
in some single language is composed wholly of an ordered arrangement of phono-
logic elements drawn from some strictly finite stock, which are the only ones of
relevance for the language in question. If the total stock of actual and possible
utterances in a language were finite, this second point would not be an assump-
tion but would be a theorem provable on the basis of the fundamental assump-
tion. For what one would do would be to compare the whole stock of utterances
in pairs, listing the shared phonetic features and those which differentiate utter-
ances, and continuing this until all possible pairs of utterances had been covered.
148 A MANUAL OF PHONOLOGY

What the analyst actually does resembles this save for one point: the analyst
can never observe more than a sample of the whole stock of actual and possible
utterances, since in every language the whole stock is transfinite in number. As
the investigator slowly enlarges the sample which he examines, the rate at which
new contrasts are revealed steadily decreases; eventually he is forced to stop
work and to predict that nothing new is going to turn up no matter how many
additional observations are made. The steady decrease in new contrasts affords
empiric support though not logical proof for this subordinate hypothesis.
Practically, what is usually assumed is that once one has tabulated everything
which can differentiate macrosegments, one has tabulated everything. This
implies that the total stock of actual or possible utterances of not greater than
some maximum length is indeed finite, and that the only reason the whole stock
of actual and possible utterances, regardless of length, is transfinite is because
there is no arbitrary limit to the length of an individual utterance; it also implies
that longer utterances will always consist of an integral number of macroseg-
ments.
The procedure of the analyst in the field reflects all of these primary and
secondary assumptions. One begins almost always by eliciting quite brief utter-
ances, of approximately "word" length, for the practical reason that the in-
vestigator can hardly hope to hear with accuracy any much longer stretch of
had practice. Indeed, in some cases (perhaps not all) a
alien speech until he has
new language sounds to an investigator for the first few hours or days like a
completely confused and continuous flow of sound, entirely free of any segmenta-
tion. This fuzziness disappears in time, not because the investigator is trained
especially in analytic techniques but simply because he is a human being; the
work until the fuzziness has gone
analytic techniques cannot even be put to
away, and by that time some "problems" are already solved, without any
conscious effort.
As work proceeds, one does not delay too long over indeterminate points. If
one cannot decide whether a particular pair of forms sound "the same" or not,
usually the best practical course is to pretend for the nonce that they do further ;

hearing of the language will often settle the matter, and one can always go back
through one's notes and enter corrections.
Subject to such temporary decisions which may later be altered, one constantly
tabulates what is turning up: vowels under stress (assuming that one hears a
stress), consonantism between beginning of utterance and first vowel, consonant-
ism between successive vowels, and so on. It is always possible to reach conclu-
sions too soon. In particular, although sufficiently long utterances are assumed
never to involve contrasting phonologic material which does not turn up in
shorter ones, there are often features which are of relevance but which cannot
appear in the very short "word"-like utterances with which one begins: junctural
features and the like, and many intonational contrasts. This is a point on which
we are more careful today than we were, by and large, ten or fifteen years ago:
the older "Wortphonologie" constantly missed subtleties of some importance, or
else brought them belatedly into the picture in a lopsided and disorganized way.
 A MANUAL OF PHONOLOGY 149

If it were possible to separate gathering and collation, we would say that

gathering is completed when one is certain that new contrasts will no longer

turn up, and when one has collected a large number of utterances, short and long,
in an allophonically correct transcription. An allophone is a "target," as it were,
at which a speaker aims as he produces an utterance: the /p/ of spill is an al-
lophone, the /p/ of pill a different allophone, the /p/ of stop still a third. A
transcription is allophonically correct if it differentiates, at each position in an
utterance, between the allophones which can contrast in that position. Thus
all

such a transcription for English would have to differentiate between /p/ and /t/
and certain other possibilities in the position after initial /s/ before a stressed
vowel, but would not have to distinguish between the /p/ of spill and that of
it

stop, because thosefall in different environments; and it would not necessarily

have to use the same symbol for, or in any other way imply some kind of identity
between, the /p/ of spill and that of pill or stop, though in this case it probably
would. Another kind of leeway in allophonically correct transcriptions has to do
with the size of the unit taken as a single allophone. One might use unit symbols,
rather than sequences of two symbols, for the syllable-peaks of English beet, bait,

bite or even unit symbols for beet and boot, bait and boat, but sequences of two
for the peaks of bite and bout. It would not even matter, in fact, if one used a
unit symbol for the /br/ of bread, so long as the items kept apart in one way or
another matched those actually kept apart by speakers of the language, and so
long as a reasonably accurate phonetic description of each "unit" symbolized in
the transcription were at hand.
In other words, gathering is completed when all relevant distinctions are known
and regularly indicated; the business of identifying elements in different environ-
ments as in some sense different occurrences of "the same" element is the task
of collation. In practice, collation begins long before gathering is completed, but
any identifications made before the completion of gathering are subject to re-
vision. The total process may involve going back and forth between gathering
and collation, but in the end the final results of collation are based on presum-
ably complete gathering.

32. Collation. The purpose is to determine what elements occur in

of collation
distinctive contrast with each other, and the hierarchic arrangements in which
those elements occur. Since the assumed point of departure for collation is an
allophonically correct transcription (together, of course, with description of what
each symbol in the transcription means), collation itself takes the shape of a
readjustment of size of the units tentatively recognized in the allophonic tran-
scription, and a readjustment in the identifications of units in different environ-
ments. The procedure is one of trial and error, of successive approximations,
until finally a satisfactory portrayal is achieved.
The matter which we shall discuss in connection with collation is the
first

principles of IC-analysis. After our introductory discussion of this, we shall

describe and illustrate some of the heuristic principles which were evolved and
used before the hierarchic approach had been consciously introduced into
phonologic work, and which are still by no means to be discarded. Then we shall
 :

150 A MANUAL OF PHONOLOGY

show how certain recurrent types of problems which gave trouble under other

approaches are often resolved sometimes in surprising ways by the hierarchic
approach. Then we shall discuss the analysis of junctural phenomena, which
require different procedures from those used in the bulk of phonologic analysis.
Finally, we shall deal with the interrelationships of ultimate phonologic com-
ponents.
321. The Principles of IC- Analysis. The questions that one must ask and seek
to answer in IC-analysis can be posed if we assume that we are confronted by a

recurrent or potentially recurrent unit of some kind a macrosegment, a micro-

segment, a syllable-peak, or the like (a constitute) the unity which can be
of
assumed, but which is to be broken into its ICs. If the unit in question is an
ultimate phonologic constituent, then, of course, there is no problem of further
breakdown. If it is a unit of only two ultimate phonologic constituents, then the
breakdown is obvious. In any more complex case, one has to try various possi-
bilities, comparing both the whole unit and the tentative ICs with what can be

obtained from other comparable whole units. The questions that one asks and
seeks to answer are as follows
(1) How many ICs are there?
(2) Which portions of the unit belong to each IC?
(3) What is the physical arrangement of the ICs in the whole?
(4) In what construction do the ICs stand?
(5) Where else does each IC occur? What other elements have "privileges of
occurrence" comparable to those of each IC of the given unit?
These questions cannot be discussed one by one, for they are interrelated in
intricate ways. What we shall do is to discuss a number of examples (all in Eng-
lish), to show the possible answers which can be obtained, and the criteria on

which one bases these answers.

(1) Consider first the English syllable scrimp. Phonetically, this syllable con-
tains an onset, a peak, and a coda, respectively /skr/, /i/, and /mp/. Here,
and generally, it is wise to follow the phonetic evidence unless there is very
convincing reason to do otherwise, and in the present case there is not. So we
assume that /skr/, /i/, and /mp/ (rather than, for example, /ri/ or /im/ or the
like) are all units on some level of structure. But this does not answer the first
question of those listed earlier. The ICs of the syllable may be these three smaller
units; but it may
also be that two of them belong together as one of two ICs
of the whole. Examining English syllables in general, we find that while a great
variety of microsegment-initial syllables occur with a peak not preceded by an
onset, only certain special types of microsegment-final syllables have a peak not
followed by a coda. The bondage of peak and coda seems to be slightly stronger
than that between onset and peak. This may lead us to say that the ICs of
/skrimp/ are the onset /skr/ and the peak-coda sequence /imp/, leaving the
subdivision of the latter to a later step. The fact that peaks (or peak-coda
sequences) occur without onsets, but not vice versa, leads us to say that the
construction of the two ICs is subordinate, with the peak-coda constituent as
nucleus, the onset as satellite. As for question (4), the answer is in this case quite


A MANUAL OF PHONOLOGY 151

obvious: the subordinate onset /skr/ precedes the nuclear peak-coda combination
/imp/. And question (5) /skr/ recurs as onset in other syllables (scratch, scrape),
:

as an interlude (describe, escrow),

and is comparable to other units which likewise
have these and only these privileges of occurrence, e.g. /spl/ (splash, display),
/skw/ (squelch, esquire), or /kr/ (crate, acrid). A unit such as /sk/ is not com-
parable, since it occurs also as a coda (ask). The other IC, /imp/, recurs and is

matched by comparable units in much the same way.

(2) Next let us examine /imp/, extracted as one IC of /skrimp/, but also
occurrent as a whole syllable. We have already seen that this consists phonetically
of peak /i/ and coda /mp/, a guide to formal analysis which we will not reject
unless forced to. But in this case formal considerations reinforce the phonetic
guide. If we were to divide into /im/ and /p/, and to divide other sequences in a

comparable way say ant into /sen/ and /t/, (f)ond into /an/ and /d/, and so

on we would find ourselves with two types of units of little phonetic homo-
geneity and capable of occurring together in sequence only in a rather random
manner, /im/, for example, would occur before /p/, but not before /t/ or /d/;
/p/ would occur after /im/ but not after /sen/ or /an/. There is no possible
division which will yield complete freedom of combination of the units obtained,
but we can seek something as systematic as possible. The phonetically most
realistic division, into /i/ and /mp/, /ae/ and /nt/, /a/ and /nd/, yields types

of units which are at least freer in their combinations than the alternative pro-
posed above: witness imp, ramp, romp, hint, ant, aunt (for some speakers),
sinned, manned, fond. Furthermore, this way of cutting the whole yields units,
like /mp/, of phonetic homogeneity: /m/ and /p/ are both bilabial, and in most
(not all) nasal-plus-stop sequences in English the two are in the same position
of articulation.
Since syllables do occur with a peak followed neither by coda nor interlude,
we shall say that in /imp/ the second constituent, /mp/, is satellite to the
first, /i/.

(3) Next let us consider a macrosegment, say

2 31
I don't want to j

The ICs of this are the intonation and the remainder. This is phonetically the
simplest analysis, and is supported structurally by the fact that every intonation
occurs with a large variety of remainders, while many a sequence of "words"
can be "intoned" in two or more ways. The physical arrangement of the two ICs
isdifferent from that in previous examples: the two occur simultaneously. This is
not a sufficiently specific statement of the arrangement, however, since the same
intonation and the same remainder might be joined together with the center at a
different point
I n don
J
2
t want to J,

The specification of arrangement, therefore, has to include indication as to where

the center is. Here we have our first example but not our last of two ICs
which can participate more) arrangements, the
in a constitute in either of two (or

resulting constitutes being distinct and in contrast with each other.

Since every macrosegment includes both an intonation and a remainder, the
152 A MANUAL OF PHONOLOGY

construction in which the two stand is presumably coordinate rather than sub-
ordinate. There is, in English, one phenomenon which suggests otherwise: we
have an indefinite number of macrosegments of the type of
i3
hm |
31
mm I
3l
mm T
where the phonemic structure (vowels and consonants) is aberrent: in the
writer's speech, the segmental "carrier" of the intonation is necessarily nasal
and voiced, but the contour of the oral cavity is irrelevant. The intonational
parts of these macrosegments, however, are like the intonations of any other
macrosegment. This might lead us to say that in macrosegments in general, the
remainder is a satellite to the intonation as nucleus.
(4) Take next the microsegment believer (as in, say, a macrosegment He's a
believer in modern ways). Phonemically, this is /biliyvar/, with (nonphonemic)
loud stress on the second syllable. Since loud stress is isolable, while weak stress
is not, we conclude that the first and third syllables are satellites in some sub-
ordinate construction. It might still be that the first and second syllables should
be taken as one IC of the whole, the third syllable as the other, or else the other
way around. But weak-stressed syllables preceding a loud-stressed syllable in a
microsegment are just as common as weak-stressed syllables following the strong
stress; there seems to be no basis for a bipartite rather than a tripartite cut.
So we recognize three ICs, the first and third each attributive to the second.
(5) In believe we have an interlude /l/ between two peaks. The first syllable
is one IC of the whole, the second syllable the other, and the first is attributive

to the second. But does the /l/ belong to the first IC or to the second (question
2)? The answer is that it belongs to both at the same time: it is a shared constitu-
ent or pivot. The breakdown can be indicated as follows: First, into /bi(l)/ and
/(l)iyv/, where the parentheses indicate that there is just one /l/ but that it
has to be assigned to both ICs. Second, /bi(l)/ into attributive /b/ and nucleus
/i(l)/; /(l)iyv/ into attributive /(l)/ and nucleus /iyv/. Third, /i(l)/ into /i/
and /(l)/; /iyv/ into /iy/ and /v/; in both cases the second constituent is attribu-
tive to the first.

(6) The opposite of a pivot turns up in the form slyness, which is two micro-
segments, each a whole syllable. The first constituent is the whole syllable sly,
and the second (subordinate) constituent is the whole syllable -ness. This leaves
unassigned the juncture which separates the two constituents. Instead of arbi-
trarily assigning it to either IC, we regard it as standing apart from both, as a
marker of the point of contact of the two the physical arrangement
; is sequential
with a gap.
(7) In such forms as cats, cads, docks, dogs, nips, nibs, apt, act, ebbed, nagged
we find codas /ts dz ks gz ps bz pt kt bd gd/. Since in each case the coda as a
whole is voiceless or voiced, it is useful to take as the ICs of the codas not that
obtained by a vertical cut, e.g. /ts/ into /t/ and
but rather a voice-irrelevant
/s/,
sequence of oral articulations on the one hand, and a voicing-component on the
other. Representing the voice-irrelevant constituents with capital letters, voice-
lessness by /H/, and voicing by /A/, we thus break /ts/ into /TS/ and /H/,
 A MANUAL OF PHONOLOGY 153

/dz/ into /TS/ and /A/, and so on. The arrangement of the constituents, as in
the case of intonation and remainder, is simultaneous, but unlike the intonational
situation, there are no further The construction in which
specifications necessary.
/TS/ and /H/ occur in coda /ts/
presumably coordinate, since the contrast
is

between voicing and voicelessness is meaningless in English except in such a

context, and the voice-irrelevant constituents occur only with one or the other
of the two voicing-components. In the case of the coda of width (or widths or
midst), we first divide /d0/ (/d0s/, /dst/) into /d/ and /Bf (/d/ and /0s/ or /st/),
and then subdivide the constituents accordingly. In the case of the coda of
twelfths, or any other coda, interlude, or onset which has both sonorants and
obstruents in it, we similarly separate the sonorants from the obstruents (/lf0s/
into /l/ and /ids/), and then proceed with the indicated analysis of the obstru-
ents, the further analysis of the sonorants being another matter.
Now let us summarize the criteria which have emerged in the examples.
(1) We base
our cuts between ICs on the phonetic facts except where there is
overwhelming reason (stemming from other criteria) not to. No example has
appeared in which there was such reason; the fact that we divide imp into /i/
and /mp/, but Ike into /ay/ and /k/ might seem inconsistent from the phonetic
angle, but structural evidence supports the difference of treatment (see 241).
(2) When confronted with alternative possible cuts into ICs, we prefer that
cut which yields elements of greater internal cohesion (/[/ and /mp/ rather than
/im/ and /p/, because the common bilabial character of /m/ and /p/ consti-
tutes "greater internal cohesion"), and with greater freedom of recombination
(same example, or our treatment of codas /ts dz ks gz/ etc.). We try to cut in
such a way that comparable limitations on recombination are dealt with on the
same level. Since codas involving obstruents in general contain only voiced or
only voiceless obstruents, it seems that the voicing or voicelessness is a feature

of the whole sequence of one or more obstruents, rather than of each obstruent
individually; we cut accordingly.
(3) A construction of two ICs is subordinate if one of the ICs is of a distribu-
tional class the members of which can appear without aiy member of the class
of the other IC; it is coordinate if each requires the other, or, theoretically, if

either can occur without the other.

To the five questions with whichwe began, clearly no general answers are
possible: the answers depend on the details in each language. Yet we may note
the range of possible answers to some of the questions. Question (1) is How many
ICs are there? The answer is often two, sometimes three, and perhaps sometimes
more than three. Question (2) is Which portions of the whole belong to each IC?
We have seen, at least, that a cut may fall neatly between two contiguous but
non-overlapping portions, or that there may be an overlap or pivot, or that there
may be a gap and marker. Question (4) is In what construction do the ICs stand?
For a bipartite form there are two answers: coordinate and subordinate. For a
tripartite form the only possibility we found was one with a nucleus and two
independent satellites, but there may well be other possibilities. Question (5)
requires more expanded discussion, which will be given in the next section (322).
There remains question (3) What is the physical arrangement of the ICs in
:
154 A MANUAL OF PHONOLOGY

the whole? We have seen instances of sequential arrangement, in immediate

juxtaposition, or with a pivot, or with a gap; and we have seen instances of
simultaneous arrangement, with or without necessary additional specification as
A major point must be made here. It seems quite impossible,
to relative timing.
by any reasonable methodology, to treat the phonologic structure of an utterance
(or of a macrosegment, or of any smaller unit) in the general case as merely a
combination of elements in the mathematical sense; what we have, rather, is a
'permutation. That is: our working assumption is that any unit save an ultimate
phonologic constituent, consists of smaller elements (ICs) in a specific geometric
arrangement, and the matter of arrangement is as important as is the matter of
the constituents themselves. Now there are of course exceptions. Thus, in Fijian,
if one knows that a given consonant and a given vowel are to participate in a
syllable, then there is no further choice of arrangement: the consonant must come
first, as onset, and the vowel second, as peak. A similar, though more complex,
predictability of arrangement applies for the constituents of any syllable in
Japanese. But these are exceptions, not the rule. The intonation /2 31|/ and
the "words" / don't want to can combine in at least two contrasting arrangements.
The /SK/ of ask consists of /S/ and /K/ in one linear order; in ax we have the
same constituents in the reverse order.
There are certain matters of arrangement which are predictable, and hence
not independently important, under the IC approach, but not predictable when
one deals with whole utterances merely as intertwining strings of ultimate
phonologic constituents. Thus in Fijian the units /k/, /a/, and /a/ can be com-
bined into either of two disyllabic sequences: /kaa/ or /aka/; but if we follow
the IC approach this indeterminacy and independence appears only for the whole
syllables /ka/ and /a/ as units to be combined, and vanishes when we go down
the size-scale to attack the structure of the individual syllables. Yet, as this
example itself shows, the IC approach does not eliminate arrangement as an
independent factor: it merely assigns contrasts of arrangement to one size-level
or another.
322. Older Heuristic Principles. The fifth of the questions posed at the begin-
ning of 321 implies that one has to make identifications of elements found in
different environments: e.g., one says that the /skr/ of scrimp, of scratch, and of
describe is "the same" element in three different environments. Many such
identifications are obvious; many non -identifications are equally obvious. How-
ever, many languages seem to present also a residue of elements where the
problem of identification requires more care.
As an example of a trivial identification, we may consider such English forms
as pep, tat, kick, bib, dead, gag; one concludes immediately that English has
elements /p t k b d g/ which occur both as onsets and as codas. The /p/ at the
beginning of pep is "the same" element as the /p/ at the end of pep, and if there
is any difference between them that difference is not part of the recurrent ele-

ment, but part of the two differing environments. Confronted with pill, bill,
sill, spill, it is likewise obvious that the /sp/ of spill is a more complex onset

than the others, but that it contains, in one way or another, the same unit which
 A MANUAL OF PHONOLOGY 155

occurs as whole onset in sill, and that it also contains a unit related somehow to
the /p/ of pill or to the /b/ of bill or to boththough the precise nature of this
relationship is not so clear.
Many theoretically conceivable identifications are never considered. In many
a language one will an onset [k] and a peak [a] in complementation
find, say,
(neither occurring in any environment where the other is found), but no one in
his right mind would suggest that these two are "the same" unit in two different
environments. The reason is that collation is not pure mathematics: it is an
analytic process applied to empiric data of a certain kind, and worthwhile results
are achieved only if we insist on hugging the phonetic ground closely.
Between the two obvious extremes just illustrated there are less trivial prob-
lems of identification. In German one gets contrasting onsets /p t k b d g/, but
only something like [p t k] as codas. In this case what would it mean to say that
those units occurrent as codas were "the same" as initial /p t k/, or, alternatively,
"the same" as initial /b d g/? Could it make sense to deny both identifications or,
in some way, to make both of them at once? The bilabial stop of English spill
isnot exactly like either the /p/ of pill or the /b/ of bill; we can ask the same
questions in this case. Shall we count the onset of chick as simple or as complex,
or does it matter?
Four have been proposed as guides in solving such prob-
heuristic principles
lems; they are of unequal value, but all merit some discussion. One is the principle

of contrast and complementation, one that of phonetic realism or phonetic similarity

and dissimilarity; one that of pattern congruity; one that of economy.
3221. Contrast and Complementation. Segments which occur in identical en-
vironments are necessarily phonologically different: e.g., the /p/ and /b/ of
English pit and bit. We say "phonologically different" rather than "different
phonemes," because the contrasting segments may turn out to be either phonemes
or phonemic sequences of some sort. Thus the pair blown and groan proves the
phonologic difference of /bl/ and /gr/, both of which prove to be sequences of
two phonemes. Even the /p/ and /b/ of pit and bit are phonologically complex,
not simple: whether we call them "phonemes" or not depends in some measure
on just how we choose to use that term (see 3231).
It is, of course, not implied that segments which are not in contrast which are
in complementation or complementary distribution in that one of them never occurs
in any environment in
which the other occurs are necessarily phonologically
identical. We may hypothetic example of an onset [k] and a peak [a]
recall the
in complementation as a case in point. Lack of contrast is a necessary but not a
sufficient condition for phonemic identity.
In making use of this heuristic principle, it is necessary to work in terms of
as much of the environment as may be relevant, and no account can be taken
of IC structure. In Fijian, [ia] and [ya] are distinct; therefore [i] and [y] are
phonologically different. It is quite irrelevant that [ia] is two syllables and [ya]
but one. But this proof of the distinctness of Fijian [i] and [y] is not yet sufficient
to define the phonologic status of the two elements. If it were to prove convenient
to recognize a syllable juncture in the language (which it does not), then the
156 A MANUAL OF PHONOLOGY

phonologic distinction between [i] and [y] could be ascribed to the juncture: [ia]

would be /i-a/, and [ya] would be /ia/. Or if it should develop that there is an
independently distinctive stress in the language, so that our original phonetic
notations had to be amended to [ia] and [ya], then the distinction which the pair
shows could be handled in terms of the stress: /ia/ versus /ia/ (this, also, does
not turn out to be the case). The mere phonetic fact that [ia] is two syllables
and [ya] one is not in itself sufficient to justify either of these two conclusions
or any other. The pair proves the phonologic distinctness of [i] and [y] a great ;

deal more collation is necessary before we can decide realistically how to provide
for the contrast. Nothing is commoner in the literature than conclusion-jumping
along this line.

Another type of problem which is apt to arise, that presented by multiple

complementation, will be dealt with below in 3232.
3222. Phonetic Realism. Segments which are in complementation are never-
theless phonetically distinct unless they resemble each other phonetically in such
a way that the differencies can be extracted from the segments themselves and
assigned, instead, to the environments, in some phonetically realistic way.
This principle bars the possible identification of onset [k] and peak [a] (or, at

least, virtually all investigators feel that it does), but would certainly often allow
an identification of onset (or coda, or complex -peak satellite) [w] and peak [u].
An example found in many languages involves voiceless stops [p t k], initially
and finally, in complementation with voiced stops [b d g] medially between vowels
or other regularly voiced sounds. There are two lines of approach which can be
taken they in general yield the same ultimate results, but their logic is somewhat
;

different. One is the method of allophones, the other the method of redrawing
boundaries.
First note that in any such case as that just described, there are two "phonetic
similarities" in the picture: (1) the phonetic similarity between the sounds which
are in complementary distribution; (2) the phonetic similarity of each of the
sounds to the environments in which it occurs. In our example, initial and final

[p] are phonetically similar to medial both are produced with bilabial
[b] in that
and velic closures [p] is like its environment on at least one side in being voiceless
;

(just as silence is voiceless), and [b] is like its environment on both sides in
being voiced.
The method of allophones works in terms of essentially "vertical" cuts (trans-
verse to the time axis). We find an allophone [p] in one set of environments, and a
similar though phonetically different allophone
[b] in another set of environments.

We then set up a single phonemic unit /p/ which is, in a sense, a "class" of
allophones, in this case including both [p] and [b], and we state the circum-
stances under which the single phoneme
be "represented by" or "actualized
will
as" one or the other of its We
have kept the requirement of phonetic
allophones.
realism in mind in that the allophone which represents phoneme /p/ in a given
environment is phonetically similar to that environment. If the distribution of
the two allophones were reversed, then the complementation would not be sup-
ported by phonetic realism and the identification would probably be rejected (I
know of no language where such a peculiar distribution occurs).
 A MANUAL OF PHONOLOGY 157

The method of redrawing boundaries works differently. Instead of setting up a

"phoneme" which has two "allophones," what we do is readjust the
class-type
boundary between "sound" and "environment." Specifically, we can delete the
voicing of[b] from the [b] itself, leaving [p], and assign the voicing to the environ-

ment. This needs to be demonstrated graphically. For the purpose, let us consider
a word which we have transcribed linearly as [pabap]. This transcription can be
expanded as follows:
|V|V|V|
P|A|P|A|P,
where "P" means bilabial closure, "A" low central tongue position, and "V"
vibration of the vocal cords. (Other symbols could be added say an "0" for
closure of the velic
but would be the same in all five columns and would not
bear on the point at issue.) Now to say that and
complementary
[p] [b] are in
distribution in this language is to say that the simultaneous bundle of
"V" and
"P" occurs only where both the preceding and the following bundle involve "V,"
and that "P" without "V" does not so occur. The vertical lines in the expanded
transcription above represent the usual sort of boundaries between successive
sounds. We can now redraw those boundaries as follows:

|V V V|
P A P A P.
I
I
[
j

Having done this, we see that the element "P" recurs three times in the form,
once in an environment of preceding, simultaneous, and following voicing, twice
without such voicing. Instead of ha\ing two distinct units to deal with we have
just one. By a rather easy definition, we can arrive at a linear transcription which
shows all of this: /papap/.
Most operations involving the criterion of "phonetic realism" can be performed
by this method rather than by the method of allophones. However, in examining
our raw materials (utterances in allophonically correct transcription; 31), and
searching for the specific items which call for this sort of readjustment, it is
convenient to think largely in terms of block-like phones following each other
without (structural) overlap. One looks for phonetic overlap, bearing in mind
Pike's suggestive aphorism "Sounds tend to slur into their environments."
Having found instances of what can be described in this rather non-rigorous
manner, one can then optionally apply the method of redrawing boundaries.
There is rather general agreement about the application of the principle of
phonetic realism, but there is much disagreement as to just what constitutes
"phonetic similarity." Some investigators would like to say that phones cannot
be phonetically similar unless they share one or more "distinctive features" (in
our terms, ultimate phonologic components) this would also imply that ultimate
;

phonologic components cannot be phonetically similar at all. This requirement

was clearly met in our [pabap] example, but in general we cannot expect things
to be so simple. There is often, apparently, some choice as to what ultimate
phonologic constituents shall be recognized, and our choice along this line may
go hand in hand with our choice of larger units which are to be taken as "pho-
netically similar."
 ;

158 A MANUAL OF PHONOLOGY

Consider, for instance, the long syllable peaks of Fox: [i- o- ae- a]. If we slice

[i-] into two consecutive halves, then to what is the second half phonetically
similar? Since the tongue position is high and front, it is like the first half of
the same long peak, and also like the short peak [i]. Since it adds to the length
of the peak, it is phonetically like the second half of [o], [a], or [ae]. Since it is

vocalic, it is like the intervocalic [h] of Fox, the latter being a brief phase of
voicelessness. Any of these phonetic similarities may be evoked in our analysis;
our choice depends on distributional matters rather than on a decision that one
of the phonetic similarities is somehow "closer" than the others; the particular
way which we view the ultimate phonologic constituents of the system will
in
come about as a result of our choice.
It seems better, on the whole, not to insist on the stand taken by the investi-
gators mentioned above, though wherever their requirements are easily met the
ensuing steps are perhaps clearer. In dealing with the old Germanic languages,
Moulton refuses to set up a phonemic identity between a prevocalic voiceless-
vocoid onset [h] and a dorso- velar spirant [x], despite complementation, on the
grounds that [h] and [x] share no distinctive feature. The particular identification,
for the particular languages (e.g., Old English) may well not be suitable, but we

should not bar it on the grounds given especially when we find, in some lan-
guages, a single unit in a single position showing free variation through a range
which includes [h] and [x] (e.g., the onset /x/ of Mandarin).
It should hardly be necessary to add that failure to see phonetic similarity
can be induced by bad phonetics, particularly by an uncritical reliance on tradi-
tional use of phonetic descriptive terms. The initial [h] of English he, haw, who
has often been called a "glottal spirant"; the association of the term "spirant"
with sounds like [f 6 s s x] can lead one to overestimate the phonetic difference
between the [h] and the voiced postvocalic glide of yeah, law, bah.
There is one aspect of phonologic analysis where the usual principle of phonetic
realism, in the writer's opinion, must be not merely suspended but almost re-
versed. This is in the search for junctures, to be discussed in 324.

3223. Pattern Congruity. Roughly speaking, this principle states that, of

several alternative analyses which equally well meet other requirements, one
should choose that which yields the greatest symmetry, both of phonemes and
of allophonic variation with phonemes. Heuristically, all that the principle can
do is guarantee that all the symmetries in the system be discovered and reported
it is important not to let the analyst's esthetic preferences distort the facts and
"find" greater symmetry than
is actually there. In gathering data, one makes,

and constantly showing the segments which occur in contrast with

revises, tables

each other in apparently relevant environments as onsets, as codas, and so on.
If in one such table one has, let us say, voiceless and voiced stops in three posi-
tions of articulation, but glottalized stops only in two of the three, one suspects

that perhaps not necessarily, but perhaps the third glottalized stop will turn
up, rounding out the pattern. Now this "turning up" can come about in several
ways. A form not previously recorded may turn up and contain the "missing"
unit. Or a form already recorded may turn out to have been heard wrongly,
and upon correct hearing may show the missing unit. Or, if neither of these
 A MANUAL OF PHONOLOGY 159

events comes about, the analyst may conclude that something which he has
been dealing with in other terms actually belongs in the empty pigeonhole.
It is at this point that the danger of forcing one's data becomes greatest many ;

systems, as we saw in 2, do have holes. Yet the analyst has every right to play
around with any notions of the kind which occur to him they may prove fruitful. :

An example is Taos, with five voiceless unaspirated stops (and affricates), five
glottalized ones to match, and only two voiceless aspirated ones. It may make
sense to "fill out the pattern" by regarding /p' t'/ and the three spirants /s x x w
/
as a third obstruent type, matching the other two in all positions. This does not
make aspirated stops out of /s x x w nor spirants out of /p' t'/. What it does is
/,
to treat, say, /x/ as differing from /k/ primarily in the presence of a certain
feature in /x/ lacking in /k/, /p'/ as differing from /p/ primarily in the presence
versus absence of this same feature. There is still a difference between /p'/ and
/x/ in the former, the arrangement of the /p/ and the added feature is sequential,
:

whereas in /x/ the arrangement is simultaneous. This difference of arrangement,

however, is relegated to secondary status.
Pattern congruity is also appealed to as a guide in cases of multiple comple-
mentation. The unaspirated voiceless stops of spill, still, skill are in complemen-
tation both with the aspirated voiceless stops of pill, till, kill, and with the un-
aspirated voiced stops of bill, dill, gill. This is an example we have given before,
and which will come up later for what the writer believes to be the most satis-
factory treatment. For the moment, let us assume that we must insist on identi-
fying the unaspirated stops either with the aspirated ones or with the voiced
ones, with no third alternative. Now we have already seen that, by and large,

clusters of obstruents as codas are voiced or voiceless as wholes: /st/ past and
/zd/ razzed, but not /sd/ or /zt/. The /s/ of past is surely the same as the /s/
of sill or of spill, still, skill. In codas it is /p t k/ which go with /s/, /b d g/ going
rather with /z/. Therefore, for the sake of neatness of pattern, we conclude that
the stops in spill, still, skill are phonemically /p t k/.

Pattern congruity is also evoked when one confronts a decision between single
phoneme and cluster: e.g., a [k w which may be /k w/ (unit) or /kw/; a [6] which
]

may be /6/ or /ts/ or /ty/; an [a-] which may be a unit or a cluster of /a/ plus
something (another /a/, a covowel //, a semiconsonant). A [6] cannot be inter-
preted as /ts/ if there is a [ts] in contrast, of course (the principle of contrast
and complementation) but in some cases there is only one segment of the kind
;

and still there may be grounds for one interpretation or the other. In German,
[6] is a cluster /t/, because a whole series of clusters of
stop plus spirant occur:
/pf ps ts ts" ks/. In Fox, [6] is usually taken as a unit because it parallels closely

the pattern of positions of occurrence of obviously unitary /p t k/, and because

treating it as a cluster /ts/ or /ty/ gives less parallelismless "pattern con-
gruity." We have more to say below about this type of problem also, and
shall
the above examples should not be accepted as definitive discussions.
3224. Economy. The general law of parsimony applies in phonologic analysis
as it does elsewhere, but in actual practice it is of remarkably little help. No one
sets up fifty phonemes by dealing with, say, onsets and codas in complete inde-
160 A MANUAL OF PHONOLOGY

pendence, when the number can be halved by identifications of elements in onset

position and in coda position. Sometimes an investigator will establish two
separate entities where the recognition of one actually eliminates the necessity
of the other: this is unparsimonious and subject to correction, providing the
evidence is all in. However in the course of analysis it is often very wise to recog-

nize more items than will probably ultimately be proved independent. Thus in
working with Cantonese it would be wise at first to keep syllable-juncture and
also to indicate the distinction between onsets /y w/ and peaks /i u/. In due
time, however, it turns out that, given syllable juncture, [y] and [i] are in comple-
mentation, as are [w] and [uj. Whenever one becomes sure of such a situation,
one reduces the number of recognized elements.
Parsimony is sometimes evoked where it is not relevant. Some analysts prefer
to list a large number of, say, consonant "phonemes," and have few or no clusters
to deal with others prefer to treat as
; many segments as possible as clusters, giv-
ing a smaller number of "phonemes." The latter group cannot rightly evoke
parsimony in defense of their position. The difference lies entirely in the level of
complexity at which the term "phoneme" is applied. If "phoneme" refers to
anything of a higher status in the hierarchic pattern than that of ultimate
phonologic constituents, then part of the total task of description is the specifi-
cation of the constituent elements of the phonemes themselves. The worker who
setsup a larger number of "phonemes" has to present more information of this
kind, and less at the "superphonemic" level (clusters etc.); the worker who sets
up fewer "phonemes" must cover less "subphonemically" but correspondingly
more "superphonemically." Quite obviously, equally accurate descriptions can
be presented either way, and there is no way to show that one way is any more
economic than the other.
Recognition of this relativity in the size-level at which one introduces the term
"phoneme" should knock the wind out of the sails of the proponents of all sides
of the argument about the "phonemic" status of English syllables. One group
argues that the syllables (that is, in our terms, the peaks, simple or complex) of
beat, boot, bate, boat, bite, bout, Hoyt, and so on, should all be called single "pho-
nemes," along with those of bit, bet, bat, book, buck, bock, wash. A second group
argues for the breakdown of the syllables in the first list into two "phonemes"
each. Still a third group wants to call the peaks of beat, boot, bate, boat, fa, law
all single"phonemes," but recognize two each in the peaks of bite, bout, Hoyt.
Now first group can be defended as a consistent use of the
the position of the
term "phoneme" to cover all the units which appear distinct and in contrast
when a certain level has been reached in working down the IC scale; it remains
for them to show the complex structure of some of these "phonemes" in terms of
ultimate phonologic features, in simultaneous or successive bundles. The position
of the second group can likewise be defended as a consistent use of the term
"phoneme." Only the position of the third group is indefensible, and that not
because they use the word "phoneme" in a slightly different way but because they
mix IC-levels. We shall see a more thoroughly worked-out example of this sort
of thing shortly.

323. Crucial Problems. There are three types of problems which have recurred
 A MANUAL OF PHONOLOGY 161

constantly in phonologic work and theory. One is that of phoneme versus cluster;
another that of multiple complementation; the third that of marked and unmarked.
In each case, most systems of phonologic analysis have erred by providing for
fewer alternative possible answers than the facts seem to require, thus forcing
arbitrary decisions. The strict IC approach by no means eliminates all the diffi-

culty, but it seems at least to afford greater flexibility.

3231. Phoneme Versus Cluster. The very way in which this question is usually
posed indicates misplaced emphasis. The phraseology implies that a [S]-type
phone, for example, must either be a unit phoneme or a cluster of an integral
number (presumably two) unit phonemes. By any one precise set of rules for
the use of the term "phoneme," this is perhaps true, but this simply shifts the
arbitrariness to the rules instead of to their application. The question "phoneme
or cluster?" is only apt to be asked of a segment which seems in some ways to
work like items that have already been accepted as unit phonemes, in other ways
like items that have already been interpreted as clusters. What the IC approach
adds to this is primarily a recognition that the ICs of a unit occur in that unit
in a specific arrangement, and that the arrangement itself may be a matter of
primary concern. The IC approach also allows greater flexibility in that there is
no insistence on exhausting all possible "vertical" cuts between ICs (cuts trans-
verse to the time axis, giving ICs in sequential arrangement) before cuts of other
kinds, yielding other arrangements, are made.
We by presenting, first, a typical discussion of the status of [6]
shall illustrate
in Fox, followed by a statement in IC terms.
Traditional Statement. The Fox segment [6] occurs as an onset, as an interlude,
and as part of an interlude [h6]. These positions of occurrence are shared by
/p t k/. Unlike /p k/, [<$] does not occur before /y/; this is shared by /t/. Unlike
/p t k/, it By the use of slant lines we have indicated
does not occur before /w/.
that these other segments are being interpreted as unit phonemes; it is only
all

the status of [c] which is in doubt. [6] also shares witn /p t k/ the privilege of
occurrence before any vowel, /y/ does not occur before /i i-/, and /w/ does not
occur before /o o/.
There are three phonetically realistic alternatives to be considered. One is to

take [6] phoneme //; a second is to take it

as a unit as a cluster /tS/; a third

is to take as a cluster /ty/. Either the second or the third interpretation in-
it

volves, naturally, some redefinition of /t/ and of // or /y/. If we treat [6] as

/ty/, then we must consider a parallel treatment of /S/ before a vowel; // in
the clusters /k sky skw/ may complicate this.
Unit interpretation yields distributional results already stated above.
Interpretation as /tS/ means that this is being recognized as the only cluster
in the language consisting of stop plus spirant; this throws /t/ out of parallelism
with /p k/, and // out of parallelism with /s/.
Interpretation as /ty/ yields a neat pattern for stops, since we now have, as
occurrent margins, /p t k/, /py ty ky/, /pw tw kw/, /hp ht hk/, /hpy hty hky/,
/hpw htw hkw/. This is a clear gain in "pattern congruity." On the other hand,
the revised /y/ which is involved in the interpretation has a peculiar distribution;
 :

162 A MANUAL OF PHONOLOGY

IyI now occurs before /i i-/ providing that the /y/ is itself preceded by /t/ (or
by /s/, if we also interpret prevocalic [s] as /sy/), but not otherwise. There is

nothing else in the distributional pattern of the system which resembles this.

It is hard to be sure what conclusion would be drawn from the above consider-
ations; it would depend in part on the preferences of the analyst. The writer's

own preference, within the limitations of the "phoneme or cluster" approach, is

to settle for a unit phoneme
Another possible answer is "it doesn't matter."
/&/.
This answer is not bad if it is given after the examination of the possibilities and
what they imply; it is bad if it is given in advance and shortcuts the investiga-
tion. For, in a way, what counts is not so much the interpretation which one
accepts as the evidence on which that interpretation is based. The most fruitful
aspect of the lengthy discussions of the "phoneme or cluster" problem in the
literature has been the careful examinations of distributional facts which were
undertaken; they have posed as a byproduct, but only because we were really
asking the wrong question.
Before passing on to an IC approach to the Fox problem, let us first indicate
that the choice of notation is a separate matter and should not be confused with
what we are discussing. Fox [6] has to be indicated somehow in a transcription,
by a unit symbol, or by /ts/, or by /ty/; any of these three, or some other nota-
tion, would do perfectly well in that there would be no ambiguity. And with any
choice of notation, the problem of the structural status of [6] would remain.
IC Statement. Working from large to small in Fox, we eventually come to a
listing of all those elements which occur as onsets, as interludes, or both. At
the moment, no matter what notation we use for these, they are all on a par.
The full list is given in 2511 (we now call this List A). When we proceed to
break the elements of this set into their own ICs, we first break twelve of the
set down as follows:

/hp/ = /ty + hi /hpw/ - /ty + /pw/

/ht/
/he/ =
= N N
/ty
+
+ //
/htw/ = /ty
/hkw/ = /ty
+
+
/tw/
/kw/
/nk/ -
/hpy/ =
N
/ty
/ty
+
+ /py/
mi = HI
/Sky/ = //
+
+
/ty
/ky/
/hky/ - /ty + /ky/ /Skw/ = // + /kw/
The first ICs, /ty and //, appear also on our original list (List A) as whole

margins; the various second ICs also all appear on List A. This first breakdown
therefore leaves us with the following lower-level elements which must in turn

pt6khsmnw
be analyzed

py
if possible (List

k
ky
B)
h
hy
s m
my
n
ny
pw tw kw hw sw mw nw
The recurrence of some elements on both List A and List B, where other elements
appear on List A but not on List B, what happens in the analysis of
is typical of
any hierarchic organization. A list of all members of the clergy of the Church
of England would include the names of all bishops, archbishops, and primates.
A list of all the bishops would still include the names of all archbishops and
 : : :

A MANUAL OF PHONOLOGY 163

primates. A list of the archbishopswould still include the names of all primates.
A list of all primates would include but one name. An analogy in terms of the
areas spiritually ruled by these various members of the clergy would be even
closer.
The next IC cut applies to twelve of the elements in List B
/py/ = /P/ + 111 /pw/ = /p/ + /w/
/tw/ = /t/ + /w/
/ky/ = /k/ + /y/ Aw/ = A/ + /w/
/hy/ = /h/ + /y/ /hw/ = /h/ + /w/
/sw/ = /s/ +/w/
/my/ = /m/ + /y/ /mw/ = /m/ + /w/
/ny/ = /n/ + /y/ /nw/ = /n/ + /w/ .

Once again all the first and all the second constituents are elements which al-
ready appear on List B. That is, we state that the /y/ of /py/ (whether the latter
is a whole margin or is occurring as one IC of the more complex interlude /hpy/)

is the same element as the /y/ which occurs as a whole margin; and so forth.

However, we have not, in this step, broken /&/ down any further, precisely
because of the difficulty in "pattern congruity" which is encountered if we
reinterpret /c/ as /ty/ in a way exactly paralleling the breakdown of /py/ into
/p/ and /y/. After the second IC cut, we are left with the following elements
(List C):

ptc'khss'mnwy.
In the first and second cuts, the plus-sign has had a very simple meaning:
"followed by." That is, we meant to state that the unit /hp/ consisted of the
lower-level unit /h/ followed by the lower-level unit /p/, and so on. We are now
going to cut in a different way at a different angle, as it were though we shall

for simplicity use the plus-sign again (third breakdown)

// = N + hi // = N + hi .

Here the plus-sign means something like "simultaneous with" or "partly over-
lapping." Once again we identify the ICs with elements already in our fist (List
C), but the arrangement in which the ICs /t/ and /y/ participate in /<$/ is differ-
ent from the arrangement in which the ICs /p/ and /y/ participate in /py/.
Our residual

ptkhsmnwy.
list is now (List D)

Further cuts will necessarily be other than "vertical," and will for the most
part yield constituents which are not already on the list constituents such as
stop closure, bilabial position, and so on (262).
This treatment deals with /c/ as different both from /p t k/ (which appear on
List Das well as on all earlier fists) and from /py ky/ (which appear on Lists A
and but not on those from C on). Such a treatment matches the distributional
B
facts, which show a set of privileges of occurrence for /d/ which resemble both
those of /p t k/ and those of /py ky/, but differ from both in some ways.
Comment. Now is // a unit phoneme or a cluster? We have not used the term
"phoneme" in our IC analysis, since we do not know where to introduce it. One
164 A MANUAL OF PHONOLOGY

could logically call all the elements in List A "phonemes," or all those in List B,
or all those in List C, or all we introduce the term "phoneme"
those in List D. If
in connection with List A, B, or C, then /&/ is a phoneme except that in the
first case (List A) it is both a phoneme and also a part of the larger phoneme /h6/.

In the last case (List D), /<$/ would be a cluster but a different kind of cluster
from /py/, /ky/, and the like. We see that the question "is /c/ a phoneme or a
cluster?" is too narrow and not properly fruitful; a more productive question to
ask is "on what hierarchic level, or levels, does the given element function, and
at what step, working down the IC scale, does it break down into smaller con-
stituents?" These are the questions which we have tried to answer above; the
term "phoneme," and just how to use it, is regarded as of subsidiary importance.
A similar approach may
be revealing in other cases of the kind which have
supplied the ammunition for discussions of phoneme-or-cluster. Let us consider,
very briefly, English /6/ and /$/ (or, what amounts to the same thing, the
voicing-irrelevant unit which can be extracted from these two). In some dialects,
these units contrast with clusters /ts/ and /d2/ : 'pitcher : won't ya; ledger : did ya.
In some of these same dialects, / 3/ contrast also with /ty dy/ : choose : tune,
teutonic; juice : dune, deuteronomy. The customary conclusion, that // and /$/
are "therefore" unit phonemes, misses the main point. Neither /&/ nor /$/
participates in more complex onsets, while the most similar other elements,
/p t k b d g/, all do. /6/ and /5Y seem to occupy a position in the hierarchic scale
midway between elements like /p t k b d g/ (which remain unbroken further
down the scale) and elements like /pr tr kr br dr gr pi kl bl gl/ (which are broken
into smaller constituents at an earlier step). Perhaps we must regard /&/ and //
as something like close or intimate sequential clusters of /t/ and //, /d/ and /&/,
in contrast to the "normal" (though much rarer) clusters /ts di/. This would
involve a recognition of two kinds of sequential arrangement (intimate and
"normal"), not contrasting in most cases; and, of course, one's notation would

have to indicate the difference perhaps precisely by writing the intimate clusters
with the symbols /<$ 3/.

3232. Multiple Complementation. On this type of problem the Americanist and

the Praguian traditions have supplied different machinery. The Americanist
position has been that in English, for example, the stop after /s/ in spill, still,

skill must be "the same" phonemically either as those of pill, till, kill, or as those
of bill, dill, gill. On this assumption, we showed in 3223 that the identification
with the stops of pill, till, kill is neater. One result is that /b d g/ are somewhat
phonemes they occur for the most part in positions paralleling those of
defective :

/p but do not occur in initial clusters after a spirant.

t k/,

The Praguian tradition affords greater flexibility in the treatment of such

problems. In Prague terms, a pair of phonemes, say English /p/ and /b/, which
differ insome one constituent feature but are otherwise identical and composed
of features not all of which occur in any other phoneme, constitute an archi-
phoneme. In certain positions both phonemes of an archiphoneme occur, in con-
trast. In positions where the contrast is irrelevant, one does not say that what
occurs is either one phoneme or the other; rather one says that it represents the
archiphoneme, the contrast in that position being neutralized.
 A MANUAL OF PHONOLOGY 165

This is a neat way to avoid a spurious identification: instead of insisting that

the bilabial stop of simply must be either /p/ or /b/, one has a way of saying
spill
that it is both and neither: it is what /p/ and /b/ have in common, without the
distinguishing features of either as opposed to the other. The way is not available
in all cases of multiple complementation. In English, /p k b g/ all occur in onsets
before /l/ (plain, clay, blare, glare), whereas /t d/ do not. But one cannot speak,
say, of neutralization of the contrast between /p/ and /t/ before /l/, since
/p/
and /t/ do not constitute an archiphoneme, nor do /t/ and /k/. /p/ and /t/
have in common a manner of articulation, and differ as to position; but the man-
ner is shared by /k/. On the other hand, one may sometimes recognize three-way
neutralizations. There are dialects of Chinese which have a three-way stop
distinction initially, /p t k/, but only one stop finally ([t] or [?]) since /p t k/
;

differ as to position but are otherwise identical, and since their common features
are not shared by any other phoneme, the three can be regarded as constituting
an archiphoneme, and the contrasts can be said to be neutralized finally. (Simi-
larly in Burmese.)
Now what the Americanist and Praguian have in common is an
traditions
insistence that yertical cutting be completed before any other kind of cutting
into ICs be attempted, and that once one has shifted to any sort of "decomposi-
tion" of segmental units into simultaneous components, no more vertical cuts be
allowed. In our discussion of IC-procedure (321) we made no such blanket rule,
and, indeed, did otherwise in a number of examples. One of the examples was
that of clusters of obstruents in English. We broke such codas as /ts dz ps bz
st zd pt bd/ first into ICs /TS PS ST PT/ (voicing-irrelevant constituents) and
/H/ /A/ (voicelessness and voicing). Only after this did we proceed to break
or
/TS/, by a vertical cut, into a voicing-irrelevant apical stop /T/ and a voicing-
irrelevant spirant /S/, in that order.
Let us examine the onsets of spill, still, skill in this light. These are, respec-
tively, /SP/, /ST/, and /SK/ plus /H/; the first ICs then break into /S/ and,
respectively, /P/, /T/, and /K/. There is no question of neutralization between
/p/ and /b/ after /s/: /sp/ contains neither /p/ nor /b/, but /SP/ and /H/,
and the first of these in turn contains neither /p/ nor /b/, but the voicing-
irrelevant stop /P/. However, there is a question of limited distribution: the
elements /SP ST SK/ in onsets are never accompanied by /A/, always by /H/.
If neutralization is to be spoken of here, it is a neutralization between /H/ and
/A/. But /H/ and /A/ are not phonemes that have something unique in com-
mon and otherwise differ: they have nothing at all in common phonetically
(that is, though they are both effects produced by differing functions of the
glottis, they cannot be broken down into smaller constituents such that some of
the smaller constituents occur in both, others in just one or the other) The basis .

for the Prague terminology is therefore lacking.

It is important to recognize the difference between an environment in which a
particular contrast is relevant and one in which it is not. The Americanist tradi-
tion does this in terms of phonemes and full or defective distributions. The
Praguian tradition does it somewhat more neatly in some cases, in terms of
phonemes, archiphonemes, and neutralization, but some cases are not subject to
 ;

166 A MANUAL OF PHONOLOGY

this treatment (non-occurrence of initial /tl/, for example, in English), and for
them one has to revert to a procedure like that used in the Americanist tradition
for all cases. The IC-approach cannot eliminate defective distributions, but with
care it is possible to perceive them on the proper hierarchic level. Thus in English
there are no microsegments of the shapes /gliyt/, /tlip/, or /zdil/. But these
three lacks are of three different kinds, /gliyt/, if it occurred, would be composed
of an onset which occurs, a peak /iy/, which not only occurs but occurs
/gl/,
after /gl/ (gleam), and a coda /t/, which not only occurs but occurs with onset
/gl/ (glut) and with peak /iy/ (eat). The "hole in the pattern" is obviously at a
fairly high size-level, /tlip/, if it occurred, would involve an otherwise unob-
served onset /tl/; the sequence /tl/ does occur elsewhere (for some speakers:
rattling). The "hole in the pattern" in this case is level, but
not at the syllable
specifically at the level of onsets. The non-occurrence an even lower
of /zdil/ is at
level: the onset /zd/ does not occur, but /st/ does, which means that /ST/
occurs in onsets but not in the company of /A/. An even lower-level "hole" is
the non-occurrence of a spirant /x/ or /y/, which would involve a simultaneous
cluster of a position of articulation which does function in the language (/k g/)
and a manner which also functions (/f v s z/).
In any case of multiple complementation the aim of the IC approach is funda-
mentally to determine the hierarchic level on which the hole appears. If this,
as it seems to the writer, is the important issue, then both the Americanist and
the Praguian traditions have tended to obscure the problem.
3233. Marked and Unmarked; Simple and Complex. In connection with the
terminology of neutralization, the Praguian approach calls one member of certain

archiphonemes the unmarked member, the other or others marked. In Russian

/p b/, for example, /p/ is the unmarked member, while /b/ has the positive
marking of voice. Our partial rejection, in 3232, of the theory of neutralization
does not invalidate this way of speaking, which is very useful. Rather more
generally, we may distinguish between contrasting units (perhaps clusters rather
than single phonemes) where one is phonologically simpler than the other or
others. The analytical problem, under an IC approach, is to know which member
of a contrasting set is to be regarded as the simplest. The stock of ultimate
phonologic components may differ depending on how we answer this question
in any one case.
There are two criteria which can be used realistically in solving this problem
the two never seem to conflict, but in some cases neither affords an answer.
One is that, if one of two or more matching sets of units has a wider distribu-
tion than the others, the former is phonologically simplest. This applies for
Russian /p b/, /t d/, and so on: the voiceless obstruents occur before juncture
(at the end of a macrosegment), while the voiced obstruents do not. In Winnebago
we and finally; the unaspirated and
find the aspirated stops initially, medially,
glottalized stops do not occur finally. We conclude that the aspirated stops are
phonologically simplest; and it turns out that the others can all be regarded as
clusters or as the result of allophonic variation (25212). In Korean we find the
short stops /p t k/ finally, while other varieties occur only initially or medially
 A MANUAL OF PHONOLOGY 167

or both; once again, it is possible to set up /p t 5 k/ as single phonemes, and

handle the other varieties as clusters (25212).
all

The other criterion is that, if one of two or more matching sets of units has a
wider range of non-distinctive variation than the other or others, the former is
phonologically simplest. In Nootka the nonglottalized stops are unaspirated
before vowels, aspirated finally or before consonants; the glottalized stops and
the spirants show no such variation. We conclude that the plain stops are the
simplest set, and that the glottalized stops and the spirants are more complex.
Since in this case all three turn out to be unit phonemes rather than clusters (at
least in the sense that /p'/, for example, cannot be reinterpreted as a cluster
/p ? /)> we call the plain series unmarked, the other two series marked. The first

criterion in part supports this conclusion, no way goes against it: glot-
and in
talized stops do not occur after a vowel before a consonant or at end of macro-
segment. But in Georgian the same criterion gives phonetically different results.
One series of stops shows a variation between voiceless unaspirated and voiceless
unaspirated glottalized; a second is always aspirated, and a third is always
voiced. The first series, then, is unmarked, and the other two have aspiration
and voicing as positive marks. Here, also, there is some distributional support.
There are few limitations on occurrence in sequence of Georgian consonants,
but it is true that only the series of stops which we take as unmarked occur
before /./.

The decision not always easy to make, and perhaps in some cases we must
is

suspend judgment. English /p t k/, it would seem, show more nondistinctive

variation than /b d g/; but here the extraction of voicelessness as a "long com-
ponent" (3232) takes precedence, so that the question of marked and unmarked
does not come up in quite the same way.
The bearing of simple and complex on the tabulation of ultimate phonologic
constituents is easy to see. Our analysis of Nootka in 262 is based on the fact
that /p'/ is more complex than /p/. Given exactly the same stock of consonantal

units, but with a /p'/ which was sometimes glottalized, sometimes not, and a /p/
w w
which was always aspirated, we should have to analyze /p t c 6 X k k q q / as
involving the feature of which /h/ consists, and /p' t'/ etc. as involving one less
w w
feature than is recognized in 262; the spirants /s s i x x x x / could not be
recognized as containing the feature which characterizes /h/. Furthermore, /h/
would remain a manner consonant, but /?/ would cease to be one; the spirants
would not be matched by a manner consonant.
324. Junctural Analysis. Junctures are in some ways a very different kind of
phonologic entity from ordinary vowel or consonant phonemes (however one
may choose to apply the term "phoneme" in the latter case). We can well enough
call the English bilabial stop /P/ a phoneme: the /p/ of pot
consists of /P/ and

/P/ and /A/; the /p/ of clasp consists of /P/

/H/; the /b/ of bought consists of

and part of an occurrence of /H/, which latter stretches also through the pre-
ceding /S/. Now all the articulatory events which we say constitute occurrences

of this element /P/ involve velic and lip closure. An articulatory event does not
constitute an occurrence of /P/ unless it does involve both of these features, and
168 A MANUAL OF PHONOLOGY

if itinvolves them both it is by definition an occurrence of /P/; any accom-

panying articulatory motions or positions are not part of the /P/, but environ-
mental. It is by and large typical of ordinary segmental phonemic units that all
occurrences of them have something articulatory in common. This is a derivative
of the heuristic principle of "phonetic realism" (3222), and we said when dis-
cussing that principle that it had to be suspended in the analysis of junctures.
In slyness, the internal open juncture is represented by a drawl of the /ay/

which is longer than an /ay/ in a form like minus, and com-

in the first syllable,
parable to the /ay/ which occurs in macrosegment-final in a sentence like He's
very sly. The /n/ of -ness is accompanied by voicing which begins fairly weakly
and increases (crescendo voicing); this is typical of a macrosegment-initial /n/
in a form like Never do that, and different from the fairly even voicing accompany-
ing a medial /n/ as in minus. This voicing difference is subtle and perhaps not
always present. When it is present, we say that the crescendo voicing or, more
precisely, the crescendo characteristic of the voicing
of the /n/ of slyness is
part of the juncture, not part of the /n/ itself. The juncture of slyness, then,
iscomposed of two features one, always present, is the drawl on /ay/ the other,
: ;

sometimes not present, is the crescendo of the voicing accompanying the /n/.
In some utterings of the form slyness not only the second but also the first of
these features may be missing; but when this is the case the form has been said
in a phonologically different way, without any juncture, making it rhyme com-
pletely with minus.
In night-rate the juncture is composed again of two features. The first is the
type of release accompanying the /t/ of night-; this /t/ is either in-
audibly released or has slight aspiration. This is the nondistinctive range of
variation of release of a final /t/, as in He arrived last night, and differs from the
release of a medial /t/ as in nitrate or bottom. However, there is a range of varia-
tion here and this distinction is not always audible ; it is like the evanescent dis-
tinction in voicing-contour between the /n/ of slyness and that of minus. The
other is the full, slightly crescendo, voicing of the /r/ of rate, like that of an initial
/r/ as in Rates are high there, and quite unlike the largely voiceless /r/ of nitrate.
This feature is much more constant on the rare occasion when the form night-rate
;

isspoken with neither of these characteristic features, then, of course, it is being

spoken just like nitrate, with no juncture.
In these two different environments of juncture (slyness, night-rate), the ele-
ment composed of almost completely diverse phonetic features. The contrast
is

between this and the constant common denominator of a segmental unit like
/T/ is obvious.
But why should we attempt to discover and recognize any such peculiar ele-
ments a phonologic system? Why cannot one set of principles be used through-
in
out? If we think of the heuristic principles of 322 as the fundamental operating
principles of phonologic analysis, then junctures indeed stand out of line, and
it is a fact that a phonologic system can be completely analyzed without recog-
nizing any junctures at all. The contrast between nitrate and night-rate can be
provided for by setting up two different [r]-like phonemes; that between slyness
 A MANUAL OF PHONOLOGY 169

and minus by setting up /ay/ and, say, /ay-/, as different units, perhaps sub-
sequently extracting the "drawl feature" // as a lower-level unit.
However, if we think of the IC-approach as the fundamental orientation, and
the principles of 322 as ancillary, then the recognition of junctures occupies an
entirely logical place, despite the large difference which we have demonstrated.
Confronted with long utterances as well as short, the IC-approach implies that
we should break them down a bit at a time, in whatever way the facts allow.
One of the things we look for in this connection is parallelisms between medial
and terminal phenomena (where "terminal" subsumes both "utterance-initial"
and "utterance-final"), versus phenomena which occur only medially. If we
find a point in the interior of an utterance where articulatory motions are or-
ganized in a way typical of utterance-final and utterance-initial environment, in
contrast with organizations occurrent only medially, we suspect that we have
found an interior boundary of some kind at which the whole utterance can be
broken into successive ICs. This suspicion becomes stronger if we find that
either half of the whole utterance is identical with some shorter whole utterance,
although such is not always the case. The technique of junctural analysis allows
us to organize all such phenomena in a simple way, and affords an excellent
structural point of departure for IC analysis of the whole system.
Sometimes no such special interior points are to be found. This would seem to
be the case in Fox. The Fox phrase /emowrMciwrsahke-howryanonrcamesahi/
'Itwas when Wisahkeha was living with his wife and children' is uttered without
any internal pauses. There is a smaller isolable segment (i.e., one which occurs
as a whole utterance) /e-howrkici/, and likewise one /onrcamesahi/, but if
these two be extracted from the beginning and ending of the long phrase above,
the remainder is not pronounceable alone. The "same" phrase (in terms of
meaning, not phonology), pronounced a word at a time, becomes /e-howrki&,
wi-sahke-ha, owryani, onrca-nesahi/; when said rapidly, the /a/ at the end of

the second word and the /i/ at the end of the thiid disappear. The sequence
/iclwi*/ across the first "word"-boundary is not characteristic in any way what-
soever of sequences across "word"-boundary when the original long phrase is
uttered, nor is the sequence /yanoni-/ across the last "word"-boundary. There

are, true enough, certain medial sequences here which might be broken into a
possible final portion followed by a possible initial portion, and other medial
sequences which could not be so broken down. We can put a hyphen into the
phrase to indicate each place where one might interpret in terms of occurrent
final plus occurront medial: /e-ho-wrki-ci-wrsahke-ho-wi-ya-no-m-ca-ne-sa-hi/.
No hyphen follows a long vowel, because long vowels do not normally occur
finally. None follows a consonant, because consonants never occur finally. No
hyphen precedes an /h/ or a cluster of // plus consonant, because these do not
occur initially. We however, no minimal contrasts between points where a
find,

break is possible and points where a break is not, and there is little positive cor-
relation between the points where hyphens were inserted above and actual bound-
aries between segments which occur as whole utterances.
In this case, then, we say that there is no juncture in the language. There is no


170 A MANUAL OF PHONOLOGY

basis on which to divide the long Fox phase into some small number of ICs (once
we have lifted off the intonation) ; we are forced to interpret the phrase as com-
posed of as many ICs as there are syllables, namely seventeen.
The two English examples already discussed (slyness, night-rate) show clearly
the difference between Fox and English in this respect. Let us give another ex-
ample, which will also show how the extraction of certain features from certain
contexts and their assignment to junctures renders the handling of ordinary seg-
mental phonemes neater. In Ben's here we have an initial bilabial stop with
crescendo voicing, which is the only kind of voiced stop found initially in English.
In Want a cab? we have a final bilabial stop with diminuendo voicing, the only
occurrent final variety. In He's a cabby we have a medial bilabial stop with even
and fairly full voicing. In / see Ben we have a medial /b/ different from that in
He's a cabby: the voicing is not full, but crescendo, like that in Ben's here. There's
a cab over there has a medial /b/ like that of Want a cab? Medially, then, there
are three kinds of voiced bilabial stop in contrast. We can assign the evenness of
voicing of the /b/ of He's a cabby to the environment rather than to the phonemic

unit /b/ the preceding and following sounds are of the always-voiced class in
English. The crescendo quality of the voicing of the /b/ of Ben's here, and the
diminuendo quality of the voicing of the /b/ of Want a cab?, can likewise be
assigned to environment, since in the first case the /b/ is preceded by silence,
in the second case it is followed by silence. But if we want to extract the crescendo
of the voicing of the /b/ of / see Ben from the /b/ itself and assign it to the
environment, to what part of the environment does it belong? It cannot simply

be assigned to the preceding /iy/, since in rebus one has the fully voiced /b/
after /iy/. The only thing to do with the crescendo, if we don't want it to be a
part of the /b/ itself, is to set it up in its own right as an element environ-

mental to the /b/. The same argument applies to the diminuendo of the
voicing of the /b/ in There's a cab over there.
Suppose we decide to write crescendo-voicing /b/ as /-b/, diminuendo-voicing
/b/ as /b-/. In macrosegment initial and final we can omit the hyphen, since
there is no contrast with anything else. But medially we will have to write it
wherever our definition calls for it. If the three-way contrast for medial /b/ were
the only set of features in the system which could be handled in this way, we
would have accomplished nothing. But since medial /d/ and /g/, and others,
manifest this same three-way contrast, and since many other medial elements
manifest phonetically different but distributionally comparable contrasts, we
can accomplish a great deal by extracting a large number of different features
all on the same basis of comparison with utterance-terminal phenomena and
setting them up as different allophones of a single structural unit: the English
juncture.
It is also relevant that a great many, though not all, of the segments which
thus come to be transcribed between successive occurrences of /-/ (or between
macrosegment-initial and a /-/, or /-/ and macrosegment-final) are segments
which occur, save for intonation, as whole macrosegments, including precisely
the features indicated by the hyphen. One can say / see Ben; one can also say
 A MANUAL OF PHONOLOGY 171

just Ben. One can say There's a cab over there; one can also say just There's a cab,
or just cab. The correlation is not complete, but unless there were some correlation
there would be little point in the analysis.
We said above that if the three-way contrast for medial /b/ were the only one
capable of being handled juncturally in English there would likewise be little
A situation of this kind is found in some varieties of Carib-
point in the process.
bean Spanish, /n/ and /rj/ contrast in these varieties of Spanish only between
vowels, and whenever /c/ occurs before a vowel it happens that what follows the

/rj/ also occurs as a whole macrosegment; where /n/ precedes a vowel, on the
other hand, this is never true. One might propose that /i)/ be reinterpreted as
/n-/, where the /-/ is defined as "dorso-velar position." Segments between
successive /-/'s would then be isolable, as in the case of similarly bound segments
(with a differently denned /-/) in English. But there would be no gain: one would
in fact be stating the same situation with a different symbolism, which, because
it is based only on a reanalysis of prevocalic /n/ and /rj/, could be stated just

as well in terms of /n/ and /rj/ themselves.

In other words, juncture phonemes achieve their power precisely because
of their phonetic heterogeneity. When we say of English that segments bounded
by /-/ are in general isolable, we are covering a large number of specific facts : a
segment which begins with crescendo voicing on a voiced stop or spirant, with
clear voicing on a sonorant, with slight glottal catch (or at least clear point of
syllable division) before a vowel, with a voiceless stop which is well aspirated,
and so forth, and which ends with diminuendo voicing on a voiced stop or spirant,
with drawl on a stressed vowel, and so forth, is in general isolable. It takes longer
to describe all the allophones of a juncture, but once the juncture has been de-
scribed it constitutes a powerful tool.
It should also be noted that the heterogeneity of the allophones of a juncture
renders possible a much neater phonetic layout for "ordinary" phonemes, be-
cause one has withdrawn from the latter many of the messy marginal differences
which would otherwise yield a very complex system. A /b/ which has to be
denned as sometimes evenly voiced, sometimes having crescendo voicing, and
sometimes having diminuendo voicing is a heterogeneous kind of a /b/. A /b/
which can be defined as always involving bilabial closure, velic closure, and
some voicing (in contrast to /p/), is homogeneous. For that matter, if one were
to reject junctural analysis in English, there could be no recourse to a hetero-
geneous /b/: one would have to establish at least evenly voiced /b/, crescendo-
voiced /-b/, and diminuendo-voiced /b-/ as three different phonemes.
There is one way of speaking of juncture which retains phonetic homogeneity
of all its allophones. This is Harris's way. Harris sets up a juncture as a "zero"
phoneme a phoneme having no phonetic properties at all (and, because of this,
having identical phonetic properties in all environments). The only function of
the "zero" phoneme, then, is to function as environment for ordinary phonemes:
English /b/ is represented by different allophones when flanked by this "zero"
phoneme and when not so flanked, and by different allophones depending on
whether the "zero" phoneme precedes or follows. This seems to the present
172 A MANUAL OF PHONOLOGY

writer a most unfortunate and misleading kind of hocus-pocus; he feels that

setting up junctures as always involving identifiable phonetic material, no
matter how diverse, is much better. This choice is directly in line with the pro-
cedure recommended in 3222 for handling similar phones which are in comple-
mentary distribution (that is, removing part of one phone and assigning that part
to the environment, so that what is left is identical with the other phone in
question).
There is one potential source of error which ought to be avoided. Juncture
phonemes are not recognized in order to show grammatic boundaries of one or

another kind say boundaries between words. It often happens that grammatic
boundaries fall at open junctures, particularly grammatic word boundaries.
It is easy to see why this should be so. Grammatic words are by definition ("mini-
mum free form" or the like) units which are apt to occur in various different posi-
tions in utterances, including initial and final position, and in some languages
grammatic words retain much the same phonologic shape regardless of their
position (in other languages this is not true). It would be very peculiar indeed
if the purely phonologic procedures for the determination of juncture which

we have described did not reveal them at a good many such grammatic bound-
aries. But one cannot reverse this and simply assume that at grammatic bound-
aries of some sort there will necessarily be phenomena subject to junctural
treatment. The Fox case shows this. The long phrase we cited earlier contains
four words, hence three phrase-medial word boundaries, but nothing happens at
those boundaries any different phonologically from what happens in word-
interiors. In English there is some correlation, but it is not complete. On the
one hand, whereas she and is are two grammatic words, even when the latter
appears in shortened form, there is normally no juncture between the two in
such a phrase as She's a friend of mine; find her in normal speech is homophonous
with finder, once again with no juncture. On the other hand, some people pro-

nounce Cato and Plato with juncture between the a and the t a drawl on the
/ey/ and a fully aspirated /t/, not the quick voiced-flap allophone found in
matter. Cato and Plato are none the less single grammatic words, and perhaps
even single morphemes. Yet Cato and Plato share with such juncturally seg-
mented stretches as padlock and playbill the fact that the phonologic segments
separated by the junctures can (with suitable adjustment of stress-level and
intonation) stand as whole macrosegments Kay, play, toe, pad, lock, Bill. Even
:

this fails in the case of the isolated microsegment /z/ of John's going, which,
given the definition of open juncture for English which our operations and the
facts of the language force on us, has to be taken as both preceded and followed
by /-/

325. Interrelationships Among Ultimate Phonologic Constituents. There is as

yet no general agreement on the proper way to decompose phonemes, junctures,
accents,
and intonational features into ultimate phonologic constituents indeed,
there is not even complete agreement that we should do so at all. Several alterna-
tive approaches were demonstrated or referred to in 262. Jakobson has used
what is, at least in theory, a still different approach, based seemingly on impres-
 A MANUAL OF PHONOLOGY 173

sionistic acoustics. Harris has tried a technique which keeps only the vaguest
tiewith either acoustics or articulation. Our purpose here is not to legislate a
choice between the possible procedures, but rather to show certain problems of an
empiric and a logical sort which inevitably turn up when any of the possible pro-
cedures is followed. For this purpose, it will be feasible for us to keep
to a relatively simple-minded sort of decomposition based firmly on articulation.
One of the problems is that of "determining" and "determined."
In French (2521341) there are twelve obstruents: /ptkbdgfsvz2/-
It is tempting to regard each of these as a bundle of three coequal ultimate con-
stituents: a voicing-term (voiceless or voiced), an occlusion term (stop or spirant),
and one of three positions (say front, central, and back). This requires seven
ultimate constituents in all, and is in a purely logical sense quite impeccable. We
can even redefine "front" in articulatory terms as "involving lower lip as articu-

lator": then when "front" and "stop" co-occur, the upper Up is predictably the
point of articulation, while when "front" and "spirant" co-occur, the upper teeth
are predictably the point of articulation. "Back" cannot be redefined quite so
simply in articulatory terms, but logically it is perfectly possible to say that the
combination of "back" and "stop" produces dorso-velar, while the combination
of "back" and "spirant" requires lamino-alveolar. It is possible seemingly to
reinforce theargument by a claim that in some vague acoustic sense (probably
purely impressionistic), /t/is to /k/ as /s/ is to // But this impressionistic-

acoustic claim will be made only within the phonologic economy of French; it
will not be made, say, for German, where obviously one must say that /t/ is
to /k/ as /s/ is to /x/, leaving // aside.
But there is another possible decomposition of French obstruents which is
equally impeccable from a purely logistic point of view. This lines the obstruents
up as follows:

p f t s k
b v d z z g.

Each obstruent is here regarded as a bundle of just two coequal ultimate con-
stituents: a voicing term and one of six positions. The latter are bilabial, labio-

dental, apico-dental, apico-dental rill, lamino-alveolar, and dorso-velar. The

contrast between stop and spirant regarded as determined rather than
is now
determining: if "bilabial," "apico-dental," or "dorso-velar" is involved in a
bundle, then, automatically, one has "stop," whereas otherwise one equally
automatically has "spirant." True enough, this interpretation requires eight
ultimate constituents instead of seven, but that is a small difference, and the
portrayal is just as realistic as the first; furthermore, if the roster of ultimate
constituents is larger, the number which participate in each obstruent is smaller.
Both of these decompositions of the French obstruents have the odor of pure
game-playing, an odor which seemingly appetizing to some linguists. True
is

enough, in each treatment we retain complete predictability of the actual relevant

phonetic nature of the several obstruents (in the first, "front" and "stop" imply
bilabial, whereas in the second, "bilabial" implies "stop"). But with such reten-
174 A MANUAL OF PHONOLOGY

tion as a sole guide in decomposition, it is possible to be far more drastic, in a

way which guarantees an absolute minimum of determining "features" for any
system one may confront. Suppose that there are a total of sixteen phonemes, a
through p. We list them in any order we wish, in a table like the following:

a b c d e / 9 h i j k I rr i o p
A X X X X X X X X
B X X X X X X X X
C X X X X X X X
D X X X X X X X

The four "determining features" are A, B, C, and D. Each of the six-

teen phonemes is now a bundle
from none to all of these four features. All of
of
the actual articulatory features of all the phonemes are "determined." Thus
if a is a low central vowel, then we say that these phonetic characteristics are

automatically present whenever determining features A, B, C, and D are all

present in a bundle; if, on the other hand, a is a voiced dorso- velar spirant, we
say that these phonetic features are the automatic product of the fourfold con-
catenation of determining features. There are 16! different ways in which we can
list the 16 phonemes of the system across the top; there are therefore 16! dif-
ferent ways in which the system can be decomposed in terms of just four deter-
mining features.
In general, let a system involve n phonemes. Then let k be the integer such that
2 is the smallest integral power of 2 not less than n. (Thus if n is 9, 10, ... 16, ,

k is 4; if n is 17, 18, 32, k is 5; and so on.) The system can then be decom-
. . . ,

posed in terms of as few as k different "determining features," but not in terms

of any smaller number; and there are at least n\ different ways of doing it.

In other words, any phonemic system and, indeed, any finite set of things

whatsoever can be arbitrarily converted to a binary code, and there is at least
one such conversion which requires a minimum number of binary digits for each
element of the set.
The above is, I believe, a psychologic reductio ad absurdum. It should show
us all that we are not properly concerned, in phonologic analysis, with any sort
of purely arbitrary decomposition; indeed, we should like insofar as possible
to avoid such arbitrariness altogether. The opposite of this sort of game-playing
is what I mean by "hugging the phonetic ground closely."

And if we hug the phonetic ground closely, then neither of the portrayals of
French obstruents given above can be accepted. The interpretation offered in
2521341 seems to be about as far as we can go.
Furthermore, it turns out that in general we cannot divide the ostensible
ultimate phonologic constituents of a system neatly into "determining" and
"determined," assigning the latter some sort of secondary status. In the actual
complexity of speech, a given feature or difference turns up in some contexts as
of primary relevance, in other contexts as subsidiary. We are forced by the lack
of balance and neatness in the average language to recognize a larger number of
ultimate phonologic constituents than would be called for in a purely arbitrary
 A MANUAL OF PHONOLOGY 175

binary coding, and these features are more rigorously and arbitrarily limited in
their privileges of occurrence relative to each other than would be the purely
abstract ones set up by a binary coding.
Thus, for French obstruents, we have no choice but to recognize (1) two
voicing terms; (2) two occlusion terms; (3) six combinations of articulator, point

and contour of articulator ten features in all. Theoretically
of articulation,
thiswould yield twenty-four distinct bundles; in fact it yields only half that
number. The non-occurrence of bilabial spirants and labio-dental stops does not
mean either that the choice of position of articulation governs that of occlusion
term or the reverse; it is simply a limitation on privilege of occurrence.
The other matter which must be discussed is that of primary and secondary
features. This is closely related to the problem of determined and determining,
but not quite the same. Let us consider a typical 2+2+1 vowel system, in which
the second dimension of contrast is primarily unrounded versus rounded: the
/u/ and /o/ are always rounded, but in some environments are relatively front.
Possibly the fronting environments are such that we can perform the sort of
resurveying of boundaries (between item and environment) described in 3222;
for the moment this does not matter. Generally, the occasional fronting of /u/
and /o/ would either not be mentioned at all, or it would be described as a
"secondary" feature (or even as "nonphonemic"). Now the problem is this:
would we be justified in ignoring the fronting altogether? Is it proper to exclude
such "secondary" features from a strictly structural account?
Note that, by our description, there are probably some environments in which
/i/ and /u/ differ both as to rounding and as to tongue frontness and backness.
In such environments, it may sometimes be the backness, rather than the round-
ing, of an /u/ which tells a hearer that he has heard an /u/ rather than an /i/;
it may sometimes be the frontness, rather than the absence of rounding, which

tells a hearer that he has heard an /i/ rather than an /u/. Granting that this

can occur only in some of the environments in which /i/ and /u/ are found,
it would seem quite arbitrary of us, as analysts, to leave out of our account

articulatory facts which the speakers of the language quite obviously do not
ignore. It is most certainly appropriate for us to mention the constant factor
first (the lip position), but we should include the occasional factor too.
Another good example of this is the three stop-types of Georgian (2521333).

The plain (unmarked) series lack both voice and aspiration this is the primary
feature. They are often, though not always (and not just in specifically de-
limitable environments), glottalized. This is a secondary feature, but it deserves
mention.
Of course it is easy to miss some relatively subsidiary secondary features,
and this obviously impairs our description of the phonologic system of a language
less than it would to miss some clearly primary feature. But we should not in-
tentionally overlook them.

4. Structure, Pattern, and Abstraction. Any utterance in a given language has a

phonologic structure : it ICs in a given arrangement those ICs in turn
consists of ;

consist of smaller ICs in a given arrangement and so on until one reaches ultimate
;
176 A MANUAL OF PHONOLOGY

phonologic constituents. Any utterance, or IC of an utterance, or IC of an IC,

and so on, is a phonologic element; any phonologic element either has a phonologic
structure (in terms of its own ICs) or is a single ultimate constituent. An ar-
rangement of phonologic material which is not a phonologic element does not
have a phonologic structure, even though it may occur. Thus the sequence /yt/
occurs in English, for example in bait; but /yt/ is not an element and has no
structure, since the ICs of bait are /b/ and /eyt/, and the ICs of the latter are
in turn /ey/ and /t/.
The above describes, in a summary way, our operating assumptions in the
collation part of phonologic analysis; it does not imply that we can always decide
unambiguously how to cut a given element into its ICs, or even that there always
exists an unambiguous way if only we could find it. In 2 and 3 we saw many
cases of non-uniqueness, only some of which responded to our efforts to resolve
the conflict between different bits of evidence.
The phonologic pattern (or system) of a language consists of a stock of ultimate
phonologic constituents and a set of (hierarchic) arrangements in which they
occur in one or another utterance. A given phonologic element (if not ultimate)
has a structure; it conforms to the pattern. To one who has analyzed the system,
structure is directly observable; pattern can only be deduced from the comparison
of the structures of actual utterances. Pattern is habit, structure is behavior;
pattern is langue, structure is parole; pattern is culture, structure is behavioral
manifestation of culture. These successive statements are intended as approxi-
mately synonymous, and the reader can choose for himself which one, in terms
of his own background, is most meaningful. Another statement of the same kind:
structure is the spatio-temporal arrangement of the parts of a specific historical
event; pattern is that to which structure more or less closely conforms. A dress
has a structure different pieces of cloth fastened together in a certain way; the
"dress pattern" is a set of pieces of paper with a similar structure; the structure

of the paper and of the cloth both conform to a single pattern in our sense.
The aim of the investigator in phonologic work is to discover and describe
the phonologic pattern of a language. This is an empiric task, as we have insisted
time and again: he must base his judgments on observations of actual speech
(observations made by himself or by others) ; insofar as he can, he must identify
and distinguish as the speakers themselves do; he must describe in terms of
actual articulatory events, hugging the phonetic ground closely; if in the field,
he must make predictions on the basis of partly collated evidence and change his
portrayal if the predictions are not born out.
In due time, however, it becomes both possible and useful (and, as we shall
presently show, necessary) to abstract the phonologic pattern, as a system, from
the articulatory and partly acoustic framework in which the observed events
occurred. Or, to express this in another way, it is worthwhile to invent a com-
pletely abstract mathematical system, manipulable in completely abstract
mathematical terms, but which can be regarded as exemplified by the phonologic
pattern in question. We shall first show more fully what it means to do this, and
the difficulties which are encountered, and will then show why the operation
is useful.
 A MANUAL OF PHONOLOGY 177

An abstract mathematical system consists of a set of elements, of completely

unspecified characteristics, but bearing certain specified abstract relationships to
each other. The set of all positive integers constitutes such a system. Any two
different elements of this set bear an order relationship: if a and b are elements,
then either a is greater than b or vice versa. Some ordered sets of three elements
of the whole set (not necessarily all different) stand in the relationship that the
sum and second is the third; if this is true of a, b, and c, then it is also
of the first
true of b, a, and c. Given any two elements, a and b, then one can always find
an element c such that this relationship holds between a, b, and c. All of these
relationships hold in various abstract sets, not only in the positive integers,
but it is possible to establish a set of relations such that any set of elements for
which they hold are either precisely the positive integers or else "might
as well be."
As for abstraction: two oranges and two more oranges make four oranges;
two apples and two apples make four apples; two oranges and two apples make
four pieces of fruit. When, in grade school, we come to the point of saying simply
"two plus two is four," without bothering to mention apples, fingers, oranges,
or the like, we have abstracted. There is no implication that a statement like
"two plus two is four" has any truth or meaning in its bare form: it is meaning-
less, in a sense, until we reinsert apples, fingers, oranges, or something else. But

the abstract statement has a kind of derivative truth, or validity, in that if prop-
erly expanded into a reference to the actual world about us, it will be true, and
in that expansions which will render it true are indeed to be found. This is what
is meant by finding an exemplification of an abstract system: one for which no

exemplification, by definition, can be found, either in the world about us or

elsewhere in mathematics, is quite empty.
To abstract a phonologic pattern from the "phonic substance" in which it is
made manifest in speech, we set up a number of abstract elements, paired off
with the ultimate phonologic constituents of the phonologic pattern; or, if we
prefer, we may pair them off with phonologic elements larger than ultimate
phonologic constituents
say with phonemes, junctures, and the like. The rela-
tions which these abstract elements bear to each other are then similarly to be
abstracted from the relations manifested by the phonologic elements: a and b
both occur in environments E h E2 ,
En though not at the same time; c and d
. . . , ,

do not both occur in the same environment; e occurs before /; g does not occur
before h; and so on. The abstract system which is constructed in such a manner
for a given phonologic system will not bear much similarity to any of the systems
with which mathematicians usually concern themselves, but that does not
controvert their mathematical nature: mathematicians as a general rule confine
their attentions to systems of a considerable degree of homogeneity, containing
large, even transfinite, numbers of elements which "work" in much the same
way, while a phonologic system is, comparatively speaking, highly heterogeneous.
Unfortunately it is possible to devise more than one mathematical system
on the basis of any single phonologic system. This is in part true because of the
indeterminacies, of various kinds and at various IC-levels, which we have al-
ready seen in numerous instances. We are forced to say that any two systematiza-
178 A MANUAL OF PHONOLOGY

tions of a single phonology, so long a they subsume in one way or another exactly
the same totality of phonetic facts, bear a certain kind of equivalence to each
other; we shall call this kind of equivalence mutual convertibility. An analysis
which treats Fox /6/ as composed of simultaneous laminal position and stop
closure is mutually convertible with one which treats it as simultaneous cluster
of ft/ and /y/ or, rather, can be mutually convertible therewith, providing

in each case other matters are adequately covered. An analysis of English based
on an opinion that initial voiceless stops are not aspirated is not mutually con-
vertible with one based on an opinion that they are: here there is disagreement
as to fact, not just as to collation.
The equivalence-relation of mutual convertibility carries over to the abstract
mathematical systems based on phonologic patterns: we must say that any two
mathematical systems which are exemplified in mutually convertible systematiza-
tions of a single phonologic pattern are likewise mutually convertible. Now on
the abstract mathematical level it should be possible to analyze and describe
precisely what constitutes "mutual convertibility." This has not yet been done,
and it is not a simple task, largely because the systems involved are complex
and finite. But it probably can be done, and the performance of this task should
show us a great deal more clearly than we can now see just what kinds of differ-
ences there can be between different systematizations of a single phonologic
pattern. Furthermore, if an ensemble of systems are related by some equivalence-
relation, such as mutual convertibility, then there exists some abstract system,
in the ensemble or not, composed of all those elements and relations which are
invariant under any transformation from one of the systems to another. This
invariant underlying system, when it can be discovered for a given language,
will be the most accurate description obtainable of the phonology of that lan-
guage, by definition maximally free from artifact-like results of the esthetic
preferences of one or another investigator.
This in itself constitutes one of the large reasons for undertaking abstraction,
and we may hope that in the next period of years enough work will be done to
render phonologic work far less subject to personal whim than it is as yet. But
there are several other reasons for abstracting.
One reason is because we cannot perform the operations of collation at all
we do. The articulatory motions which constitute speech, and the auditory
unless
behavior which constitutes hearing speech, cannot themselves be manipulated
and displayed. Collation has to be performed on some sort of a symbolic record
of actual speech behavior, on the sheets and file slips of the investigator's notes.
There are no phonologic forms in this book there are only marks on paper which
:

represent them. If we are going to talk about a phonologic system with a set of
marks on paper assigned in some way to types of articulatory event in speech,
then those marks on paper must in some way be defined as bearing relationships
to each other abstractly like those between the articulatory events they represent.
For greatest accuracy, the phonologic pattern must be manifested not only in
"phonic substance" as speakers of the language speak, but also in the "graphic
substance" which
we use in describing the language or more particularly in
transcribing words, utterances, and longer texts.
 A MANUAL OF PHONOLOGY 179

The very act of undertaking to analyze and describe a phonologic system

therefore implies that we have assumed that the kind of abstraction we are
talking about isboth possible and necessary.
Finally, it must be recognized that the phonologic pattern of a language
manifests itself, quite apart from the observations of any investigator, not
only in articulatory motions, but also in characteristics of the sound-waves pro-
duced thereby, in the ear and associated tracts of the nervous system of a hearer,
and, indeed, in the nervous system of the speaker as he speaks. We cannot assume
any very great similarity between the articulatory system and the shape events
take in these other media, but if a hearer understands what a speaker says if,
in particular, under good hearing conditions he can tell exactly what the speaker

has said, phonologic unit by phonologic unit then all the phonologic structure
in the articulatory motions must also be represented somehow in the sound
waves, and similarly it must all be in the nervous system. Now one cannot discern
such matters as "dorsal," "apical," "voiced," and the sound waves or in
like in
nervous currents. Nervous currents are homogeneous in kind, and differ only in
their relative spatial and temporal location within the central nervous system.
The only way in which a phonologic system can be represented in these media
as well as in articulation is for the system itself to be abstract.

An analogy will be useful. The unit symbols of the Morse Code, as used in
old-fashioned telegraphy, consist of short voltage pulses ("dots"), long voltage
pulses ("dashes"), and silences of two or three distinct durations. One can trans-
mit messages, using exactly the same code, with waves of a flag: in one direction
for "dot," in another direction for "dash," and in some third direction, once,
twice, or three times, for the "silences." Or one can transmit messages in exactly
the same code using dots and hyphens and spaces written from left to right on
paper. There isno physical similarity between a dot on a telegraph wire, a dot
in flag-waving, and a dot on paper, but the code is exactly the same regardless of
the medium used. Abstractly, a dot in Morse Code can only be defined as a unit
which contrasts with a dash or a silence. A unit Morse Code symbol, or an ulti-
mate phonologic constituent, is nothing in itself: it can be defined only as some-
thing different from the various units which which it stands in contrast or in
other relationship.
In other words, the phonologic 'pattern of a language (in contrast to the phono-
logic structure of specific elements as they occur) IS an abstract system. The main
reason for abstraction away from the articulatory medium is that one cannot
describe the phonologic pattern, in the last analysis, by any other means. We were
wrong, earlier in this section, in referring to phonologic systems and matching
abstract mathematical systems as two different things; we did so for a heuristic
reason, but must now conclude that they are one and the same.
We pointed out in 04 that the terminology of phonologic description, while
largely articulatory, is not completely so, being in part acoustic
("high" and

"low" with reference to pitch) and in part of the "imitation-label" variety; and
we indicated that this mixed frame of reference did not impair results for most
before
purposes. We can now see why the latter statement is true: it is because,
we complete the portrayal of a phonologic pattern, we have abstracted the
180 A MANUAL OF PHONOLOGY

elements and relations in from the primarily articulatory frame of reference

it

used in making our initial When we begin work on a system, the

observations.
symbol [m] refers to actual articulatory events involving lip closure and open
velic, the symbol [a] to articulatory events which we describe perhaps with
only "imitation-label" validity in a certain way in articulatory terms. When
we have completed our analysis, we may symbols /m/
find ourselves using the
and /a/ (now bracketed between slant lines), and it may well be that each of
these symbols refers back to certain recurrent articulatory events, whether we
can describe those events with any complete physical accuracy or not; what
counts when this point is reached, however, is no longer the physical manifesta-
tion or "exemplification" of the elements symbolized by /m/ and /a/, but only
their abstract relations to each other and to other elements within the abstracted
phonologic pattern. Theoretically we should be able to derive the same pattern
by tapping the speech-communication channel at some other point say in
transit, as modulated sound, between speaker and hearer; or somewhere in the
central nervous system of a speaker or hearer, as spurts of nervous energy. If
we could, then our physical descriptions of /m/ and /a/ would be very different,
but the network of abstract relationships between them and the other elements
in the pattern would presumably be the same. We do not, in fact, know how to
tap the channel in any way other than that currently used, but in one sense this
is irrelevant or accidental.
In another sense it is not. It is foolish to think that, just because we must
abstract a phonologic system from the articulatory-acoustic substance in which
it is made
manifest, therefore the former is our sole object of interest and the
latter beneath our dignity. The analyst is not the only person who must start,
is

at least, with the overtly observable speech signal and motions of articulation. A
child has this same point of departure in learning a language; an adult has this
same point of departure in learning a new language. No comparison of abstract
phonologic systems will tell us anything about the probable course of events
when a speaker of language A is exposed to (and may have to learn some of)
language B. For this we must make a careful comparison of the physical mani-
festations of the phonologic units of each language. Russian /p/ is abstractly not
comparable with English /p/, because the two are defined by different inter-
locking sets of contrasts. But if a speaker of English hears a Russian word be-
ginning with /p/, he will none the less interpret it as though it were an occurrence
of his own English /p/. Abstraction must add; it must not leave behind.

5. Acoustic Phonetics. Our aim in this section is not to survey exhaustively all

that has been accomplished so far in acoustic phonetics, but merely to define
the field precisely in its relation to phonology, and to supply a frame of reference
in terms of which the results of experiments in acoustic phonetics can be sensibly
interpreted. The aim should be clearly noted. The acoustics
limitations of this
of speech is studied from a number of angles and with a number of purposes in
view, of which acoustic phonetics is only one. Sometimes work directed towards
some other goal is helpful to us, and sometimes it is not. Conversely, the phonolo-
gist sometimes can give useful advice to the acoustician and sometimes cannot.
 A MANUAL OF PHONOLOGY 181

It is a mistake to think that merely because telephone engineers to take one

example are concerned with the transmission of speech, they need help from
specialists in linguistics. Both linguists and engineers have made this mistake,
and in some instances it has led to wasteful research programs.
51. Sound and its Physical Analysis. Sound is a form of energy; acoustics is

the branch of physics devoted to its study. The sound produced by a speaking
human the speech signalcontains no physical ingredient not found in other
sound: the techniques which acousticians have developed for the observation
and description of sound in general can be applied to the speech signal, and there
is no a priori reason why their investigation should miss anything essential in it.

The proper terminology for use in the study of the speech signal is the terminology
of acoustics as a branch of physics. We can add certain terms which seem par-
ticularly useful for our special interest (as over against, say, the study of bird
calls or sonar design), providing that any new terms are hooked firmly, by meas-
urement and demonstration, into the general terminology of acoustics just
exactly as, in the study of articulation, we can use the well-defined term "velic"
despite its non-use by anatomists. But we cannot with impunity invent any
fanciful impressionistic terminology and use it in place of the physicist's technical
vocabulary.
511. The Physical Dimensions of Sound. When, in the course of phonologic

work, we call, say, a voiceless dorso-velar stop a "speech sound" or simply a

"sound," we are indulging in a convenient but inaccurate figure of speech. We
might describe this figure as a kind of metonymy: "the source for the product."
We use the same figure in everyday life. A particular sound from the next room
will evoke the "I hear someone striking a match," rather than a de-
comment
scription in terms of frequency, amplitude, harmonic structure, duration, and
phase. It would be ridiculous to scorn this useful metonymy in either context,
but when we propose to describe sound itself, we must carefully eschew it.
Sound is vibration, at the molar level, in a physical medium, of which the
most important for our purposes is the air. The vibration of a source of sound
sets the adjacent particles of air into vibration; these in turn jog those a bit
further from the source; and so on; so that the sound travels in all directions from
the source as waves, at approximately 1100 feet per second, their strength di-
minishing as the distance from the source increases. The familiar inverse-square
law governs the decrease of energy-level of such energy spreading out unguided
in all directions; besides, some energy is lost by conversion into heat,
so that the

overall swifter than the inverse-square law specifies. On the other

weakening is

hand, the inverse-square law does not apply to sound that is confined and guided,
as within a stethoscope tube.
the vibrations are regularly spaced in time, we hear a musical
If
note of a

definite pitch: the more rapid the vibrations (the higher the frequency, measured
musical pitch. Doubling the frequency
in cycles per second, "cps"), the higher the
raises the pitch one octave. The range of frequencies on a
piano tuned to concert

pitch is from 27.5 cps to 4184 cps, which is The human

just over seven octaves.

ear hears lower sounds down to 15 or 20 cps providing they are loud
182 A MANUAL OF PHONOLOGY

20 0 (00 2.00 H60 tofe 10*0 WM 10000

FREQUENCY
Fig. 13. Relation of Perceived Pitch to Frequency
The perceptual distance between two tones (as determined by psycho-acoustic experi-
ments) of specified frequencies is proportional to the scale given on the right; "mels" are
arbitrary units of measurement of perceived pitch, so designed that a 1000 cps tone has the
pitch 1000 mels and a zero-mel tone is approximately at the lowest audible frequency.
Redrawn from Stevens 40.

enough, and also somewhat higher sounds up to 15 or 20 thousand cpswith

the same proviso. The total range of a good ear is thus between ten and eleven
octaves.
In our culture, at least in musical contexts, we are trained to hear octaves as
equal intervals. Laboratory tests of pitch-differences have also shown that
octaves are about equally large to untrained listeners over the middle of the
range of audible frequencies, but that they sound smaller near each end of the
range. Figure 13 demonstrates this. Since our crucial statements about frequency
perceptions will apply to the middle of the range, say from 200 cps to 4000
all

be proper to say that the perceptual distance between two fre-

cps or so, it will
quencies is proportional to their musical distance (measured in octaves and in
fractions of an octave such as semitones) and this musical distance is of course
;

proportional to the logarithm of the frequency ratio that is, to the distance
shown when the frequencies are plotted thus on a logarithmic scale. Figure 14
shows the difference between plotting frequencies linearly and logarithmically.
When the vibrations are not regularly spaced in time, we hear not a musical
note but a noise of some sort. (This is the everyday sense of the word "noise,"
not to be confused with its special sense in communication theory, as in 02
 A MANUAL OF PHONOLOGY 183

LINEAR

S V
\ \ \ \ \ \ V N \ \ \ .

i\%%
l

i s\ S\ iv *\ ix i> * v ix %v l i & i!

\ N \ \ \ V N \ \ \ \ \ \ x
I

V
x \ s \ v V \ \ \ \ \
106ARITHWIC
V
^- ^
\
>^. ^ 'V y^~
\ \ *

-*"

Fig. 14. Linear versus Logarithmic Plotting

Each brace On the logarithmic scale octaves plot into
encloses a one-octave interval.
line-s jgments of equal length; on the linear scale they do not.

and 6). Even noises, however, can often be classed roughly as higher or lower in
pitch, as
when we assert quite correctly from the physical point of view that
the sound of a hammer striking a nail is higher in pitch than the sound of surf
booming along a beach.
More technically, in a musical note almost all the energy is concentrated at a
set of frequencies which are integral multiples of the basic frequency called
the fundamental. If the fundamental is, say, 100 cps, then the energy will be
concentrated at 100 cps, 200 cps (the second harmonic; the expression
"firstharmonic" is not used), 300 cps (the third harmonic), and so on. There need
not be energy at all these frequencies, but it is essential that there be virtually
none at, say, 150 cps or 220 cps. If the relative concentration at the different
frequencies is altered, the difference is perceived as a change in tone quality. If
the energy is all concentrated at a single frequency, the result is a pure tone.

In a non-musical noise, on the other hand, the energy is scattered more ir-
regularly through a band of frequencies. If there is energy throughout the band
from 100 cps to 300 cps, one hears a noise of indefinite but nevertheless relatively
low pitch; if there is energy throughout the band from 1000 cps to 3000 cps, one
hears a noise of indefinite but nevertheless relatively higher pitch. If the energy
is spread evenly through all audible frequencies, the sound is called white noise.
At a given frequency, a greater amount of energy produces greater amplitude,
and, other things being equal, the sound is louder than that involving a lower
energy-level. Instead of speaking of amplitude, we can alternatively speak of
intensity,which is by definition proportional to the square of the amplitude.
Suppose that a violin string of fixed length vibrates in such a way that the point
halfway between the ends passes back and forth through an arc one sixteenth
of an inch in length. Then the sound will have twice the amplitude, and four
times the intensity, that it would have if the arc were only one thirty-second of
an inch in length.
We see from the above discussion that the physical dimensions of sound, in
184 A MANUAL OF PHONOLOGY

addition to duration, are frequency and amplitude or intensity; any other property
of some sound can be expressed as a function of these three. There is
specific
one additional independent factor, phase-relationship, with which it will be more
convenient to deal a bit later; but there is much evidence to show that human
ears do not perceive differences of phase-relationship, so that it can hardly play
any relevant role in the speech signal. For the sake of completeness, we had
better mention also the three spatial dimensions in terms of which the source of a

sound can be located especially since binaural hearing does make it possible
sometimes for us to localize the source of a sound, or at least its direction, without
the evidence of other senses. But it is quite obvious that nothing further need
be said along this line for our purposes.
512. Laboratory Apparatus for the Study of Sound. The acoustician can make
"direct" observations of sound only with approximately the same apparatus
which is built into all of us. For his purposes such observations are not suffi-
ciently precise. Actually they are not "direct" at all, in one sense, since between
the sound which impinges on his ears and the statements which he can make

Fig. 15. Idealized Oscillograms

B lasts longer than A; C has greater amplitude than A; D has higher frequency than A.
E and F differ from each other, and from A, B, C, and D, in harmonic structure.
 A MANUAL OF PHONOLOGY 185

about it there functions a complex, linguistically and culturally conditioned,

apparatus which distorts his report in unknown ways the common metonymy
of "source for the product" playing a prominent but variable role. Therefore he
turns to laboratory apparatus which will produce a visible record, of one sort or
another, of the sound, and which can be calibrated with at least somewhat less
difficulty than one can calibrate human beings.
The two chief devices currently used for acoustic study (at least of the speech
signal) are the oscillograph, which dates back a number of decades, and the
spectrograph, which was developed during World War II.
The oscillograph produces a visible record called an oscillogram which is, to
within a measurable degree of accuracy, a kind of facsimile of the actual wave-
motion which constitutes the sound. If the sound is not too complex, then it is
relatively easy to measure off frequency, amplitude, and something of the har-
monic structure. In Figure 15 we show, in a somewhat idealized way, how differ-
ences of frequency, amplitude, and harmonic structure appear on an oscillogram.
Differences of phase- relationship also show up clearly in simpler cases: Figure
16 shows how, and at the same time will serve to explain what is meant by phase.

Fig. 16. Difference of Phase-Relationship

A and B
are both pure tones; D
is the same as B but "out of phase" that is, D would
have to be shifted to the left or right before it could be placed squarely over B. C is the
sum of A and B; E the sum of A and D. C and E therefore are identical in harmonic struc-
ture, but differ in phase-relationships.
186 A MANUAL OF PHONOLOGY

The spectrograph produces a which is proving

different sort of visible record,
to be more useful than that of the oscillograph formost aspects of analysis of
the speech signal. As in an oscillogram, a spectrogram shows time along the length
of the record, but here the resemblance ends. The up-and-down scale of a spec-
trogram shows frequency, so that the spectrogram does in a precise way what
printed music does irregularly.The amount of energy present at a given time
and frequency shows up as the degree of darkness of the spectrogram at the
corresponding point. Thus a spectrogram is a three-dimensional portrayal, al-
though the third dimension, shown by the light-dark scale, cannot be read very
accurately. Figure 17 shows, in a highly stylized way, how differences of dura-

Fig. 17. Idealized Spectrogram

The spectrogram above shows exactly what the musical notation below does. The spectro-
gram shows difference of frequency by the height of the line; C is higher than A, B, or D.
It shows difference of duration by length of the line: B is twice as long
as A, C, or D. It
shows difference of amplitude by heaviness of the line: D is more intense than A, B, or
C. This last dimension is the hardest to read off accurately, just as in musical notation
the marking of "p" and "f" is least precise.
 A MANUAL OF PHONOLOGY 187

tion, frequency, and amplitude appear. Like the human ear, the spectrograph
does not respond to differences of phase-relationship.
Existing spectrographs will not produce a continuous record of a signal of
indefinite length such a machine is theoretically feasible, but the cost would be
prohibitive. Instead, the existing machines will record a signal lasting not longer
than two and four-tenths seconds, in two steps. First the signal is placed on a
circulating band of magnetic material within the machine. Then it is played
back repeatedly, within the machine, as a "band-pass" filter runs slowly up the
frequency scale. This filter passes only the energy within a very narrow band
of frequencies, and maps each narrow band in turn onto the spectrogram. Most
spectrographs have two alternative settings, one of which will plot in terms of
relatively narrower bands of frequencies, the other of which takes a broader band
at a time. The difference is shown in Figure 18.
A real oscillogram does not look too different from our idealized specimens in
Figures 15 and 16, but a real spectrogram of speech bears only the most super-
ficial resemblance to the stylized drawings in Figure 17. There are fines, hatchings,

and mottled or speckled areas all over the paper, and it requires considerable
skill (plus sometimes a bit of artistry) to discern what is relevant. To a large
extent, this fuzziness is the result of the awkward way in which energy-level is

represented by the light-dark scale. Some help is afforded by a special circuit

with which most spectrographs are now equipped, with which a different repre-
sentation can be achieved. For any selected short segment of the input (as re-
corded on the circulating magnetic band within the machine), this circuit pro-
duces a two-dimensional graph measuring energy-level against frequency. This is

Fig. 18. Difference in Width of "Band-Pass" Filter

The figure on the left is a segment of a spectrogram made with a "band-pass" filter which
passed a band 340 cps in width. That in the center was made with a filter passing a band
58 cps wide, and that on the right with a band passing 5.8 cps. All three are of exactly the
same segment of the same signal. Note that with the broadest band frequency is smeared
but time is precise, while with the narrowest band the time of a bit of energy is smeared
and frequency is precise. No physical operation can eliminate such smear, but we can choose
whether to smear time more and frequency less or vice versa. Reproduced, by permission,
from Gordon E. Peterson, "Parameter Relationships in the Portrayal of Signals with
Sound Spectrograph Techniques," Journal of Speech and Hearing Disorders 17.427-32,
1953.
188 A MANUAL OF PHONOLOGY

equivalent to taking a narrow vertical slice out of a spectrogram and replacing

the light-dark scale by a second dimension on the paper. The result is shown
schematically in Figure 19. By making such a representation for a large number
of successive narrow and measuring the resulting graphs carefully, it is
slices,

possible to build, out ofsome suitable material, a "spectrogram" in three spatial

dimensions, a third spatial dimension replacing the light-dark scale. Examples
will be found in Kersta; it would cost too much to reproduce a half-tone picture
of one here. Furthermore, the process of preparing such a "solid spectrogram"
is extremely tedious, and there is no reason to believe that it can yield results
that cannot be obtained more easily by other methods.
Whenever a relatively musical tone is involved in the input to a spectrograph,
the spectrogram shows a set of one or more approximately horizontal lines from
left to right, representing concentrations of energy at the various harmonics
of the tone. When the fundamental drops, the lines drop and move closer to-

Fig. 19. Energy-Level against Frequency

To
the left is shown (schematically) a short segment of a spectrogram. To the right is
a curve in which the intensity of the signal at a frequency is shown by the height of the
curve from the (vertical) base-line. Where the spectrogram is darker, the curve is higher,
and conversely. To make the curve look like such curves usually do, turn the page ninety
degrees counterclockwise.
 A MANUAL OF PHONOLOGY 189

Fig. 20. Harmonic Streaks and a Formant

gether; when the fundamental rises, the lines rise and move further apart. By-
careful measurement, under favorable conditions, it is possible to determine
the frequency of the fundamental at any point where these harmonic streaks
appear.
In addition to these, and often cutting across them, the eye seems to see
rather broader and less distinct dark bands, called formant lines. Closer scrutiny
reveals that these are not additional lines at all, but rather merely the result of
variations in darkness of the individual harmonic streaks (though if the latter
become dark and heavy enough, they may overlap and lose their distinctiveness).
Figure 20 shows, with some magnification, a sheaf of upwards-sweeping har-
monic streaks, each varying in darkness from point to point. When held at arm's
length, the darker segments will seem to coalesce into a dark band curving down-
wards from left to right. Acoustically, all this means is that the amount of energy
at a given time is concentrated more at certain harmonics than at others; a band
of frequencies within which the harmonics are relatively intense constitutes a
formant. A formant, thus, is not a fundamental feature of sound in addition to
those itemized in 511, but simply the relatively great intensity of the harmonics
at certain frequencies.
A third type of apparatus, developed very recently but already proving to be
invaluable, is the pattern playback. This device will "read" a (photographically
enlarged) spectrogram and produce the appropriate sound. Not all of the original
input to the spectrograph is thus read out: the pitch of the fundamental is lost,

so that speech comes out in a monotone. The greatest utility of this device has
been in psycho-acoustic experiments. Rather than using an actual spectrogram,
one can draw pseudo-spectrograms freehand, play them, and record the responses
of subjects. Thus one can eliminate one or another feature of what appears on
actual spectrograms, in order to test the relevance of what is retained.
 :

190 A MANUAL OF PHONOLOGY

52. The Problem of Acoustic Phonetics. The fundamental problem of acoustic

phonetics
the problem which defines the field can be described as follows:
given the phonologic structure of utterances in a certain language, what are the
acoustic correlates of the different phonologic elements?
This can be restated in terms of the discussion of 02 and the diagram in
Figure 1. In 02 we defined the phonetic system of a language as the code by

which an individual's Speech Transmitter transduces a discrete flow of phono-

logic units into a continuous speech signal, and by which an individual's Speech
Receiver performs the inverse transduction to recover the discrete flow of phono-
logic unitsfrom the continuous speech signal. The problem of acoustic phonetics,
then, is to determine
and describe this code to determine and describe the
phonetic system of a language.
In attacking this problem there are certain general considerations to be kept
in view
considerations which apply axiomatically to any study of any sort of
transducer in any sort of communicative system. A transducer is a device (of
hardware or of flesh) which converts an input signal into an output signal; the
only transducers with which we are concerned are non-singular, which means
that their output allows, by the application of a suitable second transduction,
of the virtual recovery of the original input. The conventions according to which
a given transducer converts input into output constitute a code. The physical
construction of a transducer determines the code by which it works in the sense
that if one knows the former, one can determine the latter. The reverse is not
true: transducers of varying physical construction can function according to a
single code. However, one can define as communicatively equivalent all those
transducers, real or possible, which operate or would operate according to a given
code, and one can then be concerned with the code rather than with the trans-
ducer as a physical entity: in communicative theory, this is generally what one

does, and in effect it amounts to treating the transducer itself as a "black box."
As between input, output, and code, there is a sort of two-to-one determinism
if the investigator knows any two of these three, the third can be predicted.

Knowing input and code, one can predict what output will be produced by the
input; knowing output and code, one can similarly predict (in a sort of inverse
way) what input produced the output. Or, knowing input and output, one can
describe the code. On the other hand, knowledge, no matter how complete, of
just one of these three factors is equivalent to no knowledge at all, since nothing
can be deduced about either of the others.
Often enough, the knowledge which is available at the beginning of an investi-
gation falls short of complete information about two of the three factors, without
amounting to knowledge of one of the three factors and no knowledge about
either of the other two. In military cryptanalysis, for example, one may have
(1) a fair sample of outputs (messages in encoded form) (2) general information
;

as to the sort of input which might be responsible for the outputs (the messages
certainly concern ship-movements, plans of attack, and so forth) and (3) some
;

general information as to the sort of code which has been used. The "breaking"
of the code
and the interpretation of the specific outputs then involves a
 A MANUAL OF PHONOLOGY 191

trial-and-error process; there is a well-developed mathematical theory of the

conditions under which the trial-and-error can be successful, and much practical
know-how for making the period of trial-and-error as short and effortless as pos-
sible. In theory, all that the trial-and-error process can ever lead to is a solution
of a certain degree of probability not to an absolutely certain solution. In prac-
tice,however, the degree of probability of correctness is often high enough to
eliminate any reasonable doubt.
Let us apply this to our present problem. Oscillograms and spectrograms afford
us as large a sample as we wish to obtain of the output of an individual's Speech
Transmitter. By one of the principles stated above, this sample in itself is quite
useless. But we can supplement it, two ways: (1) by
theoretically, in either of
obtaining information as to which utterances (of those of which we have the
acoustic records) sound the same to a native speaker and which ones sound dif-
ferent. We then have precise information about output and statistical informa-
tion about input, and can proceed to a probabilistic determination of the code.
The logic of this is the same as the logic of ordinary methods of phonologic analy-
sis (3), except that acoustic specification of output replaces articulatory, im-
pressionistic,and imitation-label specification; in theory, the procedure should
yield in due time not only a probabilistic specification of the phonetic system
(as a code), but also of the phonologic system itself, as an abstract network of
elements; (2) by accepting the phonologist's description of a phonologic system
(as attained by his ordinary methods, and perhaps as abstracted along the lines
discussed in 4), as a presumably accurate description of the economy of the
signals which are emitted by an individual's Phoneme Source and Speech Re-
ceiver. We then have precise information both about output and about input,
and can proceed to a presumably precise (rather than probabilistic) determina-
tion of the code.
The first of these two procedures has never been used. Theoretically it should
be possible, but there is serious reason to doubt its practicality. In any case, it

would be vastly more than the second procedure, and it is hard to see
difficult

how it could possibly yield any information which the second, and much simpler,
procedure cannot be made to yield. Thus, while it is not logically valid to say
that the acoustician must accept the findings of the phonologist as a stable frame
of reference, it is an empiric fact that doing so constitutes a practical short-cut
that the acoustician would be foolish to reject.
521. Methods of Attack. There are two major lines of experimentation which
can be followed in seeking an answer to the fundamental problem of acoustic
phonetics, under the practical assumption, just discussed, that the acoustician
will accept the phonologist's analysis as a stable frame of reference:

(1) A set of utterances is selected, each one clearly and completely identified
in phonologic terms. The selection is such that each low-level phonologic element
(each phoneme, each juncture, each intonation pattern, and the like) occurs in
various different environments. A number of recordings are made of each utter-
ance, perhaps as said on successive occasions by a single voice (at least this is a
good way to start: the complication of differing voices can well be postponed).
192 A MANUAL OF PHONOLOGY

Each recording of each utterance is then checked, by listening, to be sure that,

as actually delivered, the phonologic structure is just as planned; deviants, in
which the speaker has said the wrong thing, are discarded. Then acoustic records
(particularly spectrograms) are made for each uttering of each utterance. This
gives the analyst two sorts of material with which to work: the phonologic
specification of each utterance; and the measurable acoustic record of each utter-
ing of each utterance. The necessary procedure in searching through this evi-
dence for the acoustic correlates of the phonologic elements is obvious in general
outline, though tedious and very difficult in detail.
(2) The second line of investigation is quite different, and can be profitably
undertaken only after some tentative results have been achieved by the first
method. Subjects are selected who are native speakers of the specific language
under investigation (and of the proper dialect of that language.) Hand-drawn
pseudo-spectrograms are prepared, varying in some systematic way one or more
of the features which have been shown, by experimentation of the first kind, to
be of probable relevance. These are then played on the pattern-playback to the
subjects, who are instructed to identify what they think they hear. The instruc-
tions given the subjects have to be phrased with great care, and their answers
have to be carefully interpreted, since the frame of reference in terms of which
the responses are to be evaluated (namely, their phonologic system) is one which
they possess and use but about which they are not trained to verbalize ac-
curately.

522. General Character of Work Done So Far. Almost all investigations so far
reported in the field of acoustic phonetics have been of an essentially preliminary
nature, and have involved certain simplifications which must be borne in mind.
These simplifications are of three sorts:
(1) In work along the first line of experimentation, there has been a great deal
of "spoon-feeding" of spectrographs: that is, inputs have consisted of utterances
spoken very slowly and carefully, or even of pseudo-utterances such as artifi-
cially prolonged vowels. There are certain purposes (other than those of acoustic
phonetics) for which this sort of spoon-feeding is legitimate. For example, if one
is trying to work out a "visible speech" system for the deaf, it is legitimate to

require especially careful articulation: if the system could be made to work only
with this restriction, the small imposition thus made on anyone who wished to
communicate via such apparatus with the deaf would be far more than counter-
balanced by the gain. But for acoustic phonetics the results are only of the
roughest preliminary value. Eventually, it is imperative that naturally flowing
speech which means rapid, carelessly articulated speechbe examined.
(2) first sort, there has in general been an inade-
Also in experiments of the
quate understanding of the full complexity of phonologic systems. Studies have
been made, for example, of the acoustic correlates of "English [ae]," with no
realization that as used by the investigator this single symbol may have sub-
sumed as many as four phonologically distinct English stressed syllable nuclei:
/eh/, /ae/, /aeh/, and /aey/. Eventually it is essential that such lumping be
abolished. In the meantime, however, the results obtained are by no means to
be discarded th^y are
; of value, despite the lumping, for reasons connected with
 A MANUAL OF PHONOLOGY 193

functional load and with the difference between "frequency norm" and "clarity
norm" phonology (both of which will concern us in 6).
(3) A similar oversimplification has been involved in work along the second
line of investigation. Instead of taking the actual phonologic system of some real
language as frame of reference, what the experimenters have done is to devise

one or another miniature system turning largely on a traditional misunder-
standing of "IPA" and its "cardinal vowels." Thus one study of this sort, which
we shall discuss later, used syllables beginning with [p t k] and ending with one
of a set of seven "cardinal vowels" [i e e a o o u]. Fortunately, as shown by our
survey in 2, the vast majority of known languages include a stop subsystem
/p t k/, and a good many include vocalic or which can be roughly
syllabic units
identified with the seven "cardinal vowels" just listed. The latter comment ap-
plies even to English the speaker whose system has not been modified by train-
:

ing in "general phonetics" of the IPA sort can identify the seven "vowels" with
his /iy ey eh ah oh ow uw/.
The nature of the misunderstanding which leads to this sort of simplification
and artificiality deserves brief comment here. What we may call the "IPA gen-
eral phonetic" misunderstanding assumes that there is some even
strictly finite,
if large, total set of "speech sounds," as auditorily distinguishable units, from

which each actual language makes some selection. The IPA "cardinal vowels"
form part of this stock, and it is supposed to make sense to say that a given IPA
"cardinal vowel," say [i] or [a] or [in], occurs, or does not occur, in a given lan-
guage. The Prague phonologists early pointed out the crucial error in this view,
as did, more or less independently, such American linguists as Bloomfield and
Sapir. A phoneme is functionally nothing at all
except insofar as it is different
from other phonemes. A phonologic system is at bottom not a set of elements,
but a network of contrasts: the elements are but the end-points of the contrasts.
The entire discussion of this manual is based on this view.
Although the above remarks may seem sharply critical, we must in fact be
thankful that acoustic phoneticians have forged ahead within their oversimplified
frame of reference. Without the results achieved through oversimplification, we
should be completely at a loss for what to look for in examining the acoustic
properties of normal rapid speech. We perhaps should also be thankful that
acoustic phonetic research has been carried out by speakers of Western languages,
in which vowels are prominent and important. It is not easy to imagine what
turn acoustic phonetic study would have taken were we speakers of a language
Coola (2213), but one suspects that the initial period of floundering-
like Bella
about and of repeated frustration would have been considerably prolonged.
53. Tentative Results. In the following subsections, we shall discuss the sorts
of partial correlationswhich have so far been discovered. These do not con-
stitutean answer to the fundamental problem of acoustic phonetics, but they
are evidence pointing towards the answer.

531. Formants and Vowels. Work with spectrography quickly strengthens a

theory first proposed on the basis of earlier experiments with cruder apparatus:
the theory that vowels (and certain kindred types of unit, roughly "vocoids" or
194 A MANUAL OF PHONOLOGY

"sonorants" or "resonants") are kept apart largely by the frequency-locations

of formants.One version of this theory holds that vowels distinguished in articu-
lation solely by lip and tongue position (setting aside retroflexion, and non-oral
differentiating factors such as nasalization, glottal stricture, pharyngealization,
and the like) are, or at least can be, distinguished completely by the two lowest
formants.
This general theory (not necessarily the two-formant version) is reasonable
when we examine the functioning of the speech tract in the production of voiced
vowels. The vibrating glottis produces a musical tone exceedingly rich in har-
monics. The particular conformation of the oral cavity as a resonance chamber,
brought about by lip and tongue position, serves to reinforce the energy at and
near certain frequencies and to damp out that at others. The reinforced fre-
quency-bands are the formants.
5311. Artificial Two-Formant Matching of "Cardinal Vowels." In ordinary
speech there rarely occurs anything which might be called a "steady state"
vowel. The articulatory organs are in motion almost constantly, and therefore
the formants are also constantly shifting their frequency-positions. This is just
as true of phonemically simple (or, phonetically, crudely "monophthongal")
vowels as it is of complex nuclei which involve an obvious glide, except that the
rate of change may be rather less, or the direction of change less obvious, for the
former. It is well known that any vowel sound, if prolonged indefinitely, soon
loses its character as a "speech-sound": what is heard, in the context of speech,
as a vowel color, comes to be heard, when the sound is prolonged, as a tone-
quality. One need only think of humming, which involves the articulatory mech-
anism ofan [m], but which does net for long continue to have any noticeable
[m]-ness about it.
Nevertheless, it is possible to prolong fixed vowels or pseudo-vowels somewhat
longer than they would normally be held in ordinary speech, without imme-
diately having them lose their vocalic character. The first experiment which we
shall report involved steady-state pseudo-vowels, produced by hand-drawn
spectrograms played on the pattern-playback. A large number of these were pre-
pared, and then a psycho-acoustic experiment was performed to determine the
optimum two-formant matching of the sixteen Jonesian "cardinal vowels." If
these sixteen "vowels" were a real vowel system, then in ordinary phonologic
terms they would constitute a 4X2X2 system, charted as follows:

i ii m U
e Y
e ce A
ae a a d .

The subjects used in the experiment knew at least something of both French
and English and had had some training in IPA phonetics. The testing procedure
guaranteed in advance that the first (lower) formant would be identical for all
four vowels of a given articulatory height. The most satisfactory pairs of form-
ants proved to be the following:
 A MANUAL OF PHONOLOGY 195

[i] 250, 2900 [u] 250, 1900 [m] 250, 1050 [u] 250, 700
[e] 360, 2400 [0] 360, 1650 [y] 360, 1100 [o] 360, 800
[e] 510, 2000 [03] 510, 1450 [a] 510, 1150 [o] 510, 950
[] 750, 1650 [a] 750, 1300 [a] 750, 1200 [n] 750, 1100.

In Figure 21 we show the pseudo-spectrograms for these sixteen vowels, and in

Figure 22 the sixteen pairs of formants are mapped in two dimensions, first
formant against second. It has become customary, in this sort of representation,
to plot the first formant vertically, the values increasing from top to bottom, and
the second formant horizontally, values increasing from right to left. This is in
three ways a reversal of the most usual conventions for graphs and charts, but
it is just as correct mathematically, and has the property (a merit if not abused)

that articulatorily high vowels place higher on the graph than low vowels, and
articulatorily front vowels further left than back vowels. Both of the outer scales
in Figure 22, and all scales in figures to follow, are logarithmic (see Figure 14).
However artificial the convention of the "cardinal vowels" may be for ordinary
phonologic analysis, in acoustic phonetics it is highly convenient to have a frame
of reference of this sort. We are going to adopt a sixteen-point grid as frame of
reference, and we shall call the points (acoustic) cardinal points, but in doing so
we shall modify slightly the sixteen discovered in the above experiment. Figure
22 shows lines adjoining the vowels of the same height and also the vowels of
the same front-back quality and the same rounding-unrounding quality (all
of this, of course, in articulatory terms). The fine joining any four vowels of the
same height is horizontal, since the first formant was prescribed in advance to
be the same for all four. The more nearly vertical lines in Figure 22 also look
straight, but they are not, as a larger-scale graph would show. However, by
keeping the first formants as fixed by the experiment, and the second formants
for the highest and lowest vowels, but by making slight adjustments in the second
formants for the higher-mid and lower-mid vowels, we can make the four vertical

i e e 89 y oe a uiyad nooa
Fig. 21. Pseudo-Spectrograms for Artificial Cardinal Vowels
Reproduced with the permission of Haskins Laboratories. The phonetic symbols under
the fifth-from-last and the last spectrograms should be interchanged.
196 A MANUAL OF PHONOLOGY

I
i i i i
I
i i i i
i
i i i i

i

i i 200^

O "

300 m

400 ~

500~

600^

700 v

800 "J

TONES-* J |_ J__l L 900 nj

OCTAVES-* >

iooo
frequency of formant two
Fig. 22. Fokmant-bt-Formant Plotting of Artificial Cardinal Vowels
Reproduced with the permission of Haskins Laboratories. The two right-hand phonetic
symbols in the bottom row should be interchanged.

lines straight. The new values for the higher-mid and lower-mid vowels are as
follows:

[e] 360, 2411 [0] 360, 1675 [y] 360, 1098 [o] 360, 813.2
[e] 510, 2011 [oe] 510, 1485 [a] 510, 1145 [o] 510, 938.4.

Figure 23 shows the resulting grid of straight lines: at each meeting-point or

intersection of two lines there is one of our acoustic cardinal points of reference.
The crosses mark the positions of the psycho-acoustically matched values where
these deviate from those arbitrarily established for our grid. There is only one
slight gain in making this readjustment: the grid is easier to draw on a graph.
But this gain is not offset by any loss at all: the "cardinal vowels" are arbitrary
to begin with, and, what is more, the very slight changes in second-formant fre-
quencies would be scarcely detectable by ear.
5312. French Vowels. Spectrograms of French vowels spoken in isolation show
general agreement with the cardinal points conventionally thought of as closest
to them. There are two spurious reasons contributing to this. One is that the
French vowel system played a major role in the development of the original
"cardinal vowel" theory. The other is that in selecting first-formant frequencies
for the first part of the psycho-acoustic tests for cardinal-vowel matching, resort
 A MANUAL OF PHONOLOGY 197

10 27 VY 2.0 /6 12 JO
1 1
P"

/o

Fig. 23. Acoustic Cardinal Points

was made to measurements of French vowels. After the best second formants had
been selected, these were held constant and the first formants varied, but except
for the four lowest vowels the original values were finally settled on as more
satisfactory than any alternatives.
However, the correlation between nine of the vowels of educated northern
French (all but the two lowest ina3+3+3+2 system) and our cardinal grid is
nevertheless interesting; we display it in Figure 24. Perhaps the most striking
deviation is the rather lower second-formant frequencies for the three French
front unrounded vowels /i e e/; we shall discuss the implications of this a bit
later.

5313. English Vowels. Spectrograms have also been made of English vowel
phonemes in isolation; there more artificiality about this than for French, since
is

the English vowel phonemes do not occur stressed before no consonant, nor pro-
longed, and we can therefore expect something a little wilder in the way of re-
sults. Figure 25 shows eight of the nine English vowel phonemes as measured in

this artificial way; there are no data for the ninth. The frequency figures for the
formants are as follows:

l'\l 400,2100 /if 300, 1500 /u/ 450, 1000

/e/ 500, 1800 /a/ 600, 1300 /o/ 550, 900
// 650, 1700 /a/ 700, 1100 hi -
The arrangement may at first seem rather random, yet certain correlations ap-
pear. Considering any three vowels of the same articulatory height, the frequency
of the second f ormant decreases as we pass from front to mid unrounded to back
198 A MANUAL OF PHONOLOGY

Fig. 24. Measured French Vowels Mapped against the Cardinal Reference Grid

Fig. 25. English Vowels Mapped against the Cardinal Reference Grid

rounded (e.g., /i/ to /i/ to /u/). Considering any three of the same front-back and
Up position, the frequency of the first formant increases as we go from high to
low (e.g., /i/ to /e/ to //).
The measurements of /a/ and /a/, as compared to the cardinal points on our
grid, are interesting. Words such as cup, but, luck have /a/ in most American
English, but /a/ in southern British English, of the sort with which Daniel Jones
usually deals. The Jonesian cardinal [a] is supposed to be quite close to the British
vowel in cup, but, luck. The test audience in the two-formant cardinal matching

was obviously largely American, for their selection of a match for cardinal [a]
reflects an identification of that with their own /a/, rather than with British
English /a/, in the key reference words.
In Figure 26 we graph, against our reference grid, one occurrence each of /u/,


A MANUAL OF PHONOLOGY 199

Fig. 26. Three English Vowels in Context

The x's locate the measured English vowels shown in Figure 25.

/e/, and /ae/ in clearly articulated, but rapid and natural, speech. The /u/ is
from the word understood, the /e/ from hotel, and the /ae/ from exactly. Each
vowel appears as a curve: there is an arrowhead at one end of the curve, and dots
appear at various places along each. The dots represent actual measurements, via
a spectrogram, taken at intervals of two hundredths of a second, throughout the
duration of the vowel; the curves are then drawn in freehand. It will be noticed
that the curve for /u/ passes fairly close to the point measured for an artificially
prolonged /u/, and that that for /ae/ is even further down and to the left than the
similarly measured point for an artificially prolonged /ae/. The curve for /e/,
however, is quite a distance from the "point" /e/. This is presumably because of
the very dark /l/ which follows it. Nevertheless, and despite the fact that the
speaker for these three vowels was different from that for the steady-state
vowel measurements, it is clear that there are very wide limits of tolerable varia-
tion which does not destroy the recognizable phonologic identity of the vowels in
actual context.
If we were to make observations and measurements of, say, a thousand occur-
rences of each of the nine English vowels, in actual speech, and to draw curves for
each occurrence of each in the style of Figure 26, we should expect to find a
distribution of each of the nine around some most favored point; this is shown
quite hypothetically
in Figure 27. The relative darkness within the region for
each vowel supposed to represent the relative frequency with which that vowel
is

shows formants at a particular point; the regions for the six vowels intersect,
implying that some /i/'s, for example, pass through areas also passed through
by some occurrences of /e/ or of /i/. Remember that Figure 27 is not based on
actual measurements; it is only an attempt to show the sort of picture which we
should expect actual measurements to yield.
200 A MANUAL OF PHONOLOGY

Fig. 27. Hypothetical Graph of Frequency-Distribution of English Vowels

This portrays what we should expect to find if we made measurements of vast numbers
of occurrences of each of the nine English vowels. The portrayal is not based on actual
data.

5314. How Many Formantsf Although all of the experiments and observations
reported on in the above have tacitly assumed the two-formant theory, it should

be noted that the results do not in any sense constitute proof of that theory.
Even for languages like French and English, we have demonstrated neither the
necessity nor the sufficiency of two formants.
On the score of necessity, there is evidence that reasonably adequate matching
of some vowel-colors can be achieved with but a single formant. Old disc records
may fail to reproduce the lower formant of such high vowels as [i] and [u], with-
out destroying the intelligibility of speech or even the distinctiveness of those
two vowels. The same is true of obsolete telephones. A psycho-acoustic experi-
ment for one-formant matching was carried out at the same time as the two-
formant matching already reported, with suggestive results. It was discovered
that the effect of an [a] can be achieved, more or less closely, not only by a pair
of formants of equal loudness at 750 cps and 1300 cps, but also by a single for-
mant at about 1200 cps. Similarly:
[i] 2760 cps or anything higher
[e] about 2520 cps
[e] about 2160 cps
[ae] about 1100 cps (poor)
[a] 950 to 1050 cps (poor)
[o] 510 cps (poor); about 720 cps (better)
[o] 360 to 480 cps
[u] about 240 cps.

Now it must be recognized that the acceptability of such artificially simplified

pseudo-vowels not only the one-formant ones, but the two-formant ones de-
 A MANUAL OF PHONOLOGY 201

scribed earlier to a test audience, as equivalent to one or another natural vowel

(or at least as more similar to one natural vowel than to others), depends on a
sort of complaisant listening by the audience, which is conditioned by the test
situation to hear vowel-like characteristics in the stimuli if at all possible. The
acceptance of a single formant at about 240 cps as a sort of means only that [u]

(1) this single-formant sound resembles an [u] more than it resembles any other
"natural" vowel, and that (2) this particular single-formant sound resembles an
[u] rather more than does any other single-formant sound (of those tested). Quite

similarly, the acceptance of the two-f ormant [i] (250 and 2900 cps) as an [i] means
that this stimulus was more [i]-like than any of the other two-formant stimuli
which were tested, but in no sense implies that the artificial two-formant stimulus
sounded as thoroughly [i]-like as does a "natural" [i], with some energy at various
other (and higher) frequencies.
On we can draw only some rather negative con-
the score of sufficiency, then,
clusions. Pseudo-speech, produced by playing back hand-drawn spectrograms in
which vowels are represented by at most two f ormants and sometimes only by a
single formant, ought to be intelligible providing that other factors (representa-
tion of consonants) are adequately provided for; experimentation has shown
that this is indeed the case. Similarly, actual speech, reaching the ears of a hearer
under noise conditions which effectively cancel out some of the natural f ormants
for vowels, should not thereby lose all intelligibility. On the other hand, it is
clear that under many hearing conditions vowel-sounds, as they reach a hearer,
involve more than two formants, and we have absolutely no reason to believe
that under such conditions the hearer automatically discards any information
carried by the higher formants and bases his interpretation on the two lowest
alone.
Some positive evidence for this is found in the extremely high second-f ormant

frequencies which were required by the test audience in the artificial two-formant
matching of "cardinals" [i e e], as compared with the values for French /i e e/
(Figure 24), or with other measured values for natural vowels of this general
sort
usually around 2400 cps, 2000 cps, and 1800 cps respectively. The natural
vowels have some energy at higher levels (above the second formant), which
gives them an auditory quality which we may call "shrillness." The artificial two-
formant vowels can achieve comparable "shrillness" only by having the second
formants relatively higher than do the natural vowels.
We have, therefore, no reason to believe that in general the two lowest form-
ants do all the work of identifying vowel colors in ordinary speech. Beyond this,

itmust be borne in mind that formants have not been completely specified when
we state their number and the frequency-level of each. Apparently it is not neces-
sary to add any specification as to the width of the reinforced band of frequencies
which constitute a formant, since this seems to remain fairly constant, at some-
thing between 120 and 150 cycles: the mid-point of this frequency band is chosen
as the representative frequency for the whole formant. But it is essential to pay
attention to the relative intensities of the different formants. We do not know
in full detail what effect rounding has on a vowel color (say the passage from [i]
 :: ;

202 A MANUAL OF PHONOLOGY

to [ii]), but it is at least clear that rounding somewhat weakens formant two and

any higher resonances. When the higher formant of the artificial two-formant
cardinal [i] was progressively weakened, the test audience described it first as
taking on a sort of "dull" character, and then, with sufficiently great weakening,
as becoming almost like an [u]. An [i] thus partially "dulled" can easily be heard
as a sort of [ii], if the hearer is favorably inclined towards that interpretation.
5315. Dimensionality and Formants. In the fight of all the above, we may ven-
ture a few rough tentative conclusions.
In the relatively rare case of a one-dimensional vowel system (2442), the most
important acoustic correlate will be the frequency of the first formant. Both the
location and the relative intensity of the second formant, and of any higher
resonances, will vary allophonically, as do, in articulation, the front-back nature
and the roundedness vowel phoneme, depending on environment, and will,
of the
in a sense, constitute part of the environment rather than part of the vowel
(3222).
In the common case of an articulatorily two-dimensional vowel system, with
no* more than two contrasts at any height (e.g. /i/ versus /u/, without either
/ii/ or /tn/), our usual tabular arrangement of the vowels will correlate roughly
with their positions on a two-formant graph, with both relative intensities and
any higher formants set aside: position of first formant will correlate with tongue
height, and position of second formant with the other articulatory features. This,
of course, does not mean that resonances higher than the second formant are
performing no communicative function, and perhaps in due time we will discover
some single function of the second formant and any higher resonances which
correlates even more precisely with the articulatory features other than tongue-
height and which can replace the second formant in our two-dimensional acoustic
charts.
In the case of some two-dimensional systems with three contrasts at a single
height (for example, quite possibly English), the statements just made about
two-dimensional systems with at most two contrasts at a single height may well
prove to apply. But in other two-dimensional systems (e.g., certainly French
and German, with /i ii u/), and certainly in genuinely three-dimensional systems,
a two-formant chart will not accurately portray the acoustic correlates. The full

specification of the latter will require not only the frequencies of the two lowest
formants, but also their relative intensities, or the location and intensities of one
or more higher formants, or both.
Thus our best guess, on the basis of evidence so far available, constitutes at
least a partial rejection of the two-formant theory: we are led to suspect that
there will turn out to be a fairly close correspondence between the "dimension-
ality" of a vowel system, determined articulatorily, and the number of inde-
pendent acoustic variables necessary for the specification of the individual vowels
in the system:

one-dimensional system frequency of first formant

two-dimensional system (1) frequency of first formant;


A MANUAL OF PHONOLOGY 203

(2) either frequency of second formant, or

some single-valued function of this and higher
resonances;
three-dimensional system: (1) frequency of first formant;
(2) frequency of second formant;
(3)some single-valued function of the in-
and of the fre-
tensity of the second formant
quency and intensity of higher resonance-
bands.

5316. R-color and Nasalization. Observations have led to the suspicion, partly
confirmed through psycho-acoustic experiments, that the position of the third
formant correlates with r-color and with nasalization. Just what connection there
may be between this and the relevance of the third formant for other factors of
is not yet known.
vowel-color contrast
When the third formant is relatively close to the second closer than it is
(when it can be observed at on a spectrogram) for a non-r-colored vowel
all

subjects report hearing r-color, and the closer the third formant is to the second
the stronger is the r-color. In spectrograms of actual speech, the correlate of r-
color does not always appear in sequence where the phonologist would insist it is:
one speaker's word phrases had r-color throughout the /ey/, and most promi-
nently at the end rather than at the beginning. One must assume that the hearer
sorts out what he hears and puts each bit in "proper" chronological sequence on
the basis of expectation.
The acoustic correlates of nasalization are not so clear. Formant three seems
to be raised relatively higher for a nasalized vowel than for a corresponding oral
vowel, but there are other differences too: often an additional band of reinforced
resonance appears just above formant one. French is a poor language to use in

testing this, since theFrench nasal vowels do not very closely match any of its
oral vowels in tongue and lip position; we therefore suspect that the hearer of
French identifies a nasal vowel not only on the basis of the direct acoustic cor-
but also through the somewhat different location of first
relates of nasalization,
and second formants. French spoken with the velic closed save for the nasal
consonants would still keep apart all the vowels (nasal and oral). Portuguese or
Ojibwa, where the matching of oral and nasal vowels is closer, would be better
languages to use in this connection.
The apparently opposite effects of r-color and of nasalization on the distance
between formant three and formant two also raise some interesting problems.
In Chinese one has (phonetically speaking: some of the sounds are phonemically
clusters) vowels with neither nasalization nor r-coloring, vowels with nasalization
without r-coloring, vowels with r-coloring but no nasalization, and vowels with
both the example, at the end of the syllable hengr2 'horizontal stroke'.
last, for

Thus and nasalization are not mutually exclusive. Therefore one or the
r-color
other, or both, must involve acoustic correlates in addition to the alteration of the
distance between formants two and three.
204 a^anual'of phonology

532. Noises and Noise Bursts. Voiceless consonants show up in a spectrogram

with no clearly-placed formants; rather, energy is spread more or less evenly
through whole bands of frequencies. As in the case of voiced vowels (and other
relatively resonant voiced sounds) this is entirely what we would expect from an
examination of the articulatory motions involved primarily because of the
absence of the harmonically rich glottal tone.
Voiceless spirants appear as randomly distributed energy through a relatively
wide frequency band, sometimes with points of relative concentration, but the
latter seem to depend on the identity of the speaker. English // yields the lowest
band of frequencies, centering around 1800-3000 cps. English /s/ is higher,
with some energy up to 4000-5000 cps. English /if and /$/ are even higher, and
considerably fainter (less total energy) on spectrograms made to cut off at 4000
;

cps they often do not show at all. We can be quite sure that in conversation under
ordinary conditions the distinction between /(/ and /Of is often not heard, but
is "read in" to fit the context, either on the basis of the sort of evidence discussed

in 534, or, in some cases, perhaps even without that evidence. English is of
course intelligible even if all four of these spirants are replaced in all occurrences
virtually by silence: old-fashioned phonograph records, with a low cut-off fre-
quency, can be understood, though the distortion is obvious. It is interesting to
note that the distortion is more noticeable for /s/ and // which are less
affected
than for /f/ and /#/; presumably we notice the greater distortion of
the latter two less because we are fairly accustomed to not really hearing them
much of the time anyway.
The rather common inaudibility of /f/ and /&/ (and even of the voiced part-
ners, /v/ and /(5/) can be observed without any apparatus: it is only necessary
to listen for speech-sounds, rather than for meanings, during a conversation in
noisy surroundings.
Genuinely voiceless stops (by which we mean to exclude, for example, the
medial /t/ of a Middle Western rattle, even though the phoneme may be classed
structurally as "voiceless") show a brief moment of complete silence (a blank
vertical strip on a spectrogram). This serves to tell the hearer that some voiceless
stop, or a glottal catch, has occurred, but does not tell which one. The silence in
itself is quite identical for any voiceless stop or a glottal catch, but the way in

which the preceding signal (if any) fades out, and in which the following signal
(if any) fades in serves to differentiate them. Examination of spectrograms sug-

gests that one or both of two features in the fade-out and fade-in may perform
this differentiating function. One feature is a brief burst of noise spread over some
band the other
; is the rate and direction of change of the vowel formants next to
the stop silence.
An extremely interesting psycho-acoustic experiment has been carried through
to test the auditory effect of variously placed bursts of noise before various arti-
ficial (two-formant) vowels. A set of seven such vowels were used, as follows:
[i] 270, 2720 [u] 270, 600
[e] 360, 2200 [o] 360, 720
[e] 540, 1830 [o] 540, 960
[a] 720, 1320.
 A MANUAL OF PHONOLOGY 205

JB_ c.
4S10~
t -300~
-31*0 - . 4000 ~

-1400

ft
-1200 f ^BB
- - 3000
!?20

-0
1200

-tooo oi at oj 04
JO TlHC ISCC0N05)

3 2

9S0
-1000
T20

r*o
00
1*0 S40

270
SCO

560
*T0

1 e a 9 o u
tarn AS IN * J 1*
act *T TIT ASMCSTCR*J JM SO TOOTM

Fia. 28. Artificial Vowels and Noise-Bursts

Reproduced with the permission of Haskins Laboratories.

These were hand-drawn to last .3 seconds. Figure 28 shows, on the right, the seven
vowels; on the left, the twelve bursts of noise; and in the upper right, a typical
combination of noise-burst and artificial vowel.
In the test, the subjects had already been familiarized with the artificial
vowels; they were asked only to identify the preceding "consonant" as [p], [t], or
[k]. As might be expected, there was not complete agreement in all cases, but for

certain combinations agreement was very good, and in a few cases it was com-
plete. Figure 29 graphs all the results at once. The most important individual con-
clusion to be deduced from Figure 29 is that one and the same burst may be inter-
preted fairly consistently in more than one way depending on the following
vowel: the bursts centering around 1800 cps were interpreted generally as [p]
before [i] or [u], but as [k] before [a].
533. The Voice Bar. In spectrograms made with a wide filter of carefully
enunciated speech, a relatively low band of reinforced frequencies appears
fairly regularly during all voicing. For voiced vowels this is simply the first
formant, but for voiced consonants (particularly voiced stops and spirants) it
is often called, instead, the voice bar; it is at the frequency of the fundamental

of the glottal tone. Except for the presence of the voice bar, it is possible to think
that one sees some vague similarity between the spectrogram of, say, [ba] and
that of [pa], or that of [za] and that of [sa].
However, in all of these cases there is some question as to whether the main
206 A MANUAL OF PHONOLOGY

Fig. 29. Identification of Noise-Bursts

The seven vowels are plotted on the diagram. Dotted areas are those for a
artificial
noise -burst usually identified as [p] when followed by the artificial vowel shown in the
same strip; e.g., a noise-burst in the range of 1800 cps, followed by the artificial [i], was
regularly identified as [p]. Slanted hatching marks the areas where the identification was
[k], and circles the areas where the identification was [t]. The heaviness of the dots and

hatching, and the size of the circles, is a rough index of the extent of agreement of the
test audience in making the identifications. Reproduced with the permission of Haskins
Laboratories.

acoustic clue really lies in the voice bar. It is known that the effect of voicing
can be achieved, in hand-painted spectrograms, in either of at least two other
ways. One of these consists merely in making a given configuration short enough:
if one draws what is interpreted as [fa], and then makes the [f] part very short,

one can lead subjects to hear [va]. The other way turns on the bending of vowel
formants, which we shall now discuss in this connection and others.

534. Initial and Terminal Bending of Formants. We

have already mentioned
that acoustically steady-state vowels are very rare. Any
spectrogram of natural
speech shows this; and one may also compare Figure 26. Though the formants
for, say, English /se/ may pass through certain values which serve to distinguish
this vowel from any other vowel in English, they come from certain other values,
and go to certain other values, at varying rates, and the directions from which
they come and towards which they move seem to depend on, and to help to
identify, the surrounding consonants.
 A MANUAL OF PHONOLOGY 207

Fig. 30. Artificial Vowels with Bending Onsets

Reproduced with the permission of Haskins Laboratories.

This early suggested that much of the information as to what consonant

precedes or follows a vowel might be carried within what would acoustically
be called the vowel itself, rather than within what precedes or follows. Psycho-
acoustic tests have been made of this, some of them involving nothing more than
a spectrograph. A syllable such as back is recorded on the spectrograph's internal
storage band, and then everything is erased which would ordinarily be thought of
as representing the /b/ and the /k/, leaving only the two twisting formants.
When the tape, thus doctored, is played back, subjects usually report hearing
back rather than any other English syllable. The erasing can be done progres-
sivelyfrom the two ends, bit by bit, and it is remarkable how much has to be
taken away before what remains becomes unidentifiable.
Experiments along this line have also been done with the pattern playback. A
set of hand-painted spectrograms were prepared, each consisting of one of the
seven reference vowels mentioned in 532 and charted in Figure 28, but variously
modified by curved onsets; Figure 30 gives some examples. The results varied
from one to another vowel in a way, and were not equally
fairly complicated
clear-cut for all of the vowels. The results with the artificial
[a] were fairly sharp:

when the first formant of the artificial [a] was kept the same in all samples, then a
very low beginning for the second formant gave the effect of a preceding [b], an
even or slightly high beginning sounded like a preceding [d], and a very high
beginning was interpreted as a preceding [g]. It is interesting to note an apparent
contradiction between these results for [a] and the psycho-acoustic stop-identi-
fications of bursts of noise before the same artificial [a], as shown in Figure 29.
Since a noise-burst at approximately the frequency of the second formant of the
[a], or a little above, was identified as [k], and a relatively higher noise-burst as a
[t], we should expect that an even or slightly high beginning for formant two
would be heard as [k] and a very high beginning as [t]: but the experimental
results are exactly the reverse.
There was more consistency in interpretation of curved beginnings for formant
one, regardless of the vowel.A relatively low beginning tended to be interpreted
as voicing of the consonant, while a level beginning tended to be interpreted as
voicelessness.
Figure 31 shows nine syllables involving [a] and the essential correlation be-
tween initial (or terminal) bending of formants and the consonants. It will be
noticed that the effect of a terminal nasal is not achieved by a downswing of the
first formant, but by two separately placed formants, the lower of which is the
"voice bar."
208 A MANUAL OF PHONOLOGY

PLACE OF PRODUCTION
BILABIAL ALVEOLAR VELAR

Ho.
yo
^- ^ D
D
>c/>
(r
BA
(?
DA
o GA

1 ^
^
v<.

D PA TA KA

in
_j
< ^ =^ ^
to
<
z
C Z>

AM AN AD
Fig. 31. Nine Syllables with [a]

Reproduced with the permission of Haskins Laboratories.

54. Summary. One's natural a priori expectation, when beginning the study of
acoustic phonetics, would probably be that each relatively small phonologic

element say each phoneme, or perhaps better each allophone will appear on a
spectrogram as a recognizable pattern, subject to minor variation from one occur-
rence to another but by and large clearly distinguishable from the pattern for
any other phonologic element and that the patterns for successive phonemes in a
;

particular uttering of a particular utterance will be rather clearly set off from
each other (at least as well as successive letters are in careless handwriting). One
hardly expects an [f] to show up on a spectrogram in the visual configuration of
the letter-shape "f" or "F," of course, but one assumes that there will be some
sort of visual configuration which will regularly appear in response to [f], so that,
in time, one could learn to read sequences of these configurations as though they
were a phonetic alphabet with new shapes for the separate symbols.
We have seen that the results of actual experimentation are staggeringly at
variance with any such expectations.
We might nevertheless insist on finding reasonably exact points of separation
between successive "phones," so that by dividing a spectrogram into the right
number of strips by cuts perpendicular to the time axis, we could assign one strip
to each successive phone of the utterance. Such a strip then is (or represents) one
acoustic allophone of the phoneme with which it correlates. If two strips (from
two different utterances or from different parts of a single utterance) correlate
with a single phoneme and look very much the same, we would say that they
 A MANUAL OF PHONOLOGY 209

constitute occurrences of the same acoustic allophone. But we would soon find
that, unlike anything encountered in the bulk of conventional phonologic study,
acoustic allophones are numerous, diverse, and intersecting.
By "numerous" we mean that a single phoneme, instead of being represented
by a small handful of allophones, is represented by dozens or even hundreds of
clearly different ones.
By "diverse," we mean that the whole set of acoustic allophones which repre-
sent a single phoneme by no means necessarily appear as minor variations around
some single measurable constant feature. In ordinary phonology we do not like
to accept such a situation (save for junctures, 324). In acoustic phonetics we
would have no choice.
By "intersecting," we mean that a given acoustic allophone of phoneme A may
resemble some allophone of phoneme B much more closely than it does some
other allophones of phoneme A. An [\ in one context looks virtually like an [s]

in another. This is the sort of situation which we are prepared to tolerate in

ordinary phonologic work providing that there is clear environmental difference
between a given allophone as member of one phoneme and the "same" allophone
as member of another phoneme; but in acoustic work the difference of environ-
ment is often very hard to pin down. Once again, however, we would have no
choice.
Now it is quite possible to find, or invent, signalling systems in which logically
this same Suppose that two detectives are about to enter
sort of situation occurs.
a house, where one of them, on the basis of what he observes, will have to signal
instructions to the other. They cannot know in advance just what circumstances
they will encounter, so they arrange for alternative signals for each instruction,
to guarantee that at least one of the appropriate signals can be given. The first
detective says "If I light a cigarette and have to use two matches, or if I sit
down on the arm of a chair, or if I ask for a window to be opened, you make your
excuses and get out and phone the boss. But if it takes me three matches to light a
cigarette, or if I sit down on a table, or if I open a window myself, then grab
him." Clearly, there is no physical feature shared by the three signals assigned
to the first instruction, and excluded from the three for the second. Indeed, the
two cigarette-lighting signals are more similar to each other than to any of the
other signals, even though they have been assigned opposite meanings. We might
even say that in each of the cigarette-lighting signals, one feature is (within the
economy of the system) absolutely identical actually fighting a cigarette; this
feature carries one meaning or the other depending on the number of preceding
vain attempts. This sort of multiple mapping is what we would face in seeking
correlations between acoustic allophones and phonologic elements.
Now there is no reason to believe that any approach can be developed which
will materially reduce the obvious complexity of phonologic-acoustic correlation.
However, there is at least one readjustment which can be made in our definition

of "acoustic allophone," which promises somewhat greater clarity, and also is

seemingly more realistic. This is to relinquish the specification that we break the
acoustic signal into acoustic allophones by cuts transverse to the time axis. By
210 A MANUAL OF PHONOLOGY

this change, we obtain allophones which are not only numerous, diverse, and
intersecting, but also overlapping that : is, the allophones do not necessarily each
end as the next one begins, but may have the latter part of one simultaneous
with the beginning (at least) of the next. Thus in pruning back, as mentioned
above, we get results such that we must say that the vowel-sound (defined as
the segment during which there are clear formants) contains some of the /b/
allophone and some of the /k/ allophone. What a native listener does, therefore,
to the signal which comes to his ear, is ultimately equivalent to redistributing
the sound-qualities along the time-scale, putting some part of the vowel-quality
into the preceding consonantal segment and some into the following one, so that
each is perceived as being what it must be even if the consonantal phase (acous-
tically defined) is inaudible, and leaving the residue (after those subtractions)
to be heard as the correct vowel phoneme.
Thus we must conceive of each phonologic unit in an utterance as an element
the acoustic correlate of which is spread through a rather long time an interval
considerably longer than we take to be the average duration of single successive
phonemes, as judged in the ordinary articulatory way. The Speech Transmitter of
an individual receives a discrete series of impulses from his phoneme source, but
the Speech Transmitter slurs this discrete series into a continuous muddy signal.

Often enough, indeed if we can judge from spectrograms of rapid normal speech
the speech transmitter slurs some of the successive impulses so badly that the
speech signal contains no separately detectable trace of them.
The best analogy I have been able to think of for this is a very homely one.
Imagine a row of Easter eggs carried along a moving belt the eggs are of various;

sizes, and variously colored, but not boiled. At a certain point, the belt carries
the row of eggs between the two rollers of a wringer, which quite effectively smash
them and rub them more or less into each other. The flow of eggs before the
wringer represents the series of impulses from the phoneme source; the mess that
emerges from the wringer represents the output of the speech transmitter. At a
subsequent point, we have an inspector whose task it is to examine the passing
mess and decide, on the basis of the broken and unbroken yolks, the variously
spread-out albumen, and the variously colored bits of shell, the nature of the flow
of eggs which previously arrived at the wringer. Note that he does not have to

try to put the eggs together again a manifest physical impossibility but only
to identify.
The inspector represents the hearer. We have no reason to believe that the
hearer is able to assemble scattered and thinned-out bits of evidence for a certain
impulse (a certain egg, of certain size and color) any more efficiently than the
acoustician can in his visual examination of spectrograms; rather, the hearer
understands anyway (when, indeed, he does) because what reaches him is such
as to necessitate, in terms of the whole economy of the language, the presence of
the obscured impulse. Here are what looks like two large green eggs in succession.
But on this belt (in this language) two large green eggs are never supplied in
succession: there is always a small red egg between. Therefore I must interpret
 A MANUAL OF PHONOLOGY 211

this situation as involving a small red egg, even though I see no direct evidence
of it. We shall follow this clue further in 6.
In other words, the advantage of recognizing overlapping allophones is that,
although it may be more difficult to extract and describe individual allophones of
this type than of the non- overlapping building-block type, every success in

extracting one results in a reduction of intersection necessarily so, since the only
criteria for correct separation of simultaneous allophones would be phonologic
criteria, and in phonology we do not tolerate intersection. Only thus, by going

as far as possible with splitting simultaneous allophones apart, would we have a

chance of completing the job of finding the acoustic correlates of low-level
phonologic units (components, phonemes). Whether the job actually can be
completed is not known, but at present we believe that progress can still be
made along this line, and as long as progress appears possible there will be no
need to despair of ultimate success.

6. Functional Load and Analytical Norms. The positive correlations between

phonologic structure and the speech signal, such as can be discovered via "spoon-
fed" spectrograms and via psycho-acoustic experiments with the pattern play-
back, are of challenging interest; but they are of minor importance as compared
with the negative results obtained from spectrograms of natural flowing speech.
From the latter, we discover that with surprising frequency a phonologic feature
which must be present, as judged by the responses of native speakers and by the
observations of the phonologist, by anything that we can locate
is not represented
in the speech signal. The receptive mechanisms of a native speaker, and those of
the phonologist which he has trained to mimic those of a native speaker, must
perform some sort of interpretive operation on the speech signal which hard-
ware cannot match.
In a general way, we can understand how this is done. A spectrograph is un-
prejudiced. It reacts, at each moment, just to the gross acoustic input to its
microphone. energy at a given frequency, the spectrograph reacts to
If there is
it; if there is none, it The previous history of the signal, and what
does not react.
comes later on, play virtually no role in this reaction. A native speaker of a
language, on the other hand, is strongly prejudiced. We hear speech not merely
in terms of the momentary quality of the speech signal, but in terms of our ex-
pectations and, often enough, in terms of what comes later.There seems to be
no way, at the moment, in which such a complicated set of ever-changing prej-
udices can be built into a machine. But a linguist can build them into himself,
by going through the ordinary procedures of linguistic analysis which closely
resemble, of course, the procedures by which one learns a language. For a linguist
(or a student of any other phase of culture) to be "objective" means something
rather different than for a physical scientist to be "objective." A linguist does
not attain usable "objectivity", in analyzing a given language, merely by aban-
doning his own personal prejudices: in addition, he must acquire those of the
native speaker of the language. It is for this reason that acoustic phonetics can
hardly ever be regarded as supplying any evidence for phonologic analysis.
212 A MANUAL OF PHONOLOGY

Yet acoustic phonetics plays the invaluable role of showing us not only how
thoroughly prejudiced necessary for the hearing of a native speaker of a
it is

language to be, but also, at least in the long run, something of the specific nature
of the prejudices. This is perhaps its greatest usefulness.
One factor which differentiates the hearing of a native speaker from that of
a spectrograph is certainly the presence, in the former, of apparatus for the
production and transmission of speech. A human hears a certain virtually identi-
cal burst of noise before [i], before [a], and before [u], but interprets it as [p] in
the first and third cases, as [k] in the second. He makes this interpretation be-
cause, in his own he uses a bilabial articulation to achieve the
articulation,
acoustic effect of that burst when
[i] or [u] follows, but a dorso-velar articulation

to achieve the same burst before an [a]. As pointed out in 02, it may well be
that the whole process by which the hearer's Speech Receiver decodes the speech
signal involves a constant comparison of the signal with the articulatory motions
which the hearer would himself use to produce a similar acoustic effect.
But there is more to it than this. The prejudices of a native speaker, as he
listens to a specific utterance at a specific moment, can have been determined
only by his whole past history as a speaker and hearer of the language. For this
reason, a type of statistical analysis assumes an importance which has only
recently been recognized. In this section we shall first discuss this aspect of
phonologic statistics, and then take up several unsolved problems in point of
view which are intimately related to it.
61. Contrast: Existence versus Importance. In analyzing the phonologic system
of a language, we look for differences which distinguish otherwise similar or
identical utterances. If we know that utterances A and B sound different to
native speakers, then we know that some such phonologic difference is to be
found, though it may be relatively easy or relatively difficult to pin down. By
examining many pairs of utterances, we eventually manage to tabulate all the
features, the differences between which can function to distinguish utterances.
Such a complete tabulation constitutes half of the task of analysis: the other
half is that of deciding just what features, in what arrangement, are present in
any given utterance. This is sometimes quite difficult in a practical way: one
can be convinced (and correctly so) that one has tabulated everything that
counts, and still be uncertain whether, say, a certain word contains feature x
or feature y at a given point.
In the process of tabulation, minimal pairs are the analyst's delight, and he
seeks them whenever there is any hope of finding them. Before analysis is com-
plete, one cannot be certain that a given pair is "minimal" in a strict phonologic
sense that there is but a single difference, at the level of ultimate phonologic
constituents; for this reason, we prefer to mean something more vague but oper-
ationally more useful by the term. We mean simply a pair of utterances usually
quite short that are obviously identical for the most part, though the difference
may be somewhat greater than the strict interpretation would imply. English
blown and grown would not count as a minimal pair for an investigator whose
native language was any of those of the West, but it might well function as one
 A MANUAL OF PHONOLOGY 213

for, say, a Polynesian-speaking analyst. A minimal pair, in this loose sense,

proves that the detectable articulatory difference is phonologically distinctive:
further steps are required before one can be sure whether the contrast is between
one phoneme and another, or between one cluster and another, or between a
phoneme and a cluster, or the like.
As useful as is the method of minimal pairs, it involves one important psycho-
logic drawback. Quite properly, in analyzing a language we feel that the evi-
dence of a single minimal pair (providing it is a genuine one, not the result of
mishearing) is enough to prove a phonologic contrast. But this focusses all our
attention on the fact of the existence of a contrast, and tends to conceal the
equally relevant fact that contrasts may differ in importance. If some contrasts
are more important than others, then that is part of the whole phonologic system,
and we should not overlook it.
Two considerations show us immediately that contrasts do differ in relative
importance: (1) the number of minimal pairs turning on a given pair of phono-
logic elements may be high or low; (2) most actual utterances, in real-life con-
texts, do not stand in anything like minimal contrast with other utterances which
might conceivably be uttered in the same context.
Both points can be illustrated from English. Short ("one-word") utterances
can be listed by the hundreds to show the phonological relevance of the contrast
between [p] and [b] pet bet pit bit, pat bat, flappy flabby, cop cob, pride
: :
,
: : : : :

bride, and so on. (To constitute a valid minimal pair, of course, the two words
of any such pair must be uttered with one and the same intonational pattern;
but this is easy enough.) On the other hand, search as one will, no more than a
small handful of minimal pairs can be found to attest the phonologic relevance
of the contrast between [s] and [$]. For at least some speakers of English, one
or more of the following pairs are minimal, though there are others who make
more distinction: mesher measure; Aleutian allusion; Asher azure; dilution
: : : :

delusion. In some cases an even more peculiar situation is encountered. I have

in my speech, apparently, no word which is always pronounced with /i/. Thus
one pronunciation of the adverb just, /31st/, contrasts with a pronunciation
/38st/, which is the only pronunciation for the adjective just but also an alterna-
tive pronunciation of the adverb. The only minimal pair for /i/ and /i/ is the
adverb just (when so pronounced) versus gist. Unstressed /$/ and // contrast
often in take Hm /teykim/ versus take 'em /teykam/ but the former is sometimes
;

/teykim/ and the latter sometimes /teykim/.

The other point can be illustrated by taking any middle-length English ut-
terance as a point of departure, and looking for closely similar utterances. Con-
sider, for example,

/^Srz now sig31 nifikinsin Sis]./

(which have just pulled at random from a paper-backed detective story, tran-
I

scribing as I would naturally say it in its context). One can achieve a similar,
it

though phonologically distinct, utterance by changing the intonation (say to

/2 33T/); but it is practically impossible to get another reasonable English ut-
214 A MANUAL OF PHONOLOGY

terance by deleting, or replacing, any one of the single consonants or vowels.

Of course, long minimal pairs (apart from intonational ones) can be found, or
coined the point is not that they are impossible, but that they are devised only
:

with relatively great difficulty, and that when one does succeed in devising a
pair, it turns out to be highly unlikely that any single set of circumstances would
be equally open to either member of the pair.
Most workings of language, what is actually
of the time, then, in the actual
said differs,from what might have been said but wasn't, not just in one phono-
logic feature, but in many. There is, so to speak, on the average a fair amount of
phonologic "distance" between the whole set of utterances which are at all likely
to occur in any given set of circumstances. In the technical jargon of information
theory, we say that languages (in phonologic terms) have a certain degree of
redundancy. And this redundancy implies that, most of the time, a certain amount
of distortion, added to the message at one point or another along the route it
follows, leaves the message still intelligible.
Suppose, for example, that someone says:

2
/ ders now sik 31 nifikinsin disj/

Under good hearing conditions, we would notice the peculiarities of pronuncia-

tion, but would have no difficulty in understanding what is said: the indicated
arrangement of phonologic material can represent no arrangement of morphemes
other than precisely that for which we normally expect the phonologic shape
cited earlier. Under poorer hearing conditions, with a sufficiently strong reason
to expect that something of this general nature is going to be said, we might well
even fail to hear the phonologic peculiarities; our Speech Receivers, on the basis
of expectations, w ould
T
convert the sound waves into the stream of phonologic
units first indicated. That is, we would
in fact hear wrongly in terms of the
speaker's phonology but terms of bis grammar (morphemes).
correctly in
Suppose, for a second example, that someone says There's no significance in
this in just the normal way first transcribed above, but that deterioration of the
speech signal between his lips and our ears is such that, when it reaches the
latter point, it has been worn down to something which could be schematically
indicated as follows:

2 31
/ -r- -ow s-g nif- -I-s-n -isj,/

Here we suppose that just over half of the successive vowels and consonants
have "gotten through"; there is actually, in the speech signal as it reaches our
ears, nothing whatsoever to specify the nature of the remaining vowels and con-
sonants, though the spacing-out of the clear signal is such as to show that the
gaps are located approximately as shown. Normally we would be totally unaware

of the deterioration. Not only would we understand in the sense of correctly

decoding the signal into morphemes but we would be unaware of the fact that
we had "really" not heard all the phonemes. The redundancy of the system, all
factors considered, is sufficiently great, and the efficiency with which our Speech
 A MANUAL OF PHONOLOGY 215

Receivers work (with any necessary help from any other component) is so high,
that the gaps are filled inautomatically and correctly at an entirely subcon-
scious level. We say normally, of course there are times when we strain in vain
to understand something heard poorly, and times when we have to go through
a conscious trial-and-error process, lasting a measurable number of seconds,
before we finally do get it. But in the situation as we have described it, the
analyst would have to assert that, as received by the hearer, the 'phonologic
structure of the utterance was entirely normal.
Any casual observation of language in action in everyday life shows that
acoustically accurate hearing, accompanied by misunderstanding, occurs most
often in the case of utterances with less-than-average differentiation (minimal
pairs!), or in situations where the circumstances do not help to pin down what
may reasonably be said. Consider the business of introducing people (in our cul-
ture).There are prescribed formulas for this. We know, when the circumstances
arise,approximately the words which are going to be used by the introducer and
by the two who are meeting. Consequently, no one has to speak very distinctly,
and everyone catches on anyway. But there is one point in the operation at which
our expectations are almost completely neutral, and thus of no help at all. This
is the point at which a person's name is spoken. Most of us, having slurred the
formulas, tend also to slur the person's name: our hearers, understanding the
formulas because they know what to expect, often miss the name completely.
To summarize the above: Phonologic differences function to keep utterances
apart, and differences of articulation are phonologic if they every perform this
function. But most of the time a given phonologic difference works along with
others, rather than alone; and some do more of the work of keeping utterances
apart than do others.

62. The Measurement of Functional Load. It is the last clause of the last sen-
tence above which concerns us in this subsection. The way in which some phono-
logic contrasts do more of the work of keeping utterances apart than others is
the matter of functional load.
It is easy to get some rough notion of the relative functional loads carried
by different contrasts within a phonologic system. Assuming that two phonemes,
x and y, can contrast at all, then the functional load carried by the contrast will
be greater if both x and y have relatively high text frequencies than if one has
a high frequency and the other a low frequency, and greater under those second
conditions than if both x and y have low frequencies. The large number of pairs
of English words differing only in the presence of /p/ or of /b/ at a certain point
obviously implies the potential existence of a larger number of long minimal
pairs turning on /p/ and /b/ than on, say, /p/ versus the relatively low-fre-
quency //; yet one will expect a larger number of the latter than of minimal
pairs turning on // versus the very low-frequency //. Therefore we can be
pretty safe in concluding that the /p/ /b/ contrast carries a higher functional
:

load than the /p/ // contrast, and the latter a higher load than the // /&/
: :

contrast indeed, the functional load of the last contrast must be vanishingly
small.
216 A MANUAL OF PHONOLOGY

Functional load has been discussed in the past mainly in the context of di-
achronic linguistics. In the light of the fact that the sole function of phonologic
differences is to keep utterances apart, it has been proposed that the probability
that a given phonologic contrast will be extirpated through sound change stands
in inverse ratio to the functional load carried by that contrast. The theory is

at least negatively sound : it is true that sound change never leads to a complete
obliterating of all phonologic contrasts and hence to a destruction of the lan-
guage. The notion is that certain very general factors, perhaps collectivizable
under the label "laziness," lead to as much obliteration of distinctions as speakers
can get away with, while certain other factors, chiefly the practical need for ef-
fective communication, stand in the way of obliteration, forcing speakers to
speak more clearly whenever the point of unintelligibility is reached. But if a
particular contrast carries a low enough functional load, then this correcting
factor will not be apt to come into play, and the contrast may well be lost. Of
course, it is recognized that other factors play a part too: low functional load in
itself does not guarantee loss of contrast. Thus if in a given system the difference

between /&/ and /n/ is of vanishingly small importance, there is nevertheless

no great probability that they will fall together, since they are phonetically too
diverse.
Whether this theory is sound in a positive sense is another matter. Granting
that the functional loads of some contrasts are so low that loss of contrast would
not impair communication, we must also find out whether any contrast carries
so high a functional load that loss of contrast would impair communication. In
order to test this, we must have
(1) a consistent technique for quantifying func-
showing what relative functional load is the critical
tional load; (2) evidence
maximum, which cannot be lost without intolerable impairment of communica-
tion. The procedures of modern information theory show us how to obtain (1);
(2) is a matter for the future. We
mathematical technique for
shall outline the
the measurement of functional load; unfortunately, the determination has not
actually been made yet in any case (the amount of counting and computation
necessary is formidable), so that we can give no example.
Let the phonemes of a language L be h 2 ., 4> m Let the relative frequency
<f> <f> ,
. . -

771

of 4>i be p { then ^2pi

;
= 1.

The first-order approximation to the entropy of L per phoneme-occurrence is

then
m
# = i Z Kpi)
where
I(Pi) = -Pi log2 Pi-

Next let {0i0y} be the class of all two-phoneme sequences in L; the relative
frequency of $,#/ = p#. The second-order approximation to the entropy of L
per phoneme-occurrence is
1
m
 A MANUAL OF PHONOLOGY 217

Similarly, the nth-order approximation is

1
m
Hn = -
n
2
l'i,2.*3i" ,i n =l
I[p(iii2 ... in)],

where p(i&2 . . . in ) is the relative frequency of the w-phoneme sequence

#i0t, . . fa n -

The entropy H is defined as

H = \\mH n ,

n>co

providing such a limit exists. To assume that the limit does exist is to assume
that the phonologic pattern of any language is ergodic in the weak or Wienerian
sense.
If in an imagined phonologic system L', with m phonemes, all the phonemes
were constantly equiprobable, then the entropy would be

H' =H[ =E/( P = i) -log, I

In the actual system L the relative entropy is

F- H
and the redundancy is

R = I - E.

Now suppose that in a system L we wish to measure the functional load car-
ried by the contrast between phonemes fa and fa. We construct a pseudo-system
L* in which all occurrences of phonemes fa and fa are replaced by occurrences
of a single phoneme fa, but in which everything else remains as in L. This is tanta-
mount to putting L through a single diachronic change, in which fa and fa fall
together but everything else remains unchanged.The entropy of the pseudo-
system L* can be computed as we compute that of the system L; call this en-
tropy H*.
The entropy-loss as between L and L* is then
H- H* ,

and the functional load carried by the contrast between fa and fa is

H - H*
H
That is: the functional load carried by a given contrast is the ratio of the en-
tropy which would be lost if the contrast were abolished to the entropy of the
unchanged system. This formulation has the advantage (as over against one
in which entropy-loss is directly equated with functional load) that the sum of
the functional loads carried by all contrasts is just the entropy of the whole
system.
218 A MANUAL OF PHONOLOGY

The key figure for functional load which cannot be sacrificed without in-
tolerable impairment of communication is tied up, somehow, in the relationship
between entropy-loss and redundancy. The best available estimates so far avail-
able suggest that all phonologic systems have a redundancy not far from .50,
measured, however, not in terms of phoneme-occurrence but in terms of units
of time (so that, everything else being equal, redundancy would decrease as
rate of speech increased). If this figure is anywhere near correct, then presumably,
if the redundancy mounts much higher, the system is inefficient and people be-

come lazy enough to bring it down again, while if it falls much lower, speech
becomes unintelligible and people speak with enough care to bring it up again.
The problem, then, is to find the functional relationship between functional load
and redundancy, so that one can predict how redundancy would be altered by
the loss of a contrast bearing any given functional load; to determine how low
the redundancy can go without serious impairment of communication; and,
finally, to see whether any one contrast in any language ever carries a great

enough proportion of the total functional load for its loss to have such disastrous
consequences. Neither the mathematical nor the empiric aspect of this problem
is clear to me at the moment.
It should hardly be necessary to dwell on the synchronic implications of func-
tional load. A tabulation of all the phonologic contrasts in a system, showing the
relative functional load carriedby each, would constitute a valuable addition to
any description of a phonologic system, and might help us considerably in de-
ciding what to try to find on spectrograms of natural rapid speech. Although the
amount of counting and figuring which would be necessary, even for an approx-
imation, is enormous, there are various sorts of computing machines which could
be used to reduce the labor, and we should not shy away from the task.
63. Sporadic Intersectionand Shimmering. An utterance has a phonologic
structure as produced by a speaker: this structure is dependent on the nature
it is

of the signals sent by the speaker's Phoneme Source to his Speech Transmitter
(Figure 1) and on the concurrent functioning of the speaker's Speech Receiver,
operating on feedback input. The speech signal has no phonologic structure in
its own right. But the utterance once again has a phonologic structure as a hearer

receives and interprets it. Although we can properly assume that by and large
the phonologic structure of a given utterance is the same for speaker and for
hearer, we certainly cannot assume that this is always true. Indeed, if a speaker
is addressing several hearers, then there may be, all in all, as many differing

phonologic structures for his utterance as there are participants. This raises a
problem which is probably largely theoretic, and of no great practical importance:
if the phonologic structure of an utterance differs for speaker and hearer (s),
which is the phonologic structure of the utterance? Or (and this is still part of
the question) is our first query meaningless? Must we simply recognize a multi-
plicity of simultaneous phonologic structures for a single event?
This is a large question, and only one facet of it will concern us here: the pos-
sible relationships between the speaker's actual articulatory motions and the
phonologic structure of the utterance for a single hearer. Assuming that for the
 A MANUAL OF PHONOLOGY 219

most part the relationship is just as we should normally expect it to be on the

basis of accurate hearing and interpretation we must allow for two phenomena
not normally recognized in phonologic analysis: sporadic intersection and shim-
mering.
Articulation normally sloppy and careless. In especially careful speech, one
is

may the biscuits with very clearly voiceless and aspirated /p/'s
say please pass
and a very clearly voiced /b/. In normal speech one or the other of the intended
/p/'s may in fact be not only unaspirated but also even slightly voiced; or the
/b/ may be entirely voiceless and even slightly aspirated. Regardless of the
care or carelessness of articulation, the phonologic structure of the utterance for
a given hearer depends not alone on the articulation, but also on how he inter-
prets it. Suppose that on a given occasion a speaker says please pass the biscuits
in such a way that the intended /b/ is completely voiceless and a bit aspirated.
If the hearer "notices" this, then the phonologic structure of the utterance for
him is please pass the piscuits. But often he will not "notice" it at all: the phono-
logic structure for him will be just that "intended" by the speaker, and thus in
disagreement with the speaker's actual articulation. Under these conditions, we
have an articulatory motion which would elsewhere be classed as a /p/ being
received and treated, because of the whole context, as a /b/. This is an instance
of what we mean by sporadic intersection. By definition, sporadic intersection
cannot be observed by the hearer for whom it occurs since if the deviation is

observed, then there has been no sporadic intersection. But the phenomenon is

observable, since the analyst can, so to speak, stand to one side of a speaker and
a hearer and listen with care for things which the hearer misses.
It will be noticed that sporadic intersection, as defined, bears only a super-
ficial resemblance to the old analytic mistake of "complete intersection." The

doctrine of complete intersection holds that a given bit of articulation, as heard

and as intended, may belong now to one phoneme, now to another, even in one
and the same phonologic environment. This doctrine is the product of confusion

of levels phonologic and morphophonemic. Sporadic intersection is "complete"
in the sense that not all intended-and-heard /b/'s, for example, are actually
articulated in a way which from all intended-and-heard /p/'s, but
is distinct
apart from this it is from so-called "complete intersection."
entirely different
Shimmering is closely related to sporadic intersection, but is morphophonemic
rather than phonologic. Suppose that in the circumstances described above,
where the speaker makes a completely /p/-like /b/ by accident, the hearer
"notices" the error. Because he "notices" it, the utterance is phonologically
please pass the piscuits. But this does not necessarily prevent understanding
(compare the discussion in 61). The utterance, as phonologically understood,
can still represent only one arrangement of one set of morphemes. There is no
word piscuit in contrast to biscuit: piscuit is only a rare and sporadically occurring
alternate morph representing the same morpheme usually represented by biscuit.
Shimmering, then, is a kind of
morphophonemic alternation a kind in addition
to all the sorts usually spoken about; a kind which is in a sense least relevant

in the analysis of a language, yet which cannot be ignored. Although most

220 A MANUAL OF PHONOLOGY

morphemes are represented usually by just one, or by just a few, distinct morphs,
every one of these clearly distinct morphs is surrounded by a family of closely
similar ones which occasionally occur in its place. Shimmering is sporadic and
quite unpredictable; all other types of morphophonemic alternation occur under
describable conditions with a considerable degree of predictability.

64. Analytic Norms. We come, finally, to a question for which there is cur-
rently no clear answer: the choice between frequency norms and clarity norms
in phonologic analysis.
The use of clarity norms in phonologic analysis is somewhat comparable to
the use of spoon-feeding in acoustic phonetics. In most languages, if not in all,

there is a prescribed pattern for extra-clear speech, to which one resorts when
normal rapid speech is not understood, or when certain social factors prescribe
it. In English, for example, there is a clarity-norm pronunciation of matter and
latter which involves a and aspirated apical stop in the middle,
fully voiceless
and a clarity-norm pronunciation madder and ladder which involves a slightly
of
longer stressed vowel in the first syllable, followed by a fully voiced and some-
what prolonged apical stop. Some forms may not occur at all in clarity-norm
speech: /did5/, /wud25/, and /wadiya/ {Did-ya, would-ya, what do ya) are not
simply pronounced more slowly, but are replaced by /did yuw/, /wud yuw/,
/hwgt duw yuw/ or the like. Clarity-norm speech presumably has a higher re-

dundancy (per unit of time slower articulation is almost mandatory) than
ordinary speech. In clarity-norm phonologic analysis, one deals exclusively
with clarity-norm pronunciation in the first place, until one's picture is more or
less complete; then one admits also any more rapid forms which fit easily into
the picture as already drawn (except, perhaps, for speed of speech), but excludes
others. Fast forms which do not easily fit the picture based on clarity-norm
speech are regarded as mere deviants: their phonologic structure is taken to be
the same as the clarity-norm alternants, and the fuzziness is regarded as a matter
of personal or momentary idiosyncrasy.
Frequency-norm analysis insists, in theory, on accepting for analysis any
utterance which is produced by a native speaker and understood, or under-
standable, by other native speakers. The choice of norm is involved in the gather-
ing phase of analysis, rather than in the collation phase (3); our description of
gathering in 31 assumes frequency norm rather than clarity norm. The term
"frequency" in our designation refers to the assumption that in interpreting the
speech-signal, the heareris governed by the relative frequencies with which he

has previously heard various acoustic features as representations of various

phonologic units; this, of course, is tied up intimately with the discussion of
61 and 62.
It will not do for us to sneer at either of these guiding principles in phonologic
analysis. Most today would probably feel, offhand, that clarity-norm analy-
of us
and only frequency-norm analysis worth undertaking. A few
sis is unjustified,

may have the other preference: certainly earlier phonemic work was guided
largely by the clarity norm, a fact which led Stetson to object strenuously to
the whole phonemic theory. There is actually some doubt as to whether we ever
 A MANUAL OF PHONOLOGY 221

succeed in following completely one norm or the other. We tend to prefer the
frequency norm, but we perhaps do not accept all its consequences; where we
refuse to accept its consequences, we are referring to the clarity norm instead.
Clarity norm phonology has the merit (if it is a merit) of considerable simplifica-
tion: one is faced by neither sporadic intersection nor shimmering; but one must
suspect that the resulting phonologic picture isnot "pure," having rather a
certain intermixture of what belongs properly inmorphophonemics. But if one
insists on following the frequency norm, then before proceeding from phonology
to grammar one usually undertakes some simplification of the phonologic picture
which, in effect, is a deletion of phonologic complications which seem to be of no
morphophonemic relevance, and which amounts to a partial reestablishment of
the clarity norm.

Here is an example from Midwestern American English at least my own
speech. Clarity norm procedures show only two apical stops in contrast: /t/ and
/d/. Matter and latter contain the first, madder and ladder the second. Frequency
norm procedures force the recognition of a more complicated situation. Most of
the time, my speech does not differentiate matter and madder at all, nor latter
and ladder. The single common pronunciation of matter and madder has a rela-
tively short stressed vowel followed by a voiced apical flap. But sometimes madder
and ladder are pronounced with a lengthened voiced stop, and sometimes matter
and latter are pronounced with a voiceless aspirated stop. We are forced to recog-
nize not two but three (frequency-norm) phonemes: /t/, /j/, and /d/. The
clarity-norm morph for matter includes /t/, and the clarity-norm morph for
madder includes /d/; but in each case shimmering can produce an alternate
morph with /j/. There is no valid procedure by which the flap can be assigned
to the same phoneme as one or the other of the stops, and certainly we cannot
assign it to both (since that would produce complete intersection). The func-
tional load carried by the contrast between /t/ and /j/, or by that between
/j/ and /d/, is certainly extremely small, but this does not justify any simpler
treatment so long as we stick to frequency-norm phonology.
I do not think there is any simple answer to the irethodologic questions pre-
sented by these two analytic norms. So long as phonologic analysis of a language
is only partial and the vast majority of the reports so far available are certainly

not complete the problems probably do not arise. In field work, we are by defi-
nition apt to discover first the contrasts of greatest functional importance, and
onry later those which carry less load: differences of an embarrassing nature be-
tween frequency-norm and clarity-norm analysis usually lie in the realm of con-
trasts with low functional load. But when analysis is carried far enough to dis-
cover discrepancies between the yield of the two norms, we have some sort of a
choice to make, and the choice should be based on clearly formulated reasons.
Probably, as in so many other choices of this sort, we shall in due time discover
that certain contexts and purposes justify the option of one norm, other con-
texts and purposes the option of the other.
 NOTES
Bibliographic references below refer to the Glossarial Index and Bibliography which
follows the notes. Reference is made by author's name and year of publication, years in

the twentieth century being represented by the last two digits only (e.g. "Bloomfield
33"); a lower-case letter is added when the bibliography lists more than one work by the
given author for the given year. The digits "00" in place of a year indicate information re-
ceived informally from the individual named. Page numbers are given, when necessary,
after a decimal point: "K. Pike 43.16-24."
The notes are suggestive only, and the bibliography is in no sense intended as a complete
list of publications bearing on phonology. Works which are not listed or referred to are not
thereby implied to be unimportant in my estimation in some cases, of course, they have not
:

come to my attention,and in others their content impressed me as paralleling, more or

less, that of works which are herein cited.
01. Typology: Trubetzkoy 29, 39; TCLP passim; Voegelin 51.
Hierarchical approach: Bloomfield 33 ch. 5-8; Hjelmslev 43; Kurylowicz 49; Pike and
Pike 47. See now also Prieto 54.
The quotation from Bloomfield 33.20. "Hocus-pocus" and "God's-truth" Householder
is :

52. Bloch's postulates: Bloch 48.

020. On the controversy: K. Pike 47a; Hockett 49; K. Pike 52; also compare the Prague
approach in general, perhaps as summarized in Martinet 49a, with Hockett 51.
021. The model presented here has been developing since 1936, when a very sketchy form
of it was included in my M.A. Thesis (unpublished) at Ohio State University; the original
impetus is no doubt Bloomfield's Jack and Jill diagram (33 ch. 2). Information theory (Shan-
non 49, Shannon and Weaver 49, Hockett 53a) affords an excellent frame of reference within
which to work out the tenuous notions with which I had earlier been working. In Hockett
53a, under the heading Phonology and Tactics, the present model is hinted at; a fuller form
in Hockett 52d. Also, now, compare the approach in certain sections of Osgood 54.
0210. Problems assigned to "biophysics" and to linguistics (or "sociobiology") cf. ;

Hockett 48a.
Neuron-to-wire linkage: that is, connecting a living neuron (or bundle thereof) to a wire
in such a way that a nervous pulse traveling down the neuron will induce a voltage pulse in
the wire, and vice versa. This problem is a crucial one in modern prosthetic theory, but, so
far as I know, is nowhere near solution.
0212. Experiments on feedback: Joos 00. Role of Speech Transmitter in perception and
understanding of speech: Joos 48.
0215. The term "meaning" is also used in some other ways for example, one can say that
:

the meaning of a morpheme is the transitions of G.H.Q. -state which its reception induces.
This type of use of the word (though certainly cast in far different terminology) underlies
the European tendency to think of meaning as "in" the linguistic system, in contrast to the
Bloomfieldian slant. The conflict is purely one of terminology.
022. Hjelmslev's system: Hjelmslev 43.
023. On discreteness of contrast cf. Pike 46 ch. 7; on duality of pattern, Martinet 49c.
Mazateco whistle speech: Cowan 48.
03. On code noise (or "semantic" noise) Hockett 52a.
:

04. Imitation-label: K. Pike 43.16-24. Bohtlingk 1851.

I. There have been dozens of treatments of articulatory phonetics. The most thorough

and up to date, in many respects, is part II of K. Pike 43. Heffner 49 is more recent and
covers the physiology of speech more completely; but one should consult also McQuown 51.
10. The
list of the organs of speech: Bloch 48.
II. Stetson 45; Twaddell 53.
14. Term "velic": K. Pike 43.
150. Terms "vocoid" and "contoid": K. Pike 43. However, we do not quite follow his
usage: ours is rather looser..
222
 : :

A MANUAL OF PHONOLOGY 223

152. Our classification in general follows that of Bloch and Trager 42.
164. The hat diacritic for affrication is Jakobson's usage (00).
2. The languages discussed in this section are listed alphabetically in the Glossarial In-
dex and Bibliography. There each language is identified and, if necessary, located; and
references are given to the sources of information.
211. Following Trager and Smith 51, with modifications worked out by Joos, Agard, and
myself. K. Pike (46) disagrees with the notion that the amount of terminal rise (subsumed
indifferently by our /f/) is nondistinctive. I feel no certainty on this point. I am by no
means satisfied that the system as here reported subsumes everything that must be sub-
sumed. The "metalinguistic" approach of Smith and Trager (Smith 52a,b) seems to turn up
additional relevant phenomena, but the basis on which they exclude these from the ordinary
phonologic system of the language seems very fragile.
212. Remarks on intonation in European languages are based on observations at Cornell
and on informal reports of observations at the Foreign Service Institute. On Shawnee and
Arapaho: observations made at the 1952 Linguistic Institute at Indiana University, in the
company of H. L. Smith Jr., Carl F. Voegelin, and others. Mazahua: E. V. Pike 51.
22. "Syllables" have bothered phonologists for a long time. They have formed either the
target of discussion, or an important secondary consideration, in such articles as Sommer-
felt 31, 36, Stetson 36, Hjelmslev 36, 36-7, Kurylowicz 49, O'Connor and Trim 53. The most
serious difficulty encountered in these discussions is an insufficiently wide coverage of lan-
guages of differing types. I made this point in Hockett 52b, specifically in criticism of
Kurylowicz 49. In so doing, I missed the positive features of Kurylowicz's treatment which,
with proper generalization, seem to afford a workable frame of reference, of the sort I try
to present here.
230. The notion of isolability and its use as a criterion in IC-analysis is a transferral to
phonology of the grammatical notions of freedom and bondage. On accentual systems:
Trager 41.
2311. Recognition of zero is like Harris's notion of "zeroing out," but I 1 we tried to
make the criteria more realistic (Harris 51.337f).
242, 243. This elaborates the frame r /esented in Hockett 53b.
244. A comparison with Trubetzkoy plan for the treatment of vowels (29, 39) will show
s
that "pare off" a great deal mor* material before calling the remainder a "vowel (pho-
I
neme)"; but for the rest I find it possible to follow Trubetzkoy 's plan quite closely, and in
this survey of vowel systems include languages discussed by him, on his testimony alone.
As pointed out at the end of 2440, our exclusion of the 3X2 pattern may be not only
arbitrary, but wrong. Trager proposed the 2X3 interpretation for the oral vowels of Taos
in Trager 46 (2442), but has more recently suggested the 3X2 possibility (Trager 00)

i o u
e a o .

These matters are not always determinate in a purely phonologic sense. When they are not

and perhaps even when they are it is interesting to see what is suggested by morphopho-
nemic considerations. Thus Dagor Mongol has a vowel-harmony system which pairs off
high and low vowels to give

i i u
e a o ,

although purely on their phonetic properties one might class them differently.
251. Cf. Fischer-J0rgensen 52, and her very excellent survey and bibliography of dis-
tributional studies.
2521. The procedure used in analyzing and classifying obstruent systems is in part
reminiscent of Trubetzkoy 's (39), though the terminology differs. Cf. also Catford 39,
Jakobson 39.
262. Other interesting examples of componential analysis, showing various predilections


224 A MANUAL OF PHONOLOGY

Hockett 47b; Jakobson 49; Jakobson and Lotz 49 Bloch 50 Martin 51. The strictly phonetic
; ;

componential descriptions of K. Pike 43 should also be compared.

273. Frequency figures for English spirants were obtained by converting those for all
English consonants given in Bloomfield 33.137.
30.Boas on Kwakiutl: lib, 47.
Boas (in 11a) is often thought to have presented one of the earliest formulations of the
phonemic principle. His statement (p. 16) begins as follows: "One of the most important
facts relating to the phonetics of human speech is, that every single language has a definite
and limited group and that the number of those used in any particular dialect is
of sounds,
never excessively large." This is equivalent to about one-half of the Jonesian brand of
phonemics. The assumption is clear, that one could in time make an exhaustive list of all
the "speech sounds" to be found in any language at all, so that the stock of speech sounds
used in any one language would appear as a larger or smaller selection from this stock. But
the principle of contrast and complementation, which in effect forms the other half of
Jonesian phonemics, is missing. Thus Boas's assumption would in no way lead one to regard
certain pairs or triples of speech sounds as related with each other in a particularly close
way (i.e., as allophones of a single phoneme), nor to simplify one's notation accordingly,

even as a matter of convenience. And, both for Boas and for Jones, thinking is still in terms
of things rather than differences.
In any case, throughout his linguistic career Boas never used what most of us would call
a phonemic notation. However, it must be said in his defense that part of the reason may
have been an unformulated, but instinctively sound, objection to the rather drastically
oversimplified brand of "phonemics" which was current in this country during much of the
third and fourth decades of the century.
Synchronic phonologic results sought via the philological method (interpretation of
written records) Fry 41 Harris 41 Emeneau 46; Hall 46b Moulton 48; Jensen 49; Stockwell
: ; ; ;

and Barritt 51 (criticized by Kuhn and Quirk 53) Fairbanks 52; Lunt 53. Not all of these
;

devote any discussion to the assumptions and methods involved. Bodman 54, dealing with
Chinese, is forced to proceed quite differently because of the radically different nature of
the writing-system. Similar in aim, but recessarily different in approach, are studies where
reconstruction via comparison, rather than directly the interpretation of written records, is
involved: Lehmann 52, Martin 53, Moulton 54.
31. The gathering aspect of phonologic work does not seem to have challenged the at-
tention of European investigators (despite the extensive field work which some of them

for example, Trubetzkoy himself have done). In England, similarly, there has been little
tendency to theorize and generalize about gathering; it is interesting to note how many of
Daniel Jones's studies in the phonetics of this or that language have been done in close
collaboration with a native speaker to whom a fair amount of phonetic training has been
given. The background for our discussion in this section is thus largely American: Bloch
48 (the earlier postulates) K. Pike 47a; Voegelin 49; Hockett 49; Harris 51 Hockett 52c.
; ;

321. Wells 47; Pittman 48; Hjelmslev 43.

3220. Swadesh 34a; Hockett 42; K. Pike 47b; Jones 50.
3222. Pike's aphorism: Pike 47b. Moulton 54.26.
3231. Our IC-analysis of Fox yields results somewhat different from those presented in
262.
3232. "Defective" phonemes: Swadesh 34a; in practice in such reports as Haas 41a, 41b,
Swadesh 34b. The Prague approach: TCLP passim; Martinet 49; Hockett 51.
324. The approach to juncture in Bloch and Trager 42 is not logically valid. In his postu-

lates (Bloch 48), Bloch removes the illogicality and the possibility of recognizing juncture
along with it. Harris's habit of calling juncture a "zero phoneme": Harris 51. According
to Joos (00) there is some reason to believe that English juncture does involve a certain
measurable and detectable feature in all occurrences: a slight lengthening of the "preced-
ing" phoneme, be it vowel or consonant. Our approach does not render this unpleasing, if
true; but does allow for the recognition of a juncture in some cases where perhaps no such
constant phonetic feature is to be found.
 .

A MANUAL OF PHONOLOGY 225

325. Jakobson 49; Jakobson and Lotz 49; Harris 51.

Primary and secondary features: cf. the labelling of certain features as "redundant"
in Jakobson, Fant, and Halle 52.
4. Our aim here is to show how an empiric approach can (and, indeed, must) in due time
reach the type of abstract scheme proposed by Hjelmslev (43). It should be added that an
abstract scheme not reached via an empiric approach remains pure mathematics, with no
necessary relation to language.
Mutual convertibility: Bloch 48. In variance under transformations between mutually
convertible systematizations Chao (00)
:

The relevance of articulation and acoustics despite the importance of abstraction:

Haugen 54, criticizing certain points in the generally excellent book by Weinreich 53.
5. The special r61e of Martin Joos in the preparation of this section is described in the
Preface. Joos 48 remains the standard fuller treatment, and contains references to the im-
portant earlier contributions to the field. Here we try to summarize the results of work done
since the preparation of that book; newer items include Lisker 48, Delattre 48; Peterson
61, Delattre 51, Delattre and others 51, Liberman and others 52, Cooper and others 52,
Delattre and others 52a,b, Jakobson and others 52, Schatz 54.
512. The spectrograph is more fully described in Joos 48. At the present writing, I know
of only one pattern playback: it was built and is in use at the Haskins Laboratories in New
York City. It is described somewhat more fully in Liberman 52.
520. Transducer study, as a phase of communication theory: Shannon and Weaver 49,
Shannon 49.
5311. Delattre and others 51, 52a.
5312. Delattre 48, Delattre and others 52a.
5313.The point-vowel values were measured on spectrograms in Joos 48 and in Potter,
Kopp, and Green 47 by H. A. Gleason Jr., for inclusion in a forthcoming book of his. The
vowels shown by curves are from Joos 48, Fig. 25 and accompanying text.
5314. One-formant matching: Delattre and others 52a.
5315. We are forced to reject the tentative conclusion of Delattre 51. 870f.
5316. Joos 48; Delattre and others 52b.
532. Joos 48; on bursts, Liberman and others 52, Cooper and others 52.
533. [v] attained by shortening [f]: Delattre (00).
534. Pruning back: Joos (00). Psycho-acoustic tests with the pattern playback: Cooper
52, Delattre 52b.
54. The assumption that spectrograms might be read almost as "transcriptions" in a
new phonetic alphabet apparently lies behind the "visible speech" notions in Potter, Kopp,
and Green 47. Jakobson, Fant, and Halle 52 also show an oversimplified assumption of this
sort. In both cases the factor of spoon-feeding is at least in part responsible.

61. Redundancy: as in information theory (Shannon and Weaver 49).

62. Functional load, in diachronic linguistics: Martinet 39, 49, 52, 53, and in various
other articles. The mathematics for the measurement of functional load is a simple deriva-
tion from the information -theoretical formulations of Shannon (Shannon and Weaver 49).
64. Stetson 45.9-17.
 GLOSSARIAL INDEX AND BIBLIOGRAPHY
Here we present, in a single alphabetization, everything that the reader might wish to
find in alphabetical order. The entries are of three kinds:
(1) Language names, Each is followed by: (a) indication of its affiliation, and
in italics.
ifnecessary of its location; (b) reference to my sources of information; (3) reference to the
subsection or subsections of 2 of this manual in which substantial information on the
phonology of the language is given.
Language names are spelled in the generally accepted ways, omitting superfluous dia-
critics. Those spoken in the U.S.S.R. are spelled in English following W. K. Matthews 51.
North American Indian languages are not located, since the location can be determined by
referring to Voegelin and Voegelin 44. Nor do we locate languages such as English, French,
German, Russian, Japanese.
Sources are referred to by name of author or authors, year of publication if necessary,
and following "a," "b," etc. if necessary. Years in the twentieth century are represented by
the last two digits: 43 = 1943. "00" means, however, information via personal communica-
tion or in unpublished notes. "H" means that I have heard the language; "Ha" that I have
heard it and done a little analytical work on it; "HA" that I have worked fairly extensively
on it.
(2) Technical and semi technical terms, or names of topics, in Roman type. These are
sometimes accompanied by a brief definition or comment, and always by a cross-reference to
the sections of the manual where they are defined, used, or discussed.
(3) Bibliographical listings, by author or authors, then by year of publication. Any
unobvious abbreviations are listed in proper alphabetical order, with explanation.

Aberle, D. F., and Wm. M. Austin, 51. A lexical approach to the comparison of two Mongol
social systems, SIL 9.79-90.
Abkhaz. North Caucasian. Trubetzkoy 39. 2442.
abstraction, 4.
accent d'insistence (in French), 212.
accentual system, 23.
acoustic allophones, 54.
acoustic and articulatory, 04.
acoustic phonetics, 5. Defined, in terms of problems, 520.
acoustics: the branch of physics which studies sound, 51.
actualization, 3222.
Adyge. North Caucasian. Trubetzkoy 39. 2442.
affricates, 164.
Agard, Frederick B. 53. Noun morphology in Romanian. Lg. 29.134-42.
Alabama. Muskogean, eastern division. Haas 41. 2431, 252112, 2521321, 252213, 252225.
aleph = glottal catch, 12.
Algonquian, Central: comprises Cree, Fox, Menomini, Ojibwa, Potawatomi, Shawnee.
252213.
Algonquian, Prolo -Central-. Bloomfield 46; A. 2441.
allophones, 31 end.
, acoustic, 54.
, logic of, 3222.
allophonically correct transcription, 31 end.
Amahuaca. Panoan; on Sepahua, Inuya, and Sheshea rivers in eastern Peru and into western
Brazil. Osborn 48. 2441, 252112, 2521321, 252213, 252225, 2525.
amplitude, 511.
AmSp = American Speech.
226
 A MANUAL OF PHONOLOGY 227

Andrade, Manuel. 39. Quileute. HAIL 3.

Andrews, Henrietta. 49. Phonemes and morphophonemes of Temoayan Otomi. UAL 15.213-
22.
Apachean (southern branch of Athabascan) Chiricahua, Navaho. Hoijer 43.
: 2312, 2441, 2445.
Arapaho. Plains Algonquian. Voegelin 00; H. 212, 252112, 252212, 252221.
archiphoneme, 3232.
Armenian, East. Indo-Hittite (Indo-European?); Soviet Armenia. Stevick 00; H. 252112.
arrangement of ICs, 321.
arrangements, contrasting, 321.
articulation, point of, 1520.
, position of, 1520.
articulator, 1520.
articulatory and acoustic, 04.
Arunta. North-Central Australia. Sommerfelt 39. 2441, 252112, 252131, 252213, 252225, 2525.
Aschman, Herman P. 46. Totonaco phonemes. IJAL 12.34-43.
aspirated stops, 162.
Athabaskan, Southern: see Apachean.
Athabaskan, West Coast and Northern. Li 30; Hoijer 43. 2212.
Austin, William N. 52. A brief outline of Dagor grammar. SIL 10.65-75.
Awar. North Caucasian. Trubetzkoy 29. 2442.
Aztec: see Nahuatl.

B
backness vs frontness of tongue, 1510.
Badaga. Dravidian; Nilgiri Hills in Southern India. Emeneau 39. 2431, 2442, 2445, 2446,
252112, 2521322, 252214, 252225. 25223, 2525.
balance, 271.
Bannock. Mono-Bannock, Plateau Shoshonean, Uto-Aztecan. Liljebled. 252113, 2521322,
252214,252221,2525.
Bariba. "West-Sudanic" (Afro-Asiatic); in Cercles of Parakou, Kondi, and Natitongou in
Dahomey and in adjacent parts of Nigeria. Welmers 52a. 22112, 2322, 2442, 2443, 2445,
252114, 2521322, 252213, 252225, 2525.
Bella Coola. Coast Salishan. Newman 47. 2213, 252113, 2521332, 25220, 252213, 252225, 2524,
2525.
Bender, Ernest, and Z. S. Harris. 46. The phonemes of North Carolina Cherokee. IJAL
12.14-21.
bending of formants, 534.
bilateral symmetry, of speech tract and articulation, 10.
binit: the unit of measurement and entropy; the amount of in-
of informational capacity
formation carried, on the average, by a system within which the source has only two
signals and is equally likely to emit either. All other types of system can be handled in
terms of this by proper mathematical conversion. [Usually called bit in the literature.]
Bisayan (of Cebu). HA. 2442.
Bloch, Bernard. 48. A set of postulates for phonemics. Lg. 24.3-46.
50. Studies in colloquial Japanese IV: Phonemics. Lg. 26.86-125.
,

and George L. Trager. 42. Outline of Linguistic Analysis. Baltimore.

,

Bloomfield, Leonard. 17. Tagalog Texts, with Grammatical Analysis. University of Illinois
Studies in Language and Literature 3: 2-4. Urbana.
24a. The Menomini language. Proceedings of the Twenty-first International Congress
.

of Americanists 336-43. The Hague.

24b. Notes on the Fox language. IJAL 3.219-32.
.

28. The Plains Cree language. Proceedings of the Twenty-second International

.

Congress of Americanists 427-31. Rome.

33. Language. New York.
.
228 A MANUAL OF PHONOLOGY

Bloomfield, Leonard. 39. Menomini morphophonemics. TCLP 8.105-15.

. 42. Outline of Ilooano syntax. Lg. 18.193-200.
. Spoken Dutch. New York.
45-6.
Algonquian. LSNA 85-129.
. 46.
Boas, Franz. 11a. Introduction. HAIL 1.1-83.
lib. Kwakiutl. HAIL 1.423-558.
.

47. Kwakiutl grammar, with a glossary of the

. suffixes. Transactions of the American
Philosophical Society 37:3. 199-377.
, and Ella Deloria. 41. Dakota grammar.Memoirs of the National Academy of Sciences
Washington.
23:2.
Bodman, Nicholas C. 54. A linguistic study of the Shih Ming. Harvard-Yenching Institute
Studies 11. Cambridge.
Bohtlingk, O., 1851. Die sprache der Jakuten. St. Petersburg (= v. 3 of A. T. von Midden-
dorf, Reise im aussersten Norden und Osten Sibiriens).
Bororo. Ge; south-central Matto Grosso, Brazil. Lounsbury 00. 252111.
boundaries, redrawing (between "successive" sounds), 3222.
brackets, use of, 10.
breathgroup, 11.
Breton (St. Pol-de-Lton) . Celtic, Indo-European. Brittany, France. Smith 46. 252111,
2521332, 252214, 252222, 25223, 2525.
Bright, William. 52. Linguistic innovations in Karok. UAL 18.53-62.
Bronx cheer, 153.
Bulgarian. Slavic, Indo-European. Trubetzkoy 29. 2442.
Bunzel, Ruth L. 39. Zuni. HAIL 3.389-515.
Burmese. Tibeto-Burman. Cornyn; McDavid; Ha. 2221, 252112, 2521322, 252214, 252225, 2525.
by-systems: communicative systems deriving very closely from language but not part of
language in the narrow sense. 023.

C
Campa. Arawakan; east central Peru. Dirks. 2441, 252111, 2521331, 252214, 252225, 2525.
Cantonese. Chinese; Canton. Chao; Ha. 2221, 2322, 252214, 252222.
cardinal reference points (acoustic), 5311.
cardinal vowels, two-formant matching of, 5311.
, one-formant matching of, 5314.
Carr, Denzel, 45. Notes on Marshallese consonant phonemes. Lg. 21.267-70.
Cashibo. Panoan; along Aguaitfa River and its tributaries in Peru. Shell 50. 2424, 252113,
2521321, 252214, 252223, 2525.
Catford, J. C. 39. On the classification of stop consonants. Le Maitre Phon^tique 111:65.2-
5.
cavity friction, 12.
central (tongue-position), 1510.
channel noise (information-theoretical term), 03.
Chao, Y. R. 47. Cantonese Primer. Cambridge.
Chatino. Mixtecan?; district of Juquila in Oaxaca, Mexico. McKaughan. 2441, 2442, 2445,
2523, 2525.
Chavarria-Aguilar, O. L. 51. The phonemes of Costa Rican Spanish. Lg. 27.248-53.
Chawchila: dialect of Yokuts, q.v.
Cherokee (of North Carolina). Iroquian. Bender. 252112, 252131, 252225, 2525.
chest pulse, 163.
Cheyenne. Plains Algonquian. Smith 49; Ha. 252111, 2521321, 2525.
Chickasaw: see Choctaw.
Chinese: see Cantonese, Mandarin, Tangsic. 2221.
Chipewyan. Northern Athabascan. Li 46. 2442, 252113, 252135, 252213, 252222, 2524, 2525.
Chiricahua (Apache). Apachean. Hoijer 43, 46. 2221, 2312, 2433, 252112, 25212, 252135, 252213,
252221, 2524, 2525.
 A MANUAL OF PHONOLOGY 229

Chitimacha. Tunican. Swadesh34b, 46a. 2431, 2442, 252112, 2521324, 252213, 252221, 2524, 2525.
Choctaw-Chickasaw. Muskogean. Haas 41. 2431, 252112, 2521321, 252213, 252225.
Chontal (Tequistlateco). Hokan? Waterhouse. 2431, 2442, 2525.
Choynimni: dialect of Yokuts, q.v.
Chukchansi: dialect of Yokuts, q.v.
Churchward, C. Maxwell. 41. A new Fijian grammar. (Sydney?)
clarity norm, in phonologic analysis, 64.
clear vs dark (laterals), 153.
clicks, 1521. [Greenberg (00) says that these, in languages of South Africa, are probably
clusters.]
closure, 1520.
cluster: a sequence of two or more phonemes of the same general type: e.g., of two or more
consonants, or of two or more vowels.
or single phoneme, 3231.
cluster, simultaneous a simultaneous or effectively simultaneous combination of two sets
:

of features, each of which sets occurs separately elsewhere as one phoneme, but in-
capable of being viewed structurally as an ordinary (sequential) cluster,
coarticulated stops, 1521.
coda, 220.
code noise (or "semantic" noise), 03.
Coeur d'Alene. Interior Salishan. Reichard; Swadesh 52. 2442, 252113, 2521332, 25220, 252213,
252223, 2524, 2525.
collation (phase of phonologic analysis), 30, 32. Definition and purpose, 320.
Comanche. Shoshone-Comanche, (Plateau) Shoshonean, Uto-Aztecan. Osborn and Smalley
49, Riggs. 252113, 252131, 2525.
common core (a slant in coverage of statements), 03.
complementary distribution = complementation, 3221.
complementation and contrast, 3221.
complementation, multiple, 3223, 3232.
complete intersection (or complete overlapping: a methodological error), 63.
complex (syllable peaks), 241.
componential analysis, heuristic value of, 262 end.
congruity, pattern, 3223.
consonant. Pretechnical 10; first strict definition 242; wider definition 250.
constituents, ultimate phonologic, 20.
constitute, 321.
constitutional classification (of consonants), 250, 252.
"content" (Hjelmslevian), 02 end.
continuant, nasal, 1521.
contoid, 150, 152, 153.
contrast, role of, 61.
and complementation, 3221.
Cooper, Franklin S., Pierre C. Delattre, Alvin M. Liberman, John M. Borst, and Louis J.
Gerstman. 52. Some experiments on the perception of synthetic speech sounds. JASA
24.597-606.
coordinate (construction), 2212, 321.
Cornyn, William. 44. Outline of Burmese Grammar. Language Dissertation 38; Suppl. to
Lg. 20:4.
covowel. 242.
Cowan, George M. 48. Mazateco whistle speech. Lg. 24.280-6.
Cree. Central Algonquian. Bloomfield 28, 46; Ha. 2431, 2441, 252112, 252131, 252221, 252225,
2525.
Creek-Seminole. Muskogean. Haas 41. 2431, 2441, 252112, 2521321, 252214, 252225.
Croft, Kenneth. 51. Practical orthography for Matlapa Nahuatl. IJAL 17.32-6.
Crow. (Missouri River) Siouan. Kaschube. 252112, 2521321, 252213, 252222, 2525.
230 A MANUAL OF PHONOLOGY

Crowell, Edith E. 49. A preliminary report on Kiowa structure. UAL 15.163-7.

Cuicateco. Mixtecan; northeastern Oaxaca, Mexico. Needham. 252113, 252131, 252213, 2525.
Cuitlateco. Guerrero, Mexico; affiliations unknown. McQuown 41b. 2442, 252113, 2521322,
252213, 252225, 2525.
Czech. Slavic, Indo-European. Goodison 00; H. 2442, 252112, 2521341, 252214, 252224, 25223,
2524, 2525.

D
Dakota. (Mississippi Valley) Siouan. Boas and Deloria 41. 252112, 252213, 252225, 2525.
Danish. North Germanic; Indo-European. Deardon; H. 2443.
Dargwa. North Caucasian. Trubetzkoy 29. 2442.
dark vs. clear (laterals), 153.
Deardon, Jeanette, and K. Stig-Nielson. 45-6. Spoken Danish. New York.
defective phoneme, 3232.
Delattre, Pierre. 48. Un triangle acoustique des voyelles orales du francais. French Review
21, May.
. 51. The physiological interpretation of sound spectrograms. PMLA 66.864-75.
, Alvin M. Liberman, and Franklin S. Cooper. 51. Two-formant synthetic vowels and
cardinal vowels. Le Maitre Phon^tique, July-December.
Alvin M. Liberman, Franklin S. Cooper, and Louis J. Gerstman. 52a. An experimental
,

study of the acoustic determinants of vowel color; observations on one- and two-for-
mant vowels synthesized from spectrographic patterns. Word 8.195-210. (The line draw-
ings marked "Figure 2" and "Figure 7" in this article should be interchanged but not
the captions. Also, the IPA symbols for the low back unrounded and low back rounded
vowels are reversed.]
Franklin S. Cooper, and Alvin M. Liberman. 52b. Some suggestions for teaching
,

methods arising from research on the acoustic analysis and synthesis of speech. Report
of the Third Annual Round Table Meeting on Linguistics and Language Teaching,
The Institute of Languages and Linguistics, Georgetown University, 32-45.
Delaware. Eastern Algonquian. Voegelin 46; HA. 2431, 252112, 25212, 2521321, 252225, 2525.
demivowel, 242.
descriptive, 00, 03 end.
determined vs. determining, 325.
Dibabaon. Davao, Philippines. Forster 54. 2442, 252111, 2521322, 252214, 252225, 2525.
dimensionality of a vowel system, and acoustic correlates, 5315.
Dirks, Sylvester. 53. Campa (Arawak) phonemes. UAL 19.302-4.
distribution, implications of wider, 3233.
, limited, influence on interpretation, 25212.
distributional classification, 250, 251.
domain (of an accent or the like), 2221.
drum signals, African, 023.
duration, 511.
Dutch. West Germanic, Indo-European. Bloomfield 45-6; HA. 2442, 252111, 2521341, 252214,
252225, 2525.
Duwamish. Coast Salishan. Ransom. 252113, 2521332, 252211, 252225, 2524, 2525.
Dyen, I. 45-6. Spoken Malay. New York.
49. On the history of the Trukese vowels. Lg. 25.420-36.
.

E
economy (criterion of), 3224.
Edel, May. 39. The Tillamook language. UAL 10.1-57.
element, phonologic; defined 4.
Elson, Ben. 47. Sierra Popoluca syllable structure. UAL
13.13-7.
54. Sierra Popoluca intonation.
. MA
thesis, Cornell University; unpublished.
Emeneau, Murray B. 39. The vowels of the Badaga language. Lg. 15.43-7.
 .

A MANUAL OF PHONOLOGY 231

. 44. Kota texts, part I. University of California Publications in Linguistics 2:1.1-192.

. 46. The nasal phonemes of Sanskrit. Lg. 22.86-93.
. 51. Studies in Vietnamese (Annamese) grammar. University of California Publica-
tions in Linguistics 8:1.
empathy, 31.
emphatics (pharyngealized consonants), 13.
English. West Germanic, Indo-European. Trager and Smith 51; HA. 2222, 2311, 241, 2431,
2442, 2512, 252112, 2521341, 252214, 252225, 2524.
vowels, acoustics, 5313.
English, Middle (Chaucerian). Standard manuals; A. 2442, 2445.
English, Old. Standard manuals; Stockwell and Barritt; A. 2442.
Erzya. Dialect of Mordvin, Finno-Ugrian; northern and eastern part of Mordva Autonomous
Republic, USSR. Trubetzkoy 29. 2442.
Eskimo, South Greenlandic. Eskimo-Aleut. Swadesh 46b. 2441, 252111, 2521321, 252215,
252224, 2525.
Eskimo, Unaaliq. Eskimo-Aleut. Swadesh 51. 252112.
Esthonian. Finno-Ugrian. Trubetzkoy 29. 2442.
"expression" (Hjelmslevian), 02 end.

Fairbanks, Gordon H. 52. The phonemic structure of Zographensis. Lg. 28.360-5.

feedback, role of, 0212.
field methods, 31. [Their logic, not practical procedures.]
Fijian. Malayo-Polynesian. Churchward; HA. 22114, 2442, 2511, 2513, 252111, 2523, 2525.
Finnish. Finno-Ugrian. Sebeok 47; H. 2222, 2311, 2442, 25212, 252131, 252214, 252225, 2525.
Fischer-J0rgensen, Eli. 52. On the definition of phoneme categories on a distributional basis.
Acta Linguistica 7.8-39.
"form" (vs. "substance"), 02 end.
formants and vowels, 531.
formant line (on spectrogram), 512.
formants, bending of, 534.
Forster, Alice Jannette, 54. Dibabaon texts. University of Manila Journal of East Asiatic
Studies 3 :2.232-6.
,and Howard P. McKaughan. 52. An adaptation of modern language teaching tech-
niques to certain unusual situations. MA
thesis, Cornell University; unpublished.
Fox. Central Algonquian. Bloomfield 24b; HA. 212, 22111, 2431, 2441, 2511, 2513, 252112,
2521321, 252221, 2525, 262.
Freeland, L. S. 47. Western Miwok texts with linguistic sketch. 13.31-46. UAL
51. Language of the Sierra Miwok. Indiana University Publications in Anthropology
.

and Linguistics 6.
French. Romance, Italic, Indo-European. Hall 46, 48b; Trager 44; Martinet 49b; Jakobson
and Lotz 49; HA. 212, 2442, 2443, 2445, 2446, 252111, 2521341, 252214, 252224, 252226, 2525.
vowels, acoustics, 5312.
frequency (in acoustics), 511.
frequency (statistical) norm, in phonologic analysis, 64.
fricative vs. frictionless, 153.
front (tongue position), 1510.
Fry, Allan H. 41. A phonemic interpretation of visarga. Lg. 17.194-200.
functional load, measuring, 62. Definition 61, 62.
fundamental (in sound), 511.

G
gap (in physical arrangement of ICs) 321 ,

Garvin, Paul L. 48. Kutenai I: phonemes. UAL 14.37-42.

232 A MANUAL OF PHONOLOGY

Garvin, Paul L. 50. Wichita I: phonemes. IJAL 16.179-84.

Gashowu: dialect of Yokuts, q.v.
gathering (phase in phonologic analysis), 30, 31.
geminate (cluster) a cluster of two occurrences of the same phoneme: /pp/, /aa/. 2221.
:

Georgian. South Caucasian. Vogt 36; HA. 2213, 2442, 25210, 252112, 2521333, 252135, 252213,
252222, 2525.
German. West Germanic; Indo-European. Moulton 47; Mueller; HA. 212, 2311, 2431, 2442,
252111, 2521332, 252214, 252224, 2524, 2525.
dialect of Brienze, Switzerland. Susman 51. 2521321, 252213, 252225, 2525.
Germanic: English Dutch, German, Scandinavian. 2222.
Germanic, Proto-. Standard manuals; Twaddell 48; A. 2441.
glide vs. peak vocoid, 165.
glottal catch (not a "stop" by our definition of the latter term), 12.
glottal spirant, voiced and voiceless, 12.
glottal stop = glottal catch; former term not used. 12.
glottal tone, 12.
glottalized stop, 1521.
glottis, 12.
Goodison, Ronald A. C. 51. The phonology of Czech. Ph. D. thesis, Cornell University;
unpublished.
Gourma. West Africa. Welmers 00. 252114.
grammar, defined 02.
grammatical boundaries, 324 end.
Gray, Henry. 48. Anatomy of the human body. 25th edition, edited by Charles M. Goss.
Philadelphia.
Greek (modern). Indo-European. Kahane; H. 212, 2442.
Greek, classical. Trubetzkoy 20; Sturtevant; A. 2442.
Greek, late classical and early medieval. Trubetzkoy 29. 2442.
Greenberg, Joseph H. 41. Some problems in Hausa phonology. Lg. 17.316-23.

H
[h]not necessarily a spirant, 12.
[h],"voiced," 12.
Haas, Mary R. 41a. The classification of the Muskogean languages. LCP.
41b. Tunica. HAIL IV (extract only; New York).
.

42. Types of reduplication in Thai. SIL 1:4.

.

44. Men's and women's speech in Koasati. Lg. 20.142-9.

.

46. A grammatical sketch of Tunica. LSNA 337-66.

.

and H. R. Subhanka. 45-6. Spoken Thai. New York.

,

HAIL = Handbook of American Indian Languages. HAIL I and II, Bureau of American
Ethnology Bulletin 40, parts 1 and 2, Washington 1911, 1922. HAIL III: New York
1939. HAIL IV: only one "extract" printed, New York 1941.
Hall, Robert A. Jr., 43. The unit phonemes of Brazilian Portuguese. SIL 1:16. Occurrence
and orthographical representation of phonemes in Brazilian Portuguese. SIL 2.6-13.
.44a. Hungarian grammar. Language Monograph 21 supplement to Lg. 20:4.
;

.44b. Italian phonemes and orthography. Italica 21.72-82.

.46a. Colloquial French phonology. SIL 4.70-90.
.46b. Old French phonemes and orthography. Studies in Philology 43.575-85.
.48a. Descriptive Italian grammar. Ithaca.
.48b. French. Language Monograph 24; supplement to Lg. 24:3.
. The linguistic structure of Taki-Taki. Lg. 24.92-116.
48c.
.The reconstruction of Proto-Romance. Lg. 26.6-27.
50.
Halpern, A. M. 46a. Yuma. LSNA 249-88.
46b. Yuma I: Phonemics. IJAL 12.25-33.
.
 . .

A MANUAL OF PHONOLOGY 233

hamzah = glottal catch, 12.

Hanes, Leonard C. 52. Phonemes of Motilone (Carib). IJAL 18.146-9.
Harden, Margaret. 46. Syllable structure in Terena. IJAL 12.60-3.
harmonic, 511.
streak (on spectrogram), 512.
The linguistic structure of Hebrew. JAOS 61.143-67.
Harris, Zellig S. 41.
. Navaho phonology and Hoijer's analysis. IJAL 11.239-48.
45.
.51. Methods in structural linguistics. Chicago.
Hartman, Lawton M. 3d. 44. The segmental phonemes of the Peiping dialect. Lg. 20.28-42.
Haugen, Einar. 45-6. Spoken Norwegian. New York.
.54. Rev. of Weinreich 53. Lg. 30.380-8.
Hausa (Kano dialect). "Sudanese" (Afro-Asiatic); Nigeria, British West Africa. Green-
berg; Hodge 44. 2442.
Hawaiian. Polynesian; Malayo-Polynesian. Wise and Hervey. 252111, 252131, 252213,
252223, 2525.
Heffner, R-M. S. 49. General phonetics. Madison,
height (of tongue), 1510.
Henry, Jules. 48. The Kaingang language. IJAL 14.194-204.
hesitation -forms, 210.
Hidatsa. (Missouri River) Siouan. Lowie. 252112, 2521321, 25224, 2525.
high (tongue position), 1510.
Hindustani. Indie, Indo-European. Hoenigswald; H. 25212.
Hitchiti-Mikasuki. Muskogean, Haas 41. 2431, 252112, 2521321, 252213, 252225.
Hjelmslev, L. 36. On the principles of phonematics. Proceedings of the second international
congress of phonetic sciences 49-54.
.36-7. Accent, intonation, quantity. Studi Baltici 6.27ff
. 43. Omkring Sprogteoriens Grundlaeggelse. Copenhagen. English translation by
Francis J. Whitfield, Prolegomena to a Theory of Language; Indiana University
Publications in Anthropology and Linguistics 7, 1953.
Hjelmslevian terms, correlation with those used here: 02 end.
Hockett, Charles F. 42. A system of descriptive phonology. Lg. 18.3-21.
.47a. Peiping phonology. JAOS 67.253-67.
.47b. Componential analysis of Sierra Popoluca. IJAL 13.258-67,
.48a. Biophysics, linguistics, and the unity of science. American Scientist 36.558-72.
.48b. Potawatomi I: Phonemics. IJAL 14.1-10.
.49. Two fundamental problems in phonemics. SIL 7.33ff
.51. Review of Martinet 1949a,b. Lg. 27.333-41.
. 52a. An approach to the quantification of semantic noise. Philosophy of Science
19.257-60.
. Review of TCLC V. IJAL 18.86-98.
52b.
. Review of Harris 51. AmSp 28.117-21.
52c.
.52d. [Writing and Language.] Report of the Third Annual Round Table Conference
on Linguistics and Language Teaching, Georgetown University Institute of Languages
and Linguistics. Washington.
.53a. Review of Shannon 49. Lg. 29.69-92.
. 53b. Short and long syllable nuclei. IJAL 19.165-71.
Hodge, Carleton T. 46. Serbo-Croatian phonemes. Lg. 22.112-20.
,and H. E. Hause. 44. Hausa tone. JAOS 64.51-2.
Hoenigswald, Henry. 45-6. Spoken Hindustani. New York.
Hoijer, Harry. 43. Pitch accent in the Apachean languages. Lg. 19.38-41.
. 45. Navaho phonology. University of New Mexico Publications in Anthropology 1.
Albuquerque.
.46a. Chiricahua Apache. LSNA 55-84.
. 46b. Tonkawa. LSNA 289-311.
234 A MANUAL OF PHONOLOGY

Hooier, Harry, and Edward P. Dozier. 49. The phonemes of Tewa, Santa Clara dialect. IJAL
15.139-44.
Hopi. Shoshonean, Uto-Aztecan. Whorf 46a. 2431, 252113, 2521321, 252213, 252225, 2525.
Householder, F. 52. Review of Harris 51. IJAL 18.260-8.
Huasteco. Mayan; coast of Vera Cruz, and in Potosi, Mexico. Larsen. 2442, 252113, 2521323,
252213, 252223, 2524, 2525.
Hiiichol. Uto-Aztecan; states of Nayarit and Jalisco, Mexico. Mcintosh. 2441, 252113,
252131, 252213, 252225, 2525.
Hungarian. Finno-Ugrian. Hall 44a; Sebeok 43; H. 2311, 2431, 2442, 25210, 252112, 2521322,
2521341, 252214, 252225, 2525.

IC = immediate constituent.
identification, criteria for, 322.
idiolect, 03.
IJAL = International Journal of American Linguistics.
Ilocano. Northern Luzon, Philippines. Bloomfield 42; Forster and McKaughan; H. 2442,
25111, 2521322, 252214, 252225, 2525.
and status, 04.
imitation-label; definition
immediate constituent analysis, 241. Techniques of, 321.
injectives, 1521.
inspirated sounds (made during inhalation), 11.
intensity (acoustics), 511.
interlude, 220.
intersection, acoustic, 54.
, complete (an analytical mistake, not a method), 63.
, sporadic, 63.
intonation, 210, 260.
, of English, 211.
, of languages other than English, 212.
Iowa-Oto. (Mississippi Valley) Siouan. Whitman. 2441, 2442, 2445, 252112, 2525.
IPA (= International Phonetic Association/Alphabet), misuse of, 522 (3).
Iraqi (Arabic). Arabic, Semitic; Iraq. Van Wagoner; H. 2441, 2523, 2524, 2525.
Iroquoian: Seneca, Oneida, Cherokee. (Mohawk: Ha.) 252210, 252212.
Isleta.Tiwa, Tanoan. Trager 42. 252113, 2521342, 252225.
a criterion, 230.
isolability, as
Isthmus Zapotec: see Zapotec.
Italian. Romance, Italic, Indo-European. Hall 44b, 48a; Ha. 2311, 2443, 252112, 2521322,
252214, 252222, 25223, 2525.
Ixcatepec: see Nahuatl.

Jakobson, Roman. 39. Observations sur la classement phonologique des consonnes. Pro-
ceedings of the Third International Congress of Phonetic Sciences 34-41.
. 49. On the identification of phonemic entities. TCLC 5.205-13.
, and Notes on the French phonemic pattern. Word 5.151-8.
J. Lotz. 49.
C. Gunnar M. Fant, and Morris Halle. 52. Preliminaries to speech analysis. The
,

distinctive features and their correlates. Cambridge.

JAOS = Journal of the American Oriental Society.
Japanese. Bloch 50; Ha. 2214, 2442, 252112, 2521322, 252214, 252225, 2525.
JASA = Journal of the Acoustical Society of America.
jaws, 1510.
Jensen, Ella. 49. The vowel system of the Flensborg by-laws. TCLC 5.244-55.
Jones, Daniel. 50. The phoneme: its nature and use. London.
Joos, Martin. 48. Acoustic phonetics. Language Monograph 23; supplement to Lg. 24:2.
 A MANUAL OF PHONOLOGY 235

Jukun. West Africa. Welmers 00. 252114.

juncture. 220, 2220. Techniques for the analysis of, 324.

K
Kahane, Henry, Rene Kahane, and Ralph L. Ward. 45-6. Spoken Greek. New York.
Kaingang. Ge; eastern mountain ranges of Brazil, from Sao Paulo to northern border of
Uruguay. Henry. 252112.
Kalis-pel. Interior Salishan. Vogt 40a; Swadesh 52. 252113, 2521332, 25220, 252213, 252225,
2524, 2525.
Kapauku. Interior of western New Guinea. Mary Doble 00. 252131.
Karagin. Dialect of Koryak, "Palaeoasiatic" northeastern Siberia. Trubetzkoy 29. 2442.
;

Karok. Hokan? Bright. 2431, 252112, 2521321, 252213, 252225, 2525.

Kaschube, Dorothea. 54. Examples of tone in Crow. IJAL 20.34-6.
Kechua. Quechuan. (Cuzco:) Yokoyama; (another dialect) Ha. 2441, 2442, 252112, 2521332,
252215, 252225, 25223, 2524, 2525. Dialect of Callejon de Huaylas, 252112.
Kennedy, George A. 52. Voiced gutturals in Tangsic. Lg. 28.457-64.
. 53. Two-tone patterns in Tangsic. Lg. 29.367-73.
Keres. Spencer. 2442, 252111, 2521331, 252214, 252225, 2525.
Kersta, L. G. 52. Solid Sound. Bell Laboratories Record, September.
Key, Harold, and Mary Key. 53. The phonemes of Sierra Nahuat. IJAL 19.53-6.
King, Harold V. 52. Outline of Mexican Spanish phonology. SIL 10.51-62.
Kiowa. Azteco-Tanoan(?). Crowell. 2443, 252112, 2521322, 252213, 252222, 2525.
Koasati. Muskogean. Haas 41a, 44. 2431, 252112, 2521321, 252213, 252225.
Koibal. Northeastern Turkic, "Altaic"; Siberia. Trubetzkoy 29. 2442.
Korean. Martin; Lukoff HA. 212, 2442, 252112, 25212, 25213, 252214, 252225, 2525.
;

Kota. Dra vidian; Nilgiri Hills, southern India. Emeneau 44. 2213, 2431, 2442, 252111, 2521322,
252215, 252224, 25223, 2525.
Kpelle. West Africa. Welmers 47, 00. 252114.
Kraho. Ge, location unknown. Shell 52. 252112, 202214, 252221, 2525.
Kuhn, S. M., and R. Quirk. 53. Some recent interpretations of Old English digraph spellings.
Lg. 29.143-56.
Kurylowicz, J. 49. La notion de l'isomorphisme. TCLC 5.48-60.
Kutenai. Garvin 48. 2431, 252112, 2521321, 252213, 252225, 2525.
Kwakiutl. Wakashan. Boas lib, 47; Swadesh 53. 252113, 2521342, 25220, 252213, 252225, 2524,
2525.
Kyuri. North Caucasian. Trubetzkoy 29. 2441.

Lak. North Caucasian. Trubetzkoy 29. 2441.

langue and parole, 4.
Larsen, Raymond S., and E. V. Pike. 49. Huasteco intonations and phonemes. Lg. 25.268-77.
laryngeal; here a convenient cover term for consonants whicJ are neither obstruents nor
sonorants under our definitions: 2520; 2525.
larynx, 12.
laterals, spirantal, 1522.
, vocoid, 153.
Latin. Italic, Indo-European. Sturtevant; A. 2442, 25210, 252111, 2521322, 2521332, 252213.
2525.
, late. Hall 50, Politzer. 2443.
Lauriault, James. 48. Alternate-mora timing in Shipibo. IJAL 14.22-4.
lax vs. tense, 1510.
LCP = Language, Culture, and Personality; Essays in Memory of Edward Sapir. Menasha,
1941.
Lehmann, Winfred P. 52. Proto-Indo-European phonology. Austin, Texas.
236 A MANUAL OF PHONOLOGY

length, in articulation, 161.

lengtheners (covowels), 2431.
, scalar, 2431.
Lenormand, Maurice H. 52. The phonemes of Lifu (Loyalty Islands); the shaping of a
pattern. Word 8.252-7.
Lesnin, I. M., Leonard Bloomfield, and Olga Petrova. 45-6. Spoken Russian. New York.
Lg. = Language, Journal of the Linguistic Society of America.
Li, Fang-kuei. 30. Mattole, an Athabaskan language. University of Chicago Publications
in Anthropology.
. 46. Chipewyan. LSNA 398-423.
. 48. The distribution of initials and tones in the Sui language. Lg. 24.160-7.
Liberman, Alvin M., Pierre Delattre, and Franklin S. Cooper. 52. The role of selected
stimulus-variables in the perception of the unvoiced stop consonants. The American
Journal of Psychology 65.497-516.
Lifu. Loyalty Islands. Lenormand. 2442, 252112, 2521321, 252215, 252223, 2524, 2525.
Liljebled, Sven. 50. Bannack [sic] I: phonemes. UAL 16.126-31.
linear (accentual system), 230, 231.
linguistics: alternative definitions, 02.
lips, in vocoid articulation, 1510.
Lisker, Leigh. 48. The distinction between [] and [e]: a problem in acoustic analysis. Lg.
24.397-407.
Loma. West Africa, Sadler. 2312, 2443, 252114, 2521332, 252214, 252225, 2525.
Longacre, Robert E. 52. Five phonemic pitch levels in Trique. Acta Linguistica 7.62-82.
loudness, 511.
Lounsbury, Floyd G. 53. Oneida Verb morphology. Yale University Publications in An-
thropology 48. New Haven,
low (tongue position), 1510.
lowering lengthener (in complex syllable peaks), 2431.
Lowie, Robert H. 39. Hidatsa texts, with grammatical notes and phonograph transcriptions
by Zellig Harris and C. F. Voegelin. Indiana Historical Society Prehistory Research
Series 1:6.
LSNA = Harry Hoijer and others, Linguistic structures of native America. Viking Fund
Publications in Anthropology 6, 1946.
Lukoff, Fred. 45-6. Spoken Korean. New York,
lungs, 11.
Lunt, Horace G. 53. Old Church Slavonic phonemes. Word 8.311-28.

M
macrosegment, 210. Redefinition (deleting intonation), 220.
Maidu. Penutian. Uldall. 2442, 252111, 2525.
Mak, Cornelia. 53. A comparison of two Mixtec tonemic systems. IJAL 19.85-100.
Malay. Malayo-Polynesian. Dyen 45-6. 2442, 2445.
Mandarin. Chinese. Hartmann; Hockett 47; HA. 212, 2222, 2321, 2442, 2513, 252112, 25212,
2521332, 252214, 252222, 2525.
Maninka. In and around Kankon, French Guinea. Welmers 49. 5231, 5232, 524.
manner (for consonants, esp. obstruents), 2521.
manner consonants, 2520, 2524.
Mano. West Africa. Welmers 00. 252114.
margin (of syllable; = onset, coda, or interlude), 242.
marked vs. unmarked, 3233.
markers, in IC analysis, 321.
Marshallese. Marshall Islands. Carr. 252111, 252131, 2523, 2524, 2525.
Martin, Samuel E. 51. Korean phonemics. Lg. 27.519-33.
. 53. The phonemes of Ancient Chinese. JAOS supplement 16.
 A MANUAL OP PHONOLOGY 237

Martinet, Andre\ 39. Equilibre et stability des systemes phonologiques. Transactions of the
Third International Congress of Phonetic Sciences 30-4.
.49a. Phonology as Functional Phonetics. [First section in a pamphlet bearing the
same title; Publications of the Philological Society, London.]
.49b. Les traits gen6raux de la phonologie du francais. [Second section in the same
pamphlet.]
.49c. La double articulation linguistique. TCLC 5.30-7.
.52. Function, structure, and sound change. Word 8.1-32.
.53. Concerning the preservation of useful sound features. Word 9.1-11.
Matlapa: dialect of Nahuatl, q.v.
Matthews, W. K. 51. Languages of the USSR. Cambridge.
Mazahua: dialect of Otomi, q.v.
Mazateco. Otomanguian; Oaxaca, Mexico. Pike and Pike 47. 2212, 2312, (2322), 2433, 2511,
2513, 252112, 2521321, 252214, 2525.
McDavid, Raven I., Jr. 44. Burmese phonemics. SIL 3.6-18.
Mcintosh, John B. 45. Huichol phonemes. UAL 11.31-5.
McKaughan, Howard. 54. Chatino formulas and phonemes. UAL 20.23-7.
McQuown, Norman A. 41a. La fonemica de un dialecto Nahuatl de Guerrero. El Mexico
Antiguo 5.221-32.
.41b. La fonemica del Cuitlateco. El Mexico Antiguo 5.239-54.
.42. La fonemica de un dialecto Olmeca-Mexicano de la Sierra Norte de Puebla. El
Mexico Antiguo 6.61-72.
.44-5. Spoken Turkish. New York.
. 51. Review of Heffner 49. Lg. 27.344-62.
Mende. West Africa. Welmers 00. 252114.
Menomini. Central Algonquian. Bloomfield 00, 24a, 39; Ha. 2431, 2442, 252112, 252131,
252221, 2525.
Mesquital: dialect of Otomi, q.v.
Mexicano: see Nahuatl.
microjuncture (and syllable juncture), 2222.
microsegment, 2222.
mid (tongue position), 1510.
Mikasuki: see Hitchiti.
Mil-pa Alta: dialect of Nahuatl, q.v.
minimal pair, 61.
Miwok. Penutiar Freeland 47, 51. 2431, 2442, 252225. Western: 252112, 2521333, 252213,
2524, 2525. Sierra: 252112, 2521321, 252214, 2525.
Mixteco. Mixtecan; Guerrero and Oaxaca, Mexico. Mak. 2312, 2442, 252113, 2523, 2525.
modifications of vocoids (i.e., effects achieved other than by tongue height, tongue frontness
or backness, and lip position). 1511.
Mongol. Altaic? Aberle and Austin. 244 note, 2444, 252112, 2521332.
monitoring (via feedback), 0212.
mora; unit of duration smaller than a syllable, when there is reason to distinguish the two
structurally, 2221.
morph: the phonologic shape which represents a morpheme. Definition 02.
morpheme. Definition, 02. [A unit of content, not of expression.]
morphophonemics. Definition, 02.
Motilone. Cariban; Eroca valley, Andes Mts., Department of Magdalena, Colombia. Hanes.
252112, 2521321, 252213, 252221, 2525.
Moulton, William G. 47. Juncture in modern standard German. Lg. 23.212-26.
. 48. The phonemes of Gothic. Lg. 24.76-86.
.54. The stops and
spirants of early Germanic. Lg. 30.1-42.
Mueller, Hugo. 50. Stress phonemes in German. SIL 8.82-9.
multiple complementation, 3223, 3232.
238 A MANUAL OF PHONOLOGY

murmur, 12.
murmur vowel: a slight voiced release for a consonant, either of indefinite color or of color
which seems to be determined entirely by the surrounding consonantism.
musical distance between frequencies, 511.
musical note, 511.
Muskogean: Choctaw-Chickasaw, Alabama, Koasati, Hitchiti-Mikasuki, Creek-Seminole.
Haas 41. 2441, 2525.

N
Nahuatl (Aztec, Mexicano). Uto-Aztecan. 2441, 252113, 2521321, 252213, 252225, 2525.
, Ixcatepec dialect. McQuown 41a.
, Matlapa dialect. Croft. 2431.
, Milpa Alta dialect. Whorf 46b.
, Olmeca-Mexicano dialect. McQuown 42.
, Sierra Nahuat dialect. Key.
, Tetelcingo dialect. Pittman 54. 252223.
nasal, nasality, nasal cavity, 14.
nasal continuant or sonorant, 1521, 25221.
nasalization, levels of appearance of, 2611. Acoustic correlates of, 5316.
Navaho. Apachean, Athabascan. Hoijer 45; Harris 45. 252113, 25212, 252135, 25220, 252213,
252221, 2524, 2525.
Navarro -Tomas, Tomas. 32. Manual de pronunciaci6n espanola 4 Madrid. .

Needham, Doris, and M. Davis. 46. Cuicateco phonology. IJAL 12.139-46.

Neffgen, H. 18. Grammar and vocabulary of the Samoan language. London. (Tr., from
earlier German edition, by Arnold B. Stock.)
Nellis, Jane G. 47. Sierra Zapotec forms of address. IJAL 13.231-2.
neutralization, 3232.
Newman, Stanley L. 44. The Yokuts language of California. Viking Fund Publications in
Anthropology 4. New York.
. 46. dialect of Yokuts. LSNA 222-48.
The Yawelmani
. Coola I: phonology. IJAL 13.129-34.
47. Bella
noise (everyday and acoustic sense), 511. Noise-bursts, in acoustics, 532.
noise (information-theoretical sense) : 02, 6.
non-linear (accentual systems), 230, 232; 242.
Nootka. Wakashan. Sapir and Swadesh 39. 2213, 2321, 2431, 2442, 252113, 2521332, 25220,
252213, 2524, 2525, 262.
"north" and "south" (towards the mouth and nose, versus towards the lungs, in articula-
tory description), 10.
Norwegian. North Germanic, Indo-European. Haugen; Ha. 2321, 2443.
Lyster dialect. Lyster, Sogn, northwestern Norway. Oftedal. 2443.
,

nucleus, in IC analysis, 321. Nucleus and satellite 2212.

O
obstruent, definition 2520. Obstruent systems: 25210, 2521.
O'Connor, J. D., and J. L. M. Trim. 53. Vowel, consonant, and syllable a phonological
definition. Word 9.103-22.
octave, 511.
Oftedal, Magne. 49. A Norwegian dialect in Wisconsin. Lg. 25.261-7.
Ojibwa. Central Algonquian. Bloomfield 00; HA. (2221), 2441, 252112, 25212, 2521321, 252221,
2525.
Olmeca: dialect of Nahuatl, q.v.
Olmsted, David L. 50. The phonology of Polish. Ph.D. thesis, Cornell University; un-
published.
 A MANUAL OF PHONOLOGY 239

. 51. The phonemes of Yoruba. Word 7.245-7.

omnipotent, definition 242.
Oneida. Iroquoian. Lounsbury. 2431, 2442, 252112, 252131, 252225, 2525.
onset (part of syllable), 220.
opposition: in strict Praguian theory "opposition" and "contrast" are distinguished; e.g.,
see Martinet 53, fn. 21, and compare Prieto 54. The distinction is of importance, but
does not necessarily have to be made with just these two terms. Our use of "contrast"
is general, and the word "opposition" is used only rarely, usually in the context of

stating a Praguian position,

oral (vs. nasal), 14.
organs of speech (a metonymic expression), 10, 1.
Osage. Siouan. Wolff 52. 2441, 2442, 2445, 252112, 2525.
Osborn, Henry. 48. Amahuaca phonemes. IJAL 14.188-90.
, and William H. Smalley. 49. Formulae for Comanche stem and word formation.
IJAL 15.93-9.
oscillograph, oscillogram. 512.
Osgood, Charles E., Editor. 54. Psycholinguistics. Indiana University Publications in
Anthropology and Linguistics, Memoir 10; supplement to IJAL 20:4.
Ossetic. Iranian, Indo-European; Caucasia, USSR. Vogt 44. 2520, 252112, 252135, 252213,
252225, 2525.
Oto: see Iowa.
Otomi. Otomanguian; central Mexico. 252214, 252225, 2525.
, Mazahua dialect. E. V. Pike 51; Spotts. 212, 2442, 2445, 252113, 2521322.
, Mesquital dialect. Sinclair. 252112, 2521332.

, Temoayan dialect. Andrews. 2441, 2442, 252133, 2521322.

overall pattern (one slant in coverage of statements), 03.

parole and langue, 4.

parsimony, 3224.
Pashto. Iranian, Indo-European; Afghanistan. Penzl. 252112.
"pattern" (vs. "usage"), 02 end.
pattern and structure, 4.
pattern congruity, as criterion, 3223.
pattern playback (a device for reading spectrograms into sound), 512.
pause, 210.
peak (part of syllable), 220.
peaks, simple and complex, 241. Detailed analysis, 24.
peak vs. glide vocoid, 165.
Penzl, Herbert. 43. Aspect in the morphology of the Pashto (Afghan) verb. SIL 1 :16.

perceptual distance between frequencies, 511.

Persian (modern colloquial). Iranian, Indo-European. Newman 00; H. 2442.
Peterson, Gordon E., The phonetic value of vowels. Lg. 27.541-53.
pharyngeal catch, 13.
pharyngeal spirant, 13.
pharyngealized sounds (= "emphatics"), 13.
pharynx, 13.
phase-relationship, 511; definition by example 512.
phoneme: rough definition 242. [No precise definition possible under the relativistic ap-
proach taken in this monograph.]
, relativity of, 3224.
or cluster, 3231.
phonetic realism, phonetic similarity, as criterion, 3222.
phonetic system of a language; definition, 02.
240 A MANUAL OF PHONOLOGY

phonetics, acoustic, 5.
phonetics, articulatory, 1.

phonology; definition, 02.

fundamental assumption of, 31.
,

Pickett, Velma. 51. Non -phonemic stress: a problem in stress placement in Isthmus Zapotec.
Word 7.61-5.
. Isthmus Zapotec verb analysis. IJAL 19.292-6.
53.
Picuris. Tiwa, Tanoan. Trager 42. 252113, 2521332, 252225.
Pike, Eunice V. 48. Problems in Zapotec tone analysis. IJAL 14.161-70.
51. Tonemic-Intonemic Correlation in Mazahua (Otomi). IJAL 17.37-41.
.

Pike, Kenneth L. 43. Phonetics. Ann Arbor.

. 46. The intonation of American English. Ann Arbor.
. 47a. Grammatical prerequisites to phonemic analysis. Word 3.155-72.
. 47b. Phonemics. Ann Arbor.
52. More on grammatical prerequisites. Word 8.106-21.
.

and E. V. Pike. 47. Immediate constituents of Mazateco

, syllables. IJAL 13.78-91.
pitch, mechanism of production, 12; acoustics 511.
levels (intonational), 211.
Pittman, Richard S. 48. Nuclear structures in linguistics. Lg. 24.287-92.
. 54. A grammar of Tetelcingo (Morelos) Nahuatl. Lg. Diss. 50, Suppl. to Lg. 30:1.
pivot (in IC structure), 321.
PL = pitch level (unit in an intonational system).
point of articulation, 1520.
Polish. Slavic, Indo-European. Trager 39; Olmsted 50; (local dialects) Trubetzkoy 29; H.
2442, 2443, 2512, 252112, 2521341, 252213, 2525.
Politzer, Robert L., 51. On the phonemic interpretation of Late Latin orthography. Lg.
27.151-3.
Polynesian: Samoan, Hawaiian, and many others. 2442, 252131.
Popoluca, Sierra. Popolucan; Vera Cruz, Mexico. Elson 47, 54. 212, 2431, 2442, 25210, 252112,
2521322, 2521342, 252224, 2525.
Portuguese. Romance, Italic, Indo-European. Hall 53; Agard 00; Ha. 2443, 2445, 2512, 252111,
2521341, 252214, 252222, 25223, 2525.
position, in classifying obstruents, 2521.
position of articulation, 1520.
Potawatomi. Central Algonquian. Hockett 48b; HA. 2442, 252112, 2 il2, 2521321, 252221,
2525.
Potter, R. K., G. A. Kopp, and H. C. Green. 47. Visible Speech. New York,
pressure from lungs, importance of variations in, 150.
Preston, W. D., and C. F. Voegelin. 49. Seneca I. IJAL 15.23-44.
Prieto, Luis J. 54. Traits oppositionnels et traits contrastifs. Word 10.43-59.
primary secondary feature, 325.
vs.
psycho-acoustic experiments (in acoustic phonetics), 521 (2).
pure tone, 511.

Q
Quechua: see Kechua.
Quileute. Chimakuan. Andrade. 252211.

R
lr], 153.
r-color, 5316.
raising lengthener (in complex syllable peaks), 2431.
Ransom, Jay E. 45. Notes on Duwamish phonology and morphology. IJAL 11.204-10.
redundancy, phonologic, 61.
 A MANUAL OF PHONOLOGY 241

Reichard, Gladys A. 38. Coeur d'Alene. HAIL 3.

release, of consonants, 162.
remainder (macrosegment, in original sense, minus intonation): 210. [After 21, "macro-
segment" is redefined to exclude intonation and the term "remainder" is not used.]
resonance chamber, 14, 1510.
retroflexion, 1511.
rhythm, 166.
Riggs, Venda. 49. Alternate phonemic analyses of Comanche. UAL 15.229-31.
rill (spirant), 1522.
Romance. 212. (Otherwise see separate languages.)
Romanian. Romance, Italic, Indo-European. Agard 53; H. 2442, 252112, 2521341, 252213,
252225, 2525.
rounding, 1510.
Russian. Slavic, Indo-European. Lesnin; Trager 34; Fairbanks 00; Ha. 2222, 241, 2441,
2442, 2445, 2446, 2512, 252112, 2521341, 252213, 252224, 2523, 2524.
Rutul. North Caucasian. Trubetzkoy 29. 2441.

S
Sadler, Wesley. 51. Untangled Loma. United Lutheran Board[no place given].
Saho. West Africa. Welmers 52b. 2442, 2521342, 252213, 252225, 2525.
Salishan: Bella Coola, Coeur d'Alene, Tillamook, Kalispel, Duwamish. Swadesh 52; Vogt
50b. 2213, 2441, 25220, 252211.
Samoan. Polynesian; Malayo-Polynesian. Neffgen. 252111, 2521321, 252210, 252214, 252222,
2525.
Sandia. Tiwa, Tanoan. Trager 42 252113, 2521342, 252225.
Sanskrit. Indie, Indo-European. Whitney; Emeneau 46; Fry; A. 252112, 25212, 2521322,
2525.
Santa Clara. Tewa, Tanoan. Hoijer and Dozier 49; H. 252113, 2521342, 252214, 252225, 2524,
2525.
Sapir, Edward. 22. Takelma. HAIL II.
. A Chinookan phonetic law. IJAL 4.105-10.
2?.
and Morris Swadesh. 39. Nootka texts. Philadelphia.
,

of satellite and nucleus

satellite (in IC structure), 321. Satellite and nucleus 2212. [We speak
only in an attributive or subordinate construction; in a coordinate construction the
constituents cannot be so labelled.]
scalar lengthener (in complex syllable peaks), 2431.
Scandinavian: Norwegian, Swedish, Danish. 2321.
Schatz, Carol D. 54. The role of context in the perception of stops. Lg. 30.47-56.
scrape [r], 153.
Sebeok, Thomas A. 43. The phoneme /h/ in Hungarian. SIL 1:13.
47. Spoken Finnish. New York.
.

secondary versus primary feature, 325.

segmental phoneme = our term "phoneme" in this manual, 242.
semantics: definition, 02.
semiconsonant. General definition, 242; not necessarily used in that precise sense in earlier
sections.
Seminole: see Creek,
semivowel. General definition 242; not necessarily used in that precise sense in earlier
sections.
Senadi. Senufo group; Korhogo, Ivory Coast. Welmers 50a. 22111, 2312, 2442, 2443, 2445,
252114, 2521341, 252212, 252215, 2525.
Seneca. Iroquoian. Preston and Voegelin. 2431, 252112, 2521321, 252221, 2525.
Serbo-Croatian. Slavic, Indo-European. Hodge 46; Jakobson 49. 2442, 252112, 2521341,
252214, 252224, 25223, 2525.
242 A MANUAL OF PHONOLOGY

shannon: a unit of informational entropy or capacity, equal to one binit per second.
Shannon, Claude. 49. Information theory of secrecy systems. Bell System Technical Journal
28.656-715.
and Warren Weaver. 49. The mathematical theory of communication. Urbana.
,

shared constituent, 321.

Shawnee. Central Algonquian. Voegelin 35; Ha. 212, 2431, 2441, 252112, 2521321, 252225,
2525.
Shell, Olive A. 50. phonemes. UAL 16.198-202.
Cashibo I:
Grammatical outline of Kraho (Ge family). IJAL
(after Buell Quain). 52. 18.115-29.
shimmering (kind of morphophonemic alternation), 63.
Shipibo. Panoan; around Ucuyali river in Peru. Lauriault. 252112, 2525.
Sierra Miwok: see Miwok.
Sierra Nahuat: see Nahuatl.
Sierra Popoluca: see Popoluca.
Sierra Zapotec: see Zapotec.
SIL = Studies in Linguistics.
simple (peak), 241.
Sinclair, Donald E., and K. L. Pike. 48. The tonemes of Mesquital Otomi. IJAL 14.91-8.
singing, 023.
slant lines, use of, 20.
Slavic: Russian, Polish, Czech, Bulgarian, Serbo-Croatian, Ukrainian, Zographensis.
212, 2222.
slit (spirant), 1522.
Slocum, Marianna C. 48. Tzeltal (Mayan) noun and verb morphology. IJAL 14.77-86.
slurring, in articulation, 54.
Smith, Henry Lee Jr. 52a. An outline of metalinguistic analysis. Report of the Third Annual
Round Table Conference on Linguistics and Language Teaching, Georgetown Uni-
versity Institute of Languages and Linguistics. Washington.
52b. An outline of metalinguistic analysis. Washington. [Expansion of 52a.]
.

Smith, William B. S. 46. The Breton segmental phonemes. SIL 4.52-69.

. 49. Some Cheyenne forms. SIL 7.77-85.
Snoqualmie. Swadesh 52. 252113, 2521332, 252211, 252225, 2524, 2525.
Sobotta, Johannes. 33. Atlas of human anatomy. Edited from the eighth German edition
by J. P. McMurrich; Third Revised English Edition. New York.
Sommerfelt, A. 31. Sur 1'importance g6ne>ale de la syllabe. TCLP 4.156-9.
36. Can syllable divisions have phonological importance? Proceedings of the second
.

international congress of phonetic sciences 30-3.

39. Le systeme phonologique d'une langue Australienne. TCLP 8.209-12.
.

sonorant; definition, 2520. 2522.

"south" and "north" (in articulatory description), 10.
Spanish. Romance, Italic, Indo-European. Navarro-Tomas 32, Trager 39, King 52, Chavar-
ria-Aguilar 51; HA. 22113, 2311, 2432, 252112, 2521332, 252214, 252222, 252223, 25223,
2525.
spectrograph, spectrogram, 512.
speech signal; definition, 510.
speech sound: term a figure of speech, 511.
Spencer, Robert F. 46. The phonemes of Keresan. IJAL 12.229-36.
spirantization, 1520.
sporadic intersection, 63.
Spotts, Hazel. 53. Vowel harmony and consonant sequences in Mazahua (Otomi). IJAL
19.253-8.
spread (of lips), 1510.
Stetson, R. H. 36. The relation of the phoneme and the syllable. Proceedings of the Second
International Congress of Phonetic Sciences 245-52.
. 45. Bases of phonology. Oberlin.
 A MANUAL OF PHONOLOGY 243

Stevens, S. S., and J. Volkmann. 40. The relation of pitch to frequency: a revised scale.
American Journal of Psychology 53.329-53.
Stockwell, R. P., and C. W. Barritt. 51. Some Old English graphemic-phonemic correspond-
ences: ae, ea and a. SIL Occasional Papers 4. Washington.
stop, 1520, 1521.
stops, coarticulated, 1521.
, glottalized, 1521.
stress-timed rhythm, 166.
strong insolability (of a term in an accentual system), 230.
structure and pattern, 4.
Sturtevant, Edgar H. 40. The pronunciation of Latin and Greek 2 Special Publications, .

Linguistic Society of America. Philadelphia.

subordinate (in IC structure), 321.
"substance" (vs. "form", Hjelmslevian), 02 end.
Sui. Tai; southeastern Kueichow, China. Li 48. 252112, 25212, 2521321, 252215, 252222,
2525.
Supide. Senufo group; Sikasso, French Sudan. Welmers 50b. 2442, 2443, 2445, 25210, 252112,
2521322, 2521341, 252214, 252225, 2525.
Supple, Julia, and Celia M. Douglass. 49. Tojolabal (Mayan): phonemes and verb mor-
phology. IJAL 15.168-74.
suprasegmental, 242.
surface (spirant), 1522.
Susman, Amelia. 43. The accentual system of Winnebago. New York.
[Schultz], Amelia. 51. Segmental phonemes of Brienzerdeutsch. SIL 9.34-65.
Swadesh, Morris. 34a. The phonemic principle. Lg. 10.117-29.
34b. The phonetics of Chitimacha. Lg. 10.345-62.
.

37. The phonemic interpretation of long consonants. Lg. 13.1-10.

.

. 46a. Chitimacha. LSNA 312-36.

. 46b. South Greenlandic (Eskimo). LSNA 30-54.
. 51. Kleinschmidt Centennial III: Unaaliq and Proto-Eskimo. IJAL 17.66-70.
. 52. Salish phonologic geography. Lg. 28.232-48.
. 53. Mosan I: A problem of remote common origin. IJAL 19.26-44.
Swedish. Scandinavian, Germanic, Indo-European. 2321.
syllable, 220, 22.
syllable juncture (and microjuncture), 2220.
syllable-timed rhythm, 166.
syllable-types, languages classified as to, 22.
symmetrical set (of obstruents); definition 25210.
symmetry, 272.
synchronic (vs. descriptive), 00, 03 end.

Tabassaran. North Caucasian. Trubetzkoy 29. 2441.

"tactics" (term eliminated) 02 end.
Tagalog. North-Central Luzon, Philippines. Bloomfield 17; H. 2442, 252111, 2521322, 252214,
252225, 2525.
Takelma. Penutian? Sapir 22. 25210, 252111, 25212, 2521322, 2521332, 252213, 252225, 2525.
Taki-Taki. Creolized English; Paramaribo and environs, Dutch Guiana. Hall 48c. 2442,
252111, 2521332, 252214, 252225, 2525.
Tangsic. Village of Tangsi; one of the Wu group of dialects, Chinese. Kennedy 52; Ha.
2322, 2525.
Taos. Tiwa, Tanoan. Trager 42, 46, 48. 2212, 2321, 2433, 2442, 2445, 2446, 252113, 2521332,
252225. [Trager 48 reanalyzes as clusters many of the obstruents which we have treated
as units.]
tap [r], 153.
244 A MANUAL OF PHONOLOGY

Tarascan. Michoacan, Mexico. Swadesh 00; Ha. 252112, 2521324, 252214, 252222, 2525.
TC = terminal contour (an element in an intonational system).
TCLC = Travaux du Cercle Linguistique de Copenhague.
TCLP = Travaux du Cercle Linguistique de Prague.
Temoayan: dialect of Otomi, q.v.
tense (vs. lax), 1510.
Tequistlateco: see Chontal.
Terena. Arawakan; near towns of Aquiduana, Taunay, and Miranda, in Mato Grosso,
Harden. 2442, 2445, 252111, 252131, 252213, 252223, 2525.
Brazil.
terminal contour (intonational), 211.
terminal feature (intonational), 212.
Tetelcingo: dialect of Nahuatl, q.v.
Tewa: (Tanoan), see Santa Clara.
Thai. Haas 45-0; Ha. 2221, 2442, 252112, 2521333, 252214, 252225, 2524, 2525.
Tillamook. Coast Salishan. Edel; Swadesh 52. 252112, 2521332, 252212, 252225, 2524, 2525.
timing differences, 166.
Tiwa (Tanoan): Taos, Picuris, Sandia, Isleta. 252213, 2524, 2525.
Tlingit. Na-Dene? Velten. 252113.
Tojolabal. Mayan; southeastern Chiapas, Mexico. Supple. 2431, 2442, 252112, 2521323,
252213, 252225, 2524, 2525.
tone, levels of appearance of, 2612.
"tone languages" (meaninglessness of term), 2612 end.
tone quality, 511.
tongue, in vocoid articulation, 1510.
Tonkawa. Hoijer 46. 2431, 2442, 252113, 25212, 2521321, 252213, 252225, 2525.
Totonac. Hidalgo, Puebla, and Vera Cruz (coast), Mexico. Aschman; McQuown 00; H.
2431, 2441, 252112, 2521321, 252213, 252225, 2525.
Trager, George L. 34. The phonemes of Russian. Lg. 10.334-44.
39a. La syst^matique des phonemes du Polonais. Acta Linguistica 1.179-88.
.

39b. The phonemes of Castillian Spanish. TCLP 8.217-22.

.

41. The theory of accentual systems. LCP 131-45. Menasha.

.

42. The historical phonology of the Tiwa languages. SIL 1:5.

.

44. The verb morphology of Spoken French. Lg. 20.131-41.

.

. 46.An outline of Taos grammar. LSNA 184-221.

. 48.Taos I: A language revisited. IJAL 14.155-60.
, and Henry Lee Smith Jr. 51. An outline of English structure. SIL Occasonal Papers 3.

Washington,
transition (between successive consonants), 162.
trilling, 153.
Trique. In triangle between Tlaxiaco, Putla, and Juxtlahuaca in Oaxaca, Mexico. Longacre.
2312.
Trubetzkoy, N. 29. Zur allgemeinen theorie der phonologischen vokalsysteme. TCLP
1.39-67.
39. Grundzuge der phonologic TCLP 7.
.

Trukese. Malayo-Polynesian; Truk and nearby islands. Dyen 49. 2442, 252113, 2521321,
252215, 252225, 2525.
Tsotsil. Mayan; Chiapas, Mexico. Weathers. 252112, 2521323, 252213, 252225, 2524, 2525.
Tubatulabal. Shoshonean, Uto-Aztecan. Voegelin 35b. 2431, 2521322, 252214, 252225, 2525.
Tunica. Tunican. Haas 41, 46. 2441, 2443, 2445, 252112, 2521321, 252213, 252225, 2525.
Turkish. Turkic. Voegelin 43; McQuown 45-6; Ha. 212, 2441, 252112, 2521341, 252213, 252224,
2525.
Twaddell, W. Freeman. 48. The prehistoric Germanic short syllables. Lg. 24.139-51.
. 53. Stetson's model and the 'supra-segmental phonemes'. Lg. 29.415-53.
two-formant theory. 5310. Rejected 5315.
Tzeltal. Mayan; Chiapas, Mexico. Slocum. 252112, 2521323, 252213, 252225, 2524, 2525.
 A MANUAL OF PHONOLOGY 245

u
Ubykh. North Caucasian. Trubetzkoy 39. 2442.
Ukrainian. Slavic, Indo-European. Trubetzkoy 29; Fairbanks 00; H. 2422.
Uldall, H. J. 54. Maidu phonetics. 20.8-16. UAL
ultimate phonologic constituents (or features or components), 26.
unmarked and marked, 3233.
"usage" (vs. "pattern", Hjelmslevian), 02 end.
uvular trill, 153.

Van Wagoner, Merrill Y. 49. Spoken Iraqi Arabic. New York.

variation, non-distinctive; implications of amount of, 3233.
velic, 14.
Velten, H. V. 44. Three Tlingit stories. IJAL 10.168-80.
Vietnamese. Emeneau 51; Jackson 00; H. 2221, 2322.
Viking Fund (Publications. .) = (now) Wenner-Gren.
.

Villa Alia Zapotec: see Zapotec.

vocal cords, 12.
vocoid, 150, 151; 153.
Voegelin, Carl F. 35a. Shawnee phonemes. Lg. 11.23-37.
. 35b. Tiibatulabal grammar. University of California Publications in American
Archeology and Ethnography 34:2.55-190.
.46. Delaware, an eastern Algonquian language. LSNA 130-57.
.49. Review of K. Pike 47a. IJAL 15.75-84.
.51. Inductively arrived-at models for cross-genetic comparisons of American Indian
languages. Papers from the symposium on American Indian linguistics held at Berkeley
July 7, 1951. [University of California Publications in Linguistics 10.1-68 (1954),
27-45.]
,and M. E. Ellinghausen. 43. Turkish structure. JAOS 63.34-65.
,and E. W. Voegelin. 44. Map of North American Indian Languages. American Ethno-
logical Society and Indiana University; New York.
Vogt, Hans. 36. Esquisse d'une grammaire de g^orgien moderne. Oslo. (Extrait de NTS.)
. 40a. The Kalispel language. Oslo.
. 40b. Salishan studies. Skrifter utgitt av det Norske Videnskaps-Akademi, Oslo, II
Hist.-Filos. Klasse No. 2.
. 44. Le systeme des cas on Ossete. Acta Linguistica 4.17-41.
voice bar (on spectrogram), 533.
voiced and voiceless, 12.
voiceless consonants, acoustics, 532.
voicing, acoustic correlates of, 533, 534.
, bilabial, 153.
volume, mechanism of production, 12.
Votyak. Finno-Ugrian; Kama River, USSR. Trubetzkoy 29. 2442.
vowel (pretechnical) 10; first structural definition, 242; more inclusive structural definition,
2440.
vowel systems, 2440, 244.

W
Waterhouse, Viola, and May Morrison. 50. Chontal phonemes. IJAL 16.35-9.
Weathers, Nadine. 47. Tsotsil phonemes with special reference to allophones of b. IJAL
13.108-11.
Weinreich, Uriel. 53. Languages in contact: findings and problems. Publ. of the Linguistic
Circle of New York 1.

Wells, Rulon S. 47. Immediate constituents. Lg. 23.81-117.

246 A MANUAL OF PHONOLOGY

Welmers, William E. 47. Hints from morphology for phonemic analysis. SIL 5.91-100.
49. Tonemes and tone writing in Maninka. SIL 7.1-17.
.

50a. Notes on two languages of the Senufo group I. Senadi. Lg. 26.126-46.
.

50b. Notes on two languages of the Senufo group II. Sup'ide. Lg. 26. 494-531.
.

52a. Notes on the structure of Bariba. Lg. 28.82-103.

.

52b. Notes on the structure of Saho. Word 8.145-62, 236-51.

.

Western Miwok: see Miwok.

whispering, 023, 12.
white noise, 511.
Whitman, William. 47. Descriptive grammar of Ioway-Oto. IJAL 13.233-48.
Whitney, William Dwight. 1879 (subsequent printings through 1924). A Sanskrit grammar.
Leipzig.
Whorf, Benjamin L. 46a. The Hopi language, Toreva dialect. LSNA 158-83.
46b. The Milpa Alta dialect of Aztec. LSNA 367-97.
.

Wichita. Caddoan. Garvin 50. 2431, 2441, 252113, 252131, 25224, 2525.
Wikchamni: dialect of Yokuts, q.v.
Winnebago. (Mississippi Valley) Siouan. Susman 43; H. 2432, 2441, 2445, 2446, 252112,
25212, 2521321, 252212, 252221, 2525.
Wise, C. M., and Wesley Hervey. 52. The evolution of Hawaiian orthography. Quarterly
Journal of Speech 38.311-25.
Wishram. Chinookan. Sapir 2?. 2441.
Wolff, Hans. 48. Yuchi phonemes and morphemes, with special reference to person markers.
IJAL 14.240-3.
. Osage I: phonemes and historical phonology. IJAL 18.63-8.
52.
Wonderly, William L. 46. Phonemic acculturation in Zoque. IJAL 12.92-5.
51. Zoque II: phonemes and morphophonemes. IJAL 17.105-23.
.

written sources (synchronic analysis via philologic technique), 30.

Yawelmani: dialect of Yokuts, q.v.

Yokoyama, Masako. 51. Outline of Kechua structure I: morphology. Lg. 27.38-67.
Yokuts. Penutian. Newman 44, 46. 2431, 2442, 2511, 25220, 252213, 252225, 2524, 2525.
, Chawchila-Choynimni-Gashowu dialects. 252112, 2521333.
, Chukchansi dialect. 252112, 2521333.
Wikchamni dialect. 252112, 2521333, 252214, 252221.
,

Yawelmani dialect. 2212, 252112, 25212, 2521333.

,

Yoruba. West Africa. Olmsted 51. 252114.

Yuchi. Wolff 48. 2431, 2442, 2445, 2446, 252112.
Yuma. Yuman, Hokan. Halpern 46a,b. 2431, 2442, 2445, 2446, 252112, 2521321, 252214, 252225,
25223, 2525.

Zapotec. Oaxaca, Mexico.

, Isthmus dialect. Pickett 51, 53. 252111, 2523, 2525.
, Sierra dialect. Nellis. 252113, 252213, 252223, 2525.
, Villa Alta dialect. E. V. Pike 48. 252113, 2525.
zero, accentual systems with or without. 230, 23.
Zographensis [the Slavic language of the manuscript called this]. Fairbanks 52. 2443.
Zoque. Mixe-Zoque-Huave, Penutian?; the dialect of Copainald, Chiapas, Mexico. Wonderly
46, 51. 2442, 252112, 2521322, 252215, 252225, 2525.
Zuni. Azteco-Tanoan? Bunzel; Newman 00. 2431, 2442, 252113, 2521323, 252213, 252225,
2524, 2525.
Zyryan. Uralian; northern Russia. Trubetzkoy 29. 2442.
 UNIVERSITY OF FLORIDA

3 1262 05174 1535

...

1 C < l
 Memoir 1. Paul Radin, Winnebago Hero Cycles:
A Study in Aboriginal Literature (168 pp., 1948) $2.00.

Memoir 2. Paul Radin, The Culture of the Winnebago:

As Described by Themselves (119 pp., 1949) $2.00.

Memoir 3. Paul Radin, The Origin Myth of the Medicine Rite: Three Versions.
The Historical Origins of the Medicine Rite (78 pp., 1950) $2.00.

Memoirs 4 and 5. Harold E. Driver and S. H. Riesenberg,

Hoof Rattles and Girls' Puberty Rites in North and South America (33 pp., 1950)
and Hilda J. Curry, Negative Painted Pottery of Angel Mounds Site and
its Distribution in the New World (57 pp., 1950) $1.00.

Memoir 6. L. S. Freeland, Language of the Sierra Miwok (205 pp., 1951) $3.00.

Memoir 7. Louis Hjelmslev, Prolegomena to A Theory of Language, translated by

Francis J. Whitfield (92 pp., 1953) $3.00.

Memoir 8. Claude Levi-Strauss, Roman Jakobson,

C. F. Voegelin and Thomas A. Sebeok, Results of the Conference of Anthropologists
and Linguists (67 pp. 1953) $1.00.

Memoir 9. Harold E. Driver, John M. Cooper, Paul Kirchhoff, Dorothy Rainier Libby,
William C. Massey, and Leslie Spier, Indian Tribes of North America (Map + 30
pp., 1953) $1.00. Outline maps 11 x 17 inches may be purchased separately.

Memoir 10. Charles E. Osgood and Thomas A. Sebeok, ede.,

Psycholinguistics: A Survey of Theory and Research Problems (203 pp., 1954) $2.50.

Memoir 11. Charles F. Hockett, A Manual of Phonology (246 pp., 1955) $3.50.

Orders for Memoirs may be sent to International Journal of American Linguistics, De-
partment of Anthropology, Indiana University, Bloomington, Indiana.