Linguistic Society of America

Sound Patterns in Language

Author(s): Edward Sapir
Source: Language, Vol. 1, No. 2 (Jun., 1925), pp. 37-51
Published by: Linguistic Society of America
Stable URL: [Link]
Accessed: 04-03-2015 16:52 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@[Link].

Linguistic Society of America is collaborating with JSTOR to digitize, preserve and extend access to Language.

[Link]

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE
EDWARD SAPIR

VICTORIAMUSEUM,OTTAWA

There used to be and to some extent still is a feeling among linguists

that the psychology of a language is more particularly concerned with
its grammatical features, but that its sounds and its phonetic processes
belong to a grosser physiological substratum. Thus, we sometimes
hear it said that such phonetic processesas the palatalizingof a vowel
by a followingi or other front vowel ("umlaut")or the series of shifts
in the manner of articulating the old Indo-European stopped consonants
which have become celebrated under the name of "Grimm's Law" are
merely mechanical processes, consummated by the organs of speech and
by the nerves that control them as a set of shifts in relatively simple
sensorimotor habits. It is my purpose in this paper, as briefly as
may be, to indicate that the sounds and sound processes of speech
cannot be properly understood in such simple, mechanical terms.
Perhaps the best way to pose the problem of the psychology of
speech sounds is to compare an actual speech sound with an identical
or similar one not used in a linguistic context. It will become evident
almost at once that it is a great fallacy to think of the articulation of
a speech sound as a motor habit that is merely intended to bring about
a directly significant result. A good example of superficially similar
sounds is the wh of such a word as when, as generally pronounced in
America (i.e., voiceless w or, perhaps more accurately analyzed, aspira-
tion plus voiceless w plus voiced w-glide), and the sound made in blow-
ing out a candle, with which it has often been compared. We are
not at the present moment greatly interested in whether these two
articulations are really identical or, at the least, very similar. Let us
assume that a typically pronounced wh is identical with the sound
that results from the expulsion of breath through pursed lips when a
candle is blown out. We shall assume identity of both articulation
and quality of perception. Does this identity amount to a psycho-
logical identity of the two processes? Obviously not. It is worth
pointing out, in what may seem pedantic detail, wherein they differ.
37

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
38 EDWARD SAPIR

1. The candle-blowingsound is a physical by-productof a directly

functional act, the extinguishingof the candle by means of a peculiar
method of producinga currentof air. So far as normalhumaninterest
is concerned,this sound serves merely as a sign of the blowingout, or
attempted blowing out, itself. We can abbreviate our record of the
facts a little and say that the productionof the candle-blowingsound
is a directly functional act. On the other hand, the articulation of
the wh-soundin such a word as when has no direct functional value;
it is merely a link in the constructionof a symbol, the articulatedor
perceived word when, which in turn assumes a function, symbolic at
that, only when it is experiencedin certain linguistic contexts, such
as the saying or hearing of a sentence like Whenare you coming? In
brief, the candle-blowingwh means business; the speech sound wh is
stored-upplay which can eventually fall in line in a game that merely
refers to business. Still more briefly, the former is practice; the
latter, art.
2. Each act of blowingout a candle is functionallyequivalent,more
or less, to every other such act; hence the candle-blowingwh is, in the
first instance, a sign for an act of single function. The speech sound
wh has no singleness,or ratherprimarysingleness,of reference. It is a
counter in a considerablevariety of functional symbols, e.g. when,
whiskey,wheel. A series of candle-blowingsoundshas a natural func-
tional and contextual coherence. A series of wh-soundsas employed
in actual speechhas no such coherence;e.g., the serieswh(en),wh(iskey),
wh(eel) is non-significant.
3. Every typical human reaction has a certain range of variation
and, properlyspeaking, no such reaction can be understoodexcept as
a seriesof variants distributedabout a normor type. Now the candle-
blowing wh and the speech sound wh are norms or types of entirely
distinct series of variants.
First, as to acousticquality. Owingto the fact that the blowingout
of a candle is a purely functional act, its variability is limited by the
function alone. But, obviously, it is possible to blow out a candlein
a great numberof ways. Onemay pursethe lips greatlyor only a little;
the lowerlip, or the upperlip, or neithermay protrude;the articulation
may be quite impure and accompaniedby synchronousarticulations,
such as a x-like (velar spirant) or sh-like sound. None of these and
other variationsreachesover into a class of reactionsthat differsat all
materially from the typical candle-blowingwh. The variation of wh
as speech sound is very much more restricted. A when pronounced,

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE 39

for instance, with a wh in which the lower lip protrudedor with a wh

that was contaminated with a sh-sound would be felt as distinctly
"off color." It could be tolerated only as a joke or a personalspeech
defect. But the variability of wh in language is not only less wide
than in candle-blowing,it is also different in tendency. The latter
sound varies chieflyalong the line of exact place (or places) of articula-
tion, the formerchiefly along the line of voicing. Psychologicallywh
of when and similar words is related to the w of well and similar
words. Thereis a strong tendencyto minimize the aspiration and to
voice the labial. The gamut of variations, therefore, runs roughly
from hW (I use W for voiceless w) to w. Needless to say, there is
no tendency to voicing in the candle-blowingwh, for such a tendency
would contradictthe very purpose of the reaction, which is to release
a strong and unhamperedcurrent of air.
Second, as to intensity. It is clear that in this respect the two
series of variations differ markedly. The normal intensity of the
candle-blowingsound is greater than that of the linguistic wh; this
intensity, moreover,is very much more variable, dependingas it does
on the musculartone of the blower, the size of the flame to be extin-
guished, and other factors. All in all, it is clear that the resemblance
of the two wh-soundsis really due to an intercrossingof two absolutely
independentseries, as of two independentlines in space that have one
point in common.
4. The speech sound wh has a large number of associations with
other sounds in symbolically significant sound-groups, e.g. wh-e-n,
wh-i-s-k-ey,wh-ee-l. The candle-blowingsound has no sound asso-
ciations with which it habitually coheres.
5. We now come to the most essential point of difference. The
speech sound wh is one of a definitely limited numberof sounds (e.g.
wh, s, t, 1, i, and so on) which, while differingqualitatively from one
another rather more than does wh from its candle-blowingequivalent,
nevertheless belong together in a definite system of symbolically
utilizable counters. Each memberof this system is not only character-
ized by a distinctive and slightly variable articulation and a corre-
sponding acoustic image, but also-and this is crucial-by a psycho-
logical aloofness from all the other members of the system. The
relational gaps between the sounds of a languageare just as necessary
to the psychological definition of these sounds as the articulations
and acoustic images which are customarily used to define them. A

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
40 EDWARD SAPIR

sound that is not unconsciouslyfelt as "placed"1with referenceto

other sounds is no more a true element of speech than a lifting of the
foot is a dance step unless it can be "placed"with referenceto other
movementsthat help to definethe dance. Needlessto say, the candle-
blowing sound forms no part of any such system of sounds. It is not
spacedoff from nor relatedto other sounds-say the soundof humming
and the sound of clearing one's throat-which form with it a set of
mutually necessary indices.
It should be sufficientlyclear from this one example-and there are
of course plenty of analogous ones, such as m versus the sound of
humming or an indefinite series of timbre-varyinggroans versus a
set of vowels-how little the notion of speech sound is explicable in
simple sensorimotorterms and how truly a complex psychology of
association and pattern is implicit in the utterance of the simplest
consonantor vowel. It followsat once that the psychologyof phonetic
processes is unintelligible unless the general patterning of speech
soundsis recognized. This patterninghas two phases. We have been
at particularpains to see that the sounds used by a language form a
self-contained system which makes it impossible to identify any of
them with a non-linguisticsound producedby the "organsof speech,"
no matter how great is the articulatory and acoustic resemblance
between the two. In view of the utterly distinct psychologicalback-
groundsof the two classesof soundproductionit may even be seriously
doubted whether the innervation of speech-soundarticulation is ever
actually the same type of physiological fact as the innervation of
"identical"articulationsthat have no linguistic context. But it is not
enough to pattern off all speech sounds as such against other sounds
produced by the "organs of speech." There is a second phase of
sound patterningwhich is more elusive and of correspondinglygreater
significancefor the linguist. This is the inner configurationof the
sound system of a language,the intuitive "placing"of the soundswith
referenceto one another. To this we must now turn.
Mechanical and other detached methods of studying the phonetic
elements of speech are, of course,of considerablevalue, but they have
sometimes the undesirable effect of obscuring the essential facts of
speech-soundpsychology. Too often an undue importanceis attached
to minute sound discriminationsas such; and too often phoneticians
1 This word has, of course, nothing to do here with "place of articulation."
One may feel, for instance, that sound A is to sound B as sound X is to sound Y
without having the remotest idea how and where any of them is produced.

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE 41

do not realize that it is not enough to know that a certain sound occurs
in a language, but that one must ascertain if the sound is a typical
form or one of the points in its sound pattern, or is merely a variant
of such a form. There are two types of variation that tend to obscure
the distinctiveness of the different points in the phonetic pattern of a
language. One of these is individual variation. It is true that no
two individuals have precisely the same pronunciation of a language,
but it is equally true that they aim to make the same sound discrimina-
tions, so that, if the qualitative differences of the sounds that make up
A's pattern from those that make up B's are perceptible to a minute
analysis, the relations that obtain between the elements in the two
patterns are the same. In other words, the patterns are the same
pattern. A's s, for instance, may differ quite markedly from B's s,
but if each individual keeps his s equally distinct from such points in
the pattern as th (of think) and sh and if there is a one to one cor-
respondence between the distribution of A's s and that of B's, then
the difference of pronunciation is of little or no interest for the phonetic
psychology of the language. We may go a step further. Let us
symbolize A's and B's pronunciations of s, th, and sh as follows:
A: th s sh
B: th1 si sh1
This diagram is intended to convey the fact that B's s is a lisped s
which is not identical with his interdental th, but stands nearer
objectively to this sound than to A's s; similarly, B's sh is acoustically
somewhat closer to A's s than to his sh. Obviously we cannot discover
B's phonetic pattern by identifying his sounds with their nearest
analogues in A's pronunciation, i.e. setting th1 = th, si = variant of
th, sh1 = s. If we do this, as we are quite likely to do if we are ob-
sessed, like so many linguists, by the desire to apply an absolute and
universal phonetic system to all languages, we get the following pattern
analysis:
A: th s sh
A
B: th1 S1 sh1 -

which is as psychologically perverse as it is "objectively" accurate.

Of course the true pattern analysis is:
A: th s sh
B: th1 Si sh1

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
42 EDWARD SAPIR

for the objective relations between sounds are only a first approxima-
tion to the psychological relations which constitute the true phonetic
pattern. The size of the objective differences th-s, s-sh, th-1sl,
s--sh1, th-si, si-s, s-shl, and shr-sh does not correspond to the
psychological "spacing" of the phonemes th, s, and sh in the phonetic
pattern which is common to A and B.
The second type of variation is common to all normal speakers of
the language and is dependent on the phonetic conditions in which
the fundamental sound ("point of the pattern") occurs. In most
languages, what is felt by the speakers to be the "same" sound has
perceptibly different forms as these conditions vary. Thus, in (Amer-
ican) English there is a perceptible difference in the length of the vowel
a of bad and bat, the a-vowel illustrated by these words being long or
half-long before voiced consonants and all continuants, whether voiced
or unvoiced, but short before voiceless stops. In fact, the vocalic
alternation of bad and bat is quantitatively parallel to such alternations
as bead and beat, fade and fate. The alternations are governed by
mechanical considerations that have only a subsidiary relevance for
the phonetic pattern. They take care of themselves, as it were, and
it is not always easy to convince natives of their objective reality,
however sensitive they may be to violations of the unconscious rule
in the speech of foreigners. It is very necessary to understand that
it is not because the objective difference is too slight to be readily
perceptible that such variations as the quantitative alternations in
bad and bat, bead and beat, fade and fate stand outside of the proper
phonetic pattern of the language (e.g., are not psychologically parallel
to such qualitative-quantitative alternations as bid and bead, fed and
fade, or to such quantitative alternations as German Schlaf and schlaff,
Latin ara and ara), but that the objective difference is felt to be slight
precisely because it corresponds to nothing significant in the inner
structure of the phonetic pattern. In matters of this kind, objective
estimates of similarity or difference, based either on specific linguistic
habits or on a generalized phonetic system, are utterly fallacious. As
a matter of fact, the mechanical English vocalic relation bad: bat
would in many languages be quite marked enough to indicate a rela-
tion of distinct points of the pattern, while the English pattern rela-
tion -t* -d, which seems so self-evidently real to us, has in not a few
other languages either no reality at all or only a mechanical, condi-
tional one. In Upper Chinook, for instance, t: d exists objectively
but not psychologically; one says, e.g., mat 'across,' but mad before

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE 43

wordsbeginningwith a vowel, and the two formsof the final consonant

are undoubtedlyfelt to be the "same"sound in exactly the same sense
in which the English vowels of badand batare felt by us to be identical
phonetic elements. The Upper Chinookd exists only as a mechanical
variant of t; hence this alternationis not the same psychologicallyas
the Sanskritsandhi variation-t: -d.
Individual variations and such conditional variations as we have
discussedonce clearedout of the way, we arrive at the genuine pattern
of speech sounds. After what we have said, it almost goes without
saying that two languages, A and B, may have identical sounds but
utterly distinct phonetic patterns; or they may have mutually in-
compatiblephoneticsystems, from the articulatoryand acoustic stand-
point, but identical or similar patterns. The following schematic
examplesand subjoinedcommentswill make this clear. Soundswhich
do not properly belong to the pattern or, rather, are variants within
points of the pattern are put in parentheses. Long vowels are desig-
nated as a; 77is ng of sing; 0 and 6 are voiceless and voiced interdental
spirants;x and y are voiceless and voiced guttural spirants;' is glottal
stop; ' denotes aspirated release; e and o are open e and o.

A: a (E) (e) i u (o) (o)

(a) (e ) (e.) i u (o) ( )

' h w y I m n (v)

p t k
p' t' k'
(b) (d) (g)
f O s x
(v) (8) (z) (y)
but B: a e e i u o J
(a*) (e*) (e-) (i-) (u) (o) (?*)

(' h (IU) (y) (1) m n v,

p t k
(P') (t') (k')
b d g
(f) (0) s (x)
v 3 2 y

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
44 EDWARD SAPIR

We will assume for A and B certain conditional variants which are

all of types that may be abundantly illustrated from actual languages.
For A:
e
1. occurs only as palatalized form of a when following y or i.
In many Indian languages, e.g., ys = ya.
2. e is dropped from i-position when this vowel is final. Cf. such
mechanical alternations as Eskimo -e: -i-t.
3. o is dropped from u-position when this vowel is final. Cf. 2.
4. ? occurs only as labialized form of a after w or u. Cf. 1. (In
Yahi, e g., wowi 'house' is objectively correct, but psychologically
wrong. It can easily be shown that this word is really wawi and
"feels"
o like a rhyme to such phonetic groups as lawi and bawi; short
in an open syllable is an anomaly, buta' is typical for all Yana dialects,
including Yahi.)
5. rj is merely n assimilated to following k, as in Indo-European.
6. b, d, g, v, z,6, y are voiced forms of p, t,kc, f, s,0, x respectively
when these consonants occur between vowels before the accent (cf.
Upper Chinook wa'pul 'night': wabu'lmax 'nights'). As the voiced
consonants can arise in no other way, they are not felt by the speakers
of A as specifically distinct from the voiceless consonants. They feel
sharply the difference between p and p', as do Chinese, Takelma, Yana,
and a host of other languages, but are not aware of the alternation
p:b.
And for B:
1. Long vowels can arise only when the syllable is open and stressed.
Such alternations as ma 'la: u'-mala are not felt as involving any but
stress differences. In A, mala and mala are as distinct as Latin
"apples"
' and
"bad" (fem.).
2. is not an organic consonant, but, as in North German, an attack
of initial vowels, hence 'a- is felt to be merely a-. In A, however, as
in Semitic, Nootka, Kwakiutl, Haida, and a great many other lan-
guages, such initials as 'a- are felt to be equivalent to such consonant
+ vowel groups as ma- or sa-. Here is a type of pattern difference
which even experienced linguists do not always succeed in making
clear.
3. w and y are merely semi-vocalic developments of u and i. Cf.
French oui and hier. In A, w and y are organically distinct con-
sonants. Here again linguists often blindly follow the phonetic feel-
ing of their own language instead of clearly ascertaining the behavior

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE 45

of the language investigated. The difference, e.g., between aua and

awa is a real one for some languages, a phantom for others.
4. 1 arises merely as dissimilated variant of n.
5. p, t', c' are merely p, t, k with breath release, characteristic of B
at the end of a word, e.g. ap-a: ap'. This sort of alternation is com-
mon in aboriginal America. It is the reverse of the English habit:
tame with aspirated t (t'eim) but hate with unaspirated, or very weakly
aspirated, release (heit).
6. f, 0, and x similarly arise from the unvoicing of final v, a and y;
e.g., av-a: af. z and s also alternate in this way, but there is a true s
besides. From the point of view of B, s in such phonems as sa and usa
is an utterly distinct sound, or rather point in the phonetic pattern,
from the objectively identical as which alternates with az-a.2
The true or intuitively felt phonetic systems (patterns) of A and B,
therefore, are:

2 If B ever develops an orthography, it is likely to fall into the habit of writing

az for the pronounced as in cases of type az-a: as, but as in cases of type as-a: as.
Philologists not convinced of the reality of phonetic patterns as here conceived
will then be able to "prove" from internal evidence that the change of etymologi-
cal v, z, 5, -y to -f, -s, -8, -x did not take place until after the language was reduced
to writing, because otherwise it would be "impossible" to explain why -s should
be written -z when there was a sign for s ready to hand and why signs should not
have come into use for f, 0, and x. As soon as one realizes, however, that "ideal
sounds," which are constructed from one's intuitive feeling of the significant rela-
tions between the objective sounds, are more "real" to a naive speaker than the
objective sounds themselves, such internal evidence loses much of its force.
The example of s in B was purposely chosen to illustrate an interesting phenome-
non, the crossing in a single objective phoneme of a true element of the phonetic
pattern with a secondary form of another such element. In B, e.g., objective s
is a pool of cases of "true s" and "pseudo-s." Many interesting and subtle exam-
ples could be given of psychological difference where there is objective identity,
or similarity so close as to be interpreted by the recorder as identity. In Sarcee,
an Athabaskan language with significant pitch differences, there is a true middle
tone and a pseudo-middle tone which results from the lowering of a high tone to
the middle position because of certain mechanical rules of tone sandhi. I doubt
very much if the intuitive psychology of these two middle tones is the same.
There are, of course, analogous traps for the unwary in Chinese. Had not the
Chinese kindly formalized for us their intuitive feeling about the essential
tone analysis of their language, it is exceedingly doubtful if our Occidental
ears and kymographs would have succeeded in discovering the exact patterning
of Chinese tone.

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
46 EDWARD SAPIR

A: a( i u
a* i. u

h w y I m n
t

p t k

p' t' k'

f 8 x S
B: a e e i u o 3

h m n ri

p t k
b d g
S
v 8 X 7
which show the two languages to be very much more different pho-
netically than they at first seemed to be.
The converse case is worth plotting too. C and D are languages
which have hardly any sounds in common but their patterns show a
remarkableone to one correspondence. Thus:
C: a e i u
a e

h w y I m n

p t k q (velar k)
b d g I (velar g)
f s x x (velar x)
D: a e i u
a' e

h v j3 r m

p' t' k' q'

84 5 7 Y(velar y)
f g X5 h (laryngeal h)
3 As in French jour.
4 Bilabial v, as in Spanish.
5 As in German ich.

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE 47

LanguagesC and D have far less superficialsimilarity in their sound

systems than have A and B, but it is obvious at a glance that their
patterns are built on very much more similar lines. If we allowed
ourselves to speculate genetically, we might suspect, on general prin-
ciples, that the phonetic similarities between A and B, which we will
suppose to be contiguouslanguages,are due to historical contact, but
that the deeper pattern resemblancebetween C and D is an index of
genetic relationship. It goes without saying that in the complex
world of actual linguistic history we do not often find the phonetic
facts working out along such neatly schematic lines, but it seemed
expedientto schematizehere so that the pattern conceptmight emerge
with greater clarity.
An examinationof the patterns of C and D shows that there is still
a crucial point that we have touched on only by implication. We
must now make this clear. We have arrangedthe sounds of C and D
in such a way as to suggest an equivalence of "orientation"of any
one sound of one system with some sound of the other. In comparing
the systems of A and B we did not commitourselvesto specificequiva-
lences. We did not wish to imply, for instance, that A's s was or was
not "oriented"in the same way as B's, did or did not occupy the same
relative place in A's pattern as in B's. But here we do wish to imply
not merely that, e.g., C's p correspondsto D's p' or C's h to D's h,
which one would be inclined to grant on general phonetic grounds,
but also that, e.g., C's w correspondsto D's v while C's b corresponds
to D's ,B. On general principles such pattern alignments as the
latter are unexpected,to say the least, for bilabial A resemblesw rather
more than dentolabialv does. Why, then, not allow fAto occupy the
position we have assigned to v? Again, why should D's j be supposed
to correspondto C's y when it is merely the voiced form of s? Should
it not rather be placed under s precisely as, in C's system, b is placed
under p? Naturally, there is no reason why the intuitive pattern
alignment of sounds in a given language should not be identical with
their natural phonetic arrangementand, one need hardly say, it is
almost universally true that, e.g., the vowels form both a natural and
a pattern group as against the consonants,that such stopped sounds
as p, t, k form both a natural and a pattern groupas opposedto the
equally coherentgroup b, d, g (provided,of course,the languageposses-
ses these two series of stopped consonants). And yet it is most im-
portant to emphasizethe fact, strange but indubitable,that a pattern
alignment does not need to correspondexactly to the more obvious

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
48 EDWARD SAPIR

phonetic one. It is most certainly true that, however likely it is that

at last analysis patternings of sounds are based on natural classifica-
tions, the pattern feeling, once established, may come to have a linguis-
tic reality over and above, though perhaps never entirely at variance
with, such classifications. We are not here concerned with the histor-
ical reasons for such phonetic vagaries. The fact is that, even from a
purely descriptive standpoint, it is not nonsense to say that, e.g., the
s or w of one linguistic pattern is not necessarily the same thing as the
s or w of another.
It is time to escape from a possible charge of phonetic metaphysics
and to face the question, "How can a sound be assigned a 'place' in a
phonetic pattern over and above its natural classification on organic
and acoustic grounds?" The answer is simple. "A 'place' is in-
tuitively found for a sound (which is here thought of as a true 'point
in the pattern,' not a mere conditional variant) in such a system be-
cause of a general feeling of its phonetic relationship resulting from
all the specific phonetic relationships (sich as parallelism, contrast,
combination, imperviousness to combination, and so on) to all other
sounds." These relationships may, or may not, involve morphological
processes (e.g., the fact that in English we have morphological alterna-
tions like wife: wives, sheath: to sheathe, breath: to breathe, mouse: to
mouse helps to give the sounds f, 0, s an intuitive pattern relation to
their voiced correlates v, 5, z which is specifically different from the
theoretically analogous relation p, t, k: b, d, g; in English, f is nearer
to v than p is to b, but in German this is certainly not true).
An example or two of English sound-patterning will help us to fix
our thoughts. P, t, and k belong together in a coherent set because,
among other reasons: 1, they may occur initially, medially, or finally;
2, they may be preceded by s in all positions (e.g. spoon: cusp, star:
hoist; scum: ask); 3, they may be followed by r initially and medially;
4, they may be preceded by s and followed by r initially and medially;
5, each has a voiced correspondent (b, d, g); 6, unlike such sounds as f
and 0, they cannot alternate significantly with their voiced correspon-
dents; 7, they have no tendency to be closely associated, either pho-
netically or morphologically, with corresponding spirants (p:f and t:0
are not intuitively correct for English; contrast Old Irish and Hebrew
t:0, k:x, which were intuitively felt relations-Old Irish and Hebrew
o and x were absolutely different types of sounds, psychologically,
from English 0 and German x). These are merely a few of the rela-
tions which help to give p, t, k their pattern place in English.

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE 49

A second example is v? of sing. In spite of what phoneticians tell

us about this sound (b:m as d:n as g:rl), no naive English-speaking
person can be made to feel in his bones that it belongs to a single
series with m and n. Psychologically it cannot be grouped with them
because, unlike them, it is not a freely movable consonant (there aire
no words beginning with rf). It still feels like rg, however little it sounds
like it. The relation ant:and = sink:sing is psychologically as well
as historically correct. Orthography is by no means solely responsible
for the "ng feeling" of 7. Cases like -t7g- in finger and anger do not
disprove the reality of this feeling, for there is in English a pattern
equivalence of -rpg-:-r and -nd-:-nd. What cases like singer with
-7- indicate is not so much a pattern difference -pg-:-X-, which is not
to be construed as analogous to -nd-:-n- (e.g. window:winnow), as an
analogical treatment of medial elements in terms of their final form
(singer: sing like cutter:cut).6
To return to our phonetic patterns for C and D, we can now better
understand why it is possible to consider a sibilant like j as less closely
related in pattern to its voiceless form s than to such a set of voiced
continuants as v, r, m, 77. We might find, for instance, that s never
alternates with j, but that there are cases of s:6 analogous to cases of
f:i and x: y; that ava, aja, ara alternate with au, ai, ar; that com-
binations like -and, -a8g, -ayd are possible, but that combinations
of type -ajd and -avd are unthinkable; that v- and j- are possible initials,
like r-, m-, and ?t-, but that f3-, 6-, r-, e- are not allowed. The product
of such and possibly other sound relations would induce a feeling
that j belongs with v, r, m, n; that it is related to i; and that it has
nothing to do with such spirants as s and 6. In other words, it "feels"
like the y of many other languages, and, as y itself is absent in D, we
can go so far as to say that j occupies a "place in the pattern" that
belongs to y elsewhere.
In this paper I do not wish to go into the complex and tangled
6Incidentally, if our theory is correct,such a form as singer betrays an un-
conscious analysis into a word of absolute significance sing and a semi-indepen-
dent agentive element -er, which is appended not to a stem, an abstracted radical
element, but to a true word. Hence sing: singer is not psychologically analogous
to such Latin forms as can-: can-tor. It would almost seem that the English insis-
tence on the absoluteness of its significant words tended at the same time to give
many of its derivative suffixes a secondary, revitalized reality. -er, for instance,
might almost be construed as a "word" which occurs only as the second element
of a compound, cf. -man in words like longshoreman. As Prof. L. Bloomfield
points out to me, the agentive -er contrasts with the comparative -er, which
allows the adjective to keep its radical form in -rg- (e.g., long with -X: longer
with -ng-).

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
50 EDDWARDSAPIR

problems of the nature and generality of sound changes in language.

All that I wish to point out here is that it is obviously not immaterial
to understand how a sound patterns if we are to understand its history.
Of course, it is true that mechanical sound changes may bring about
serious readjustments of phonetic pattern and may even create new
configurations within the pattern (in Modern Central Tibetan, e.g.,
we have b-, d-, g-: B'-, D'-, G'-,7 while in classical Tibetan we have, as
correspondents, mb-, nd-, 'rg-: b-, d-, g-; mb-, nd-, rtg- are here to be
morphologically analyzed as nasal prefix + b-, d-, g-). But it is
equally true that the pattern feeling acts as a hindrance of, or stimulus
to, certain sound changes and that it is not permissible to look for
universally valid sound changes under like articulatory conditions.
Certain typical mechanical tendencies there are (e.g. nb > mb or
-az> -as or tya> tsa), but a complete theory of sound change has to
take constant account of the orientation of sounds in our sense. Let
one example do for many. We do not in English feel that 0 is to be
found in the neighborhood, as it were, of s, but that it is very close to
6. In Spanish, 0 is not far from s, but is not at all close to 6.8 Is it
not therefore more than an accident that nowhere in Germanic does 0
become s or proceed from s, while in certain Spanish dialects, as so
frequently elsewhere, 0 passes into s (in Athabaskan 0 often proceeds
from s)? In English 0 tends to be vulgarized to t as a tends to be
vulgarized to d, never to s; similarly, Old Norse 0 has become t in
Swedish and Danish. Such facts are impressive. They cannot be
explained on simple mechanical principles.
Phonetic patterning helps also to explain why people find it difficult
to pronounce certain foreign sounds which they possess in their own
language. Thus, a Nootka Indian in pronouncing English words with
vi or I invariably substitutes n for each of these sounds. Yet he is
able to pronounce both X and 1. He does not use these sounds in
prose discourse, but X is very common in the chants and I is often
substituted for n in songs. His feeling for the stylistic character
of t} and for the n-l equivalence prevents him from "hearing" English
t] and I correctly. Here again we see that a speech sound is not
merely an articulation or an acoustic image, but material for sym-
bolic expression in an appropriate linguistic context. Very instruc-
tive is our attitude towards the English sounds j, 77,and ts. All
7 B, D, G represent intermediate stops, "tonlose Medien." In this series they
are followed by aspiration.
8 The slight objective differences between English and Spanish a and 5 are of

course not great enough to force a different patterning. Such a view would be
putting the cart before the horse.

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions
 SOUND PATTERNS IN LANGUAGE 51

three of these sounds are familiar to us (e.g. azure,sing, hats). None

occurs initially. For all that the attempt to pronouncethem initi-
ally in foreign words is not reacted to in the same way. ?a- and tsa-
are naively felt to be incredible,not so ja-, which is easily acquired
without replacementby dja- or ia-. Why is this? 7a- is incredible
because there is no mba-, nda-, ri(g)a- series in English. tsa- is in-
credible because there is no psa-, tsa-, ksa-, series in English; -ts is
always morphologically analyzable into -t + -s, hence no feeling
develops for ts as a simple phoneme despite the fact that its phonetic
parallel tS (ch of church)is found in all positions.9 But ja- is not
difficult, say in learning French, because its articulation and percep-
tion have been masteredby implicationin the daily use of our phonetic
pattern. This is obvious from a glance at the formula:
-j- -z- -8- -v-
z- 5- v-
which is buttressed by:
-s- -0- -f-
S- s- 0- f-
Is it not evident that the English speaker'spattern has all but taught
him j- beforehe himselfhas ever used or heardan actualj-?
There are those who are so convinced of the adequacy of purely
objective methods of studying speech sounds that they do not hesitate
to insert phonetic graphsinto the body of their descriptivegrammars.
This is to confuse linguistic structure with a particular method of
studying linguistic phenomena. If it is justifiable in a grammatical
workto describethe vocalicsystem of a languagein termsof kymograph
records,10it is also properto insert anecdotes into the morphologyto
show how certain modes or cases happenedto come in handy. And
a painter might as well be allowed to transferto his canvas his unre-
vised palette! The whole aim and spirit of this paper has been to
show that phonetic phenomenaare not physical phenomena per se,
however necessary in the preliminary stages of inductive linguistic
researchit may be to get at the phoneticfacts by way of their physical
embodiment. The present discussion is really a special illustration
of the necessity of getting behind the sense data of any type of ex-
pression in orderto grasp the intuitively felt and communicatedforms
which alone give significanceto such expression.
9 Obviously we need not expect -ts and -t< to develop analogously even if s
and I do.
10Needless to say, such records are in place in studies explicitly devoted to
experimentalphonetics.

This content downloaded from [Link] on Wed, 04 Mar 2015 [Link] UTC
All use subject to JSTOR Terms and Conditions