Perceptual similarity and articulatory reinterpretation as … · Perceptual similarity and articulatory reinterpretation as a source of phonological innovation Jonasson, J. journal:

Dept. for Speech, Music and Hearing

Quarterly Progress andStatus Report

Perceptual similarity andarticulatory reinterpretationas a source of phonological

innovationJonasson, J.

journal: STL-QPSRvolume: 12number: 1year: 1971pages: 030-042

http://www.speech.kth.se/qpsr

STL-QPSR 1/1971 30.

D. PERCEPTUAL SIMILARITY AND ARTICULATORY REINTERPRETATION AS A SOURCE OF PHONOLOGICAL INNOVATION

0. Contents.

1. The goals of phonological theory.

2. Articulatory reinterpretation can occur when different ar t iculatory g e s t u r e s have s imi lar perceptual effects.

3. Some cases where articulatory reinterpretation seems to have played a role.

4. Can marking conventions h a n d e reinterpretat ion?

1. 1 Phonological theory should enable u s to give an adequate description

of phonological processes . As pointed out by Halle (1964a, p. 329) the choice

of a specific set of distinctive features "must be understood a s a statement

about the nature of human language in general", i. e. i t i s assumed that a

res t r ic ted se t of distinctive features is capable of defining the segments of

all human language s . This claim of universali ty has been interpreted in two ways:

(1) every possible phonemic opposition should be provided fo r

(Jakobson' s approach), and

(2) every possible phonetic difference should be provided for .

Chomsky and Halle (1968, p. 297) acknowledge the la t te r level. Ladefoged

(1967, p. 57) is of the same opinion: "a linguistic theory should be able to

character ize both the oppositions within a language (the difference between

the members of the set of a l l possible sentences) and the contrasts between

languages (al l and only the features which m a r k the sounds of the language

a s being different f rom the sounds of other languages). "

1.2 Phonological theory should explain why cer tain sound pat terns a r e

favored r a the r than others , e. g. why five -vowel sys tems normally have this

shape i u ra ther than, say, i e o a e 5 o *

L

Dept. of Speech Communication, KTH and Dept. of Phonetics, Institute of Linguistics, University of Stockholm, Box 670 1, 113 85 Stockholm, Sweden.

STL-QPSR 1/197i 3 1.

1 .3 Phonological theory should offer an explanation of the fact that

children l ea rn to m a s t e r the sounds of their language in a specific o rde r

(Jakobson, 1969).

1.4 Phonological theory, being a statement about the nature of human

language, should include some cr i te r ia f o r defining natural c lasses of seg-

ments. Halle (1964a, p. 328) pointed out that the distinctive feature system

i s useful for defining these classes , the specification of a natural c lass in-

volving fewer features than an individual member of such a class.

Fur thermore a n evaluation procedure is p a r t of the generative theory.

An ea r ly proposal was described by Halle (1964b, p. 335). The guiding

principle was simplicity, in t e r m s of the number of alphabetic symbols. I I

"Given two alternative descriptions of a part icular body of data, the descr ip-

tion containing fewer such symbols will be regarded as s impler and will,

therefore, be p re fe r red over the other, I t

Halle' s suggestion i s apparently to base predictions of natural ru les ,

systems, and c lasses on 3 measure of phonological distance defined in t e r m s

of feature specifications.

2. Articulatory reinterpretation: a discontinous phenomenon of language acquisition

2.1 The evaluation procedure i s not only considered a tool for the l in-

guist, when choosing between competing g rammars , but i t i s a lso claimed

to operate when a child i s learning his native tongue ( ~ a l l e , 1964b, p. 344):

"It has been suggested by Chomsky that language acquisition by a child may

best be pictured a s a process of constructing the simplest (optimal) g rammar

capable of generating the se t of utterances, of which the utterances heard by

the child a r e a representative sample" and s imilar ly (Chomsky and Halle,

1968, p. 25 1): ltAn essential feature of our theory of language i s that i t in-

cludes an evaluation measure which makes i t possible to assign values to

alternative grammars . It i s on the basis of this evaluation measure that a

child learning a language chooses one of the g r a m m a r s (of which there a r e ,

in principle, infinitely many) compatible with the fair ly res t r ic ted body of

linguistic data to which he has been exposed. The g rammar that a child

constructs in learning his native tongue will therefore always be the one

that ranks highest in t e r m s of this evaluation measure . I t

STL-QPSR 1/1971

In Halle' s view (1964b, p. 344 ff) language change i s attcibuted to dis-

continuities in the g r a m m a r s of successive generations (for a c r i t ic i sm of

this view, see Weinreich, Labov, Herzog, 1968, p. 143 ff).

2 , 2 Each child, when learning his native tongue, has to construct his

own g rammar on the basis of scat tered and degraded linguistic data. The

p r imary data available a r e auditory stimuli, which a r e given ar t iculatory

interpretations b y the child. During this process there i s a possibility for

the child to make an articulatory reinterpretation of perceptually s imi lar

segments. Suppose that a given language uscs a segment whose phonetic

realization i s [dl!]. The child, learning this language, is exposed to a

number of utterances containing this segment in various contcxts and t r i e s

to m a s t e r the pronunciation of i t himself. In this par t icular case, there

a r e more than one articulatory configuration that will produce an acoustic

and perceptual pat tern s imi lar to that associated with the velarized l a t e ra l

of the mature speakers. It i s fo r instance possible to use a labio-velar

[w] instead. Fig. I-D-1 shows spectrograms of the nonsense words [iwi]

and [i* i] a s spoken by a phonetically trained Swedish speaker. It can be

seen that the consonantal portions a s well a s the transit ional pat terns a r e

closely s imilar . It does not seem unjustified to assume that in cases of

this sor t , that i s , in cases of perceptual s imilar i ty and ar t iculatory a m -

biguity, i t i s more likely than otherwisc that the child re in terpre ts the a r t i c

ulatory character is t ics of speech sounds.

This view i s not a t variance with the view that sound change is g rammar

change. The point i s that the grouping together of cer tain segments which

i s not naturally captured in purely ar t iculatory t e r m s can be accounted for

if the role of perception i s considered. It i s not implied that this sound

change i s gradual. The "sound change" i s brought about by a change in

g rammar , but this change i s "sugge sted" to the child by the perceptually

s imilar effect of two different ar t iculatory gestures.

If i t can be shown that segments can be grouped on the bas is of perceptual

similarity, i r respect ive of their art iculatory character is t ics , there may be

a strong argument for taking such s imilar i t ies into account, when construck-

ing a distinctive feature system.

Below we shall demonstrate a few cases where reinterpretations of this

s o r t have occurred and we shall argue that cur rent formulations of phonology

Fig. I-D- I. Spectrograms showing [il i] and [iwi], spoken by a phonetically trained Swedish speaker.

STL-QPSR 1/1971 33.

(including marking conventions) do not offer the formal machinery appro-

pr iate for predicting reinterpretations in a natural manner.

3. Some cases of articulatory reinterpretation

3.1 The articulatory correlates of the feature - flat, introduced by

Jakobson e t a1 (1963, p. 31), a r e constrictions of the front o r back orif ices

of the vocal tract. Uzbeli and Bantu languages, which lack pharyngealized

sounds, labialize the sounds that correspond to the Arabic emphatic (phar-

yngealized) sounds in words borrowed f rom Arabic, which indicates that

there i s a great perceptual simil a r i ty be tween pharyngealization and labiali-

zation. Jakobson a s s e r t s that no language can use an opposition of labiali-

zation and pharyngealization, apparently because of the perceptual similarity

of the two articulatory gestures. There i s nothing in the framework of

Chomsky and Halle to suggest such a restr ic t ion upon the occurrence of I labialization and pharyngealization.

It i s noted by Chomsky and Halle ( 1968, p. 3 10) that labialization and

velarization never act independently in a given phonological system. We

suggest that this might be due to the perceptual s imilar i ty between the two

gestures, a fact that cannot be properly expressed within a purely articula-

tory system.

On the other hand, the fact that palatalization, velarization, and phar-

yngealization a r e mutually exclusive i s expressed in a natural way with r e -

ference to articulatory factors (Chomslcy and Halle, 1968, p. 307).

3.2. 1 Latin ve lars have become labials before dentals in Rumanian,

for example /linnum/--as /limn/ and /olcto/-3 /opt/ (King, 1969, p. 1 15).

This process is expressed a s follows in Jakobson' s system:

- cont

[+camp]-+-[-comp] / .. ..

1 Apparently the process i s due to an ass i~ni la t ion , the non-compactness of

the dental influencing the velar , while the graveness i s retained. If the

fea tures a r e interpreted perceptunlly, the process is not an unnatural one.

In t e r m s of Chomslry and Hnlle the process is:

~ ~ E ~ ~ ~ ] + [ ~ E ~ f ] 1 1 - c o r [ ~ z Z ] 1 -ant i +ant - cont / +ant i j L

STL-QPSR 1/1971 35.

Chomsky and Halle (1968, p. 177) specify these segments in the following

way (only features relevant to the discussion a r e considered):

rLfl CQI Csl coronal - t f

strident t - t

Applying a feature counting procedure, one would ra ther expect LO]-$ [s]

than [ Q ] 4 [ f ] , since the f o r m e r involvc s a change in one feature and the

la t ter a change in two.

One could improve the situation by making [f] [-strident], a s suggested

by Wooley (1968, p. 141). Then both processes will involve a change in one

feature. In this case fcature counting will indicate that the processes

[ Q ] --B [ f ] and [ Q ] - a [s ] a r e equally likely, which i s contrary to observed

facts. The confusion mat r ices presented by Miller and Nicely (1955, p.

338 ff) indicate that the distinction b e t ween [ f ] and [ Q ] i s m o r e difficult

to perceive than the distinction between [f] o r [Q] and [s]. Obviously this

i s due to the fact that both [ Q ] and [f] a r c character ized by a low-intensity

noise spectrum (Heinz and Stevens, 1961; Stevens, 1968) while [s] is char -

acter ized by a high-intensity spectrum.

It appears f rom the prcceding discussion that features have different

perceptual weight - in this case stridency prevai ls over coronality. It

seems reasonable to assume that this fact has to be incorporated in the

theory, in view of the fact that phonological s t ructure i s in p a r t determined

by perceptual facto r s .

3.5. 1 The change f rom an apical [ r ] to an uvular [R] in German is ex- V

plained by Zirmunskij ( 1962, p. 377) in a manner compatible with our de - finition of ar t iculatory reinterpretation: "Es besteht im Gegenteil a l le r

Grund zu de r Annnhme, dass der Ausgangspunkt fiir dicsen Vorgang im

Deutschen bei der weit vcrbrei te ten Reduktion de s alveolarcn - -r nach Vokal

und im Vlortauslaut liegt. Der reduzierte Vokal, de r das -r in d ieser I

- Stellung erse tz t , hat srtilculatorisch and akustisch eine gewisse ~ h n l i c h k e i t

m i t dem Hinterzungen -R; e s fiihrt zum Schnarrlaut bei dem Versuch, den

Laut zu verdeutlichen, dxs heiss t ihn deutlicher a l s Konsonant auszu - sprechen."

STL-QPSR 1/197 I 3 6.

3. 5.2 There a r e two main variet ies of /r/ in Swedish. [R] i s typical

of Southern Swedish, while [ r ) i s typical of Northern Swedish. The uvular

variety i s c lear ly gaining ground, the border between the a r e a s slowly

moving north-wards. Since there i s no t rend of migration of Southerners

to the North, we mus t assume that a g rammar change of the following kind

is spreading north-wards (only features relevant to the discussion a r e con-

sidered):

(5.8 anter ior i ty i s involved in the change if we assume an alveolar, ra ther than a retroflex front /r/).

t c o r 1 -back +back

/

Despite the complexity of this rule there a r c many languages, such as

French, German, Norwegian, and Swedish, where there is a dialectal var ia -

t voc +cons -low t cont -high

tion between a front and a back /r/. I t does not scem unjustified to assume

that this fact i s explicable with reference to perceptual factors. The p e r -

ceptual s imilar i ty of a front and a back /r/ would a lso offer an explanation

to the fact that a phonemic distinction between [ r ] and p] i s extremely

r a re . In fact, w e know of only one language, P r o v e n ~ a l , where [ r ] and

[R] a r e said to contrast (Coustenrble, 1945, p. 93).

3.5.3 The change /r/* /w/ occurs in Swedish dialects and in English

child language (Weir, 1962, p. 49; Leopold, 1947, p. 71; Lewis, 1936,

p. 268; Moskowitz, 1970, p. 430). This change is expressed as follows in

the Chom sky-Halle framework:

I t voc +cons +ant t c o r - round -high 1 -back

! +round +high

i tback %. J

In view of the grea t complexity of the rule one would regard this change a s

quite improbable, but in fact the confusion between /r/ and /w/ is not un-

common a t all. We assume that perceptual fac tors play an important role

in this case.

STL-QPSR 1/197 1 37.

3.6.1 In Pol ish there has been a change f rom a velarized la tera l [+ 1 to a labio-velar semi-vowel [ w] . In t e r m s of the Chomsky-Halle framework

we get the following rule:

+ cont thigh +back

Assuming an evaluation procedure based on feature counting we would have

to regard this change a s highly improbable. We know, however, that this

change i s very common, which i s quite natural if perceptual similarity is

taken into account. Fig. I-D- 1 shows spectrograms of the utterances [ i Y i]

and [iwi] a s spoken by a phonetically trained Swedish speaker. There a r e

striking s imilar i t ies between the utterances, a s already mentioned in sec-

tion 2.2.

3.6.2 The same change i s reported fo r English child language (Leopold,

1947,. p. 65; Moskowitz, 1970, p. 428).

3.6.3 Latin /1/ was velarized before consonants, according to gram-

mar ians of the fourth and fifth century (Bouriez, 1955, p. 255). This

velarized la tera l has developed into a [w] o r [u] in old French, e. g.

alba -+ aube, pulrnone poumon, where the pronunciation was diphthongal,

ra ther than monophthongal a s in contemporary French.

3.6.4 The same process has occurred in English a s well, in words

such a s walk, -- stalk, and - talk, where the le t te rs -al- have been pronounced

Caul (Luick, 1914, p. 605).

3.7 In the Chomsky-Halle framework the features defining the places of

articulation a r e the same for plosives and nasals. This implies that changes

in place of articulation be equally likely among plosives and nasals. Green-

berg (1966, p. 65), however, remarks : "that a change m y n i s not un-

common, but b +d o r p -it i s practically unheard of.

Data published by Mohr and Wang (1968, p. 37) indicate that the distances

in a perceptual space between nasals a r e smal ler than between the co r re -

sponding plosives. The same finding i s reported by Pruzansky (197 1).

STL-QPSR 1/197 1 38.

Under the assumption that the probability of a phonologic change is inversely

correlated with the perceptual distance between segments, one would have

to account for the differences between nasals and plosives in this respect,

by including perceptual factors in distinctive feature theory.

The fact that differences in places of articulation a r e more easily per-

ceived in plosives than in nasals, can probably be explained with reference

to the undifferentiated spectral shape of the nasals, which decreases the

perceptual value of the place of articulation cues.

Another characteristic of nasals i s the fact that they a r e very often a s -

similated, a s to place of articulation, to a following stop. This fact can be

expressed a s follows:

[+nasal7

This i s descriptively correct, but in the proposal of Chomsky and Halle

there i s no particular reason why this process should not a s easily affect

[-nasal] segments. Again we suggest that the explanation of the behavior

of nasals, a s opposed to other segments, i s to be found in the small percep-

tual value of place of articulation distinctions. In the position before a stop

the articulatory effort of maintaining a separate place of articulation of the

nasal i s rewarded by a very small perceptual effect. So the "lazy mouth1'

i s allowed to minimize its articulatory effort by assimilating the place of

articulation of the nasal to the following stop.

1 -

*

aant p cor high

tj back &low

4.

To what extent will current phonology be capable of handling articulatory

reinterpretation? It might for example be proposed that rnarking conventions I

be extended so a s to include reinterpretation. In our opinion it might be

possible to do so but such modifications a r e not likely to eliminate the de-

scriptive nature of the conventions. Fo r a discussion of the intrinsic con-

tent of features, see Fromkin (1968).

r cyant pco r yhigh 6ba ck

- €low cont C

To take one example, by convention no. XI i s stated that if a language I

has five vowels, the natural case i s a vowel system of this kind i e o U ;

(2) a

STL-QPSR l/1971 40.

This determines the operational hierarchy of levels of decreasing pertinence:

perceptual, aural, acoustical, and articulatory (the la t te r carrying no direct

information to the receiver). I t

"To sum up, the specification of the phonemic oppositions may be made

with respect to any stage of the speech event f rom articulation to perception

and decoding, on the sole condition that the variables of any antecedent stage

be selected and correlated in t e r m s of the subsequent stages, given the

evident fact that we speak to be heard in ord.er to be understood."

In fact, speakers of Bantu languages, when pronouncing Arabic loan-

words, behave in accordance with the perception-oriented definition of the

flatness feature, a s already mentioned in section 3. 1.

Since the chi ld 's f i r s t contact with his mother-tongue is mediated by

auditory, not articulatory stimuli, i t should not cause surpr ise that percep-

tual factors shape phonological structure. Consequently there i s no reason

to expect that a theory based principally on articulation ( ~ h o m s k y - ~ a l l e )

should give an adequate description of the events.

On the other hand, the pr ice paid for Jakobson' s l is tener -oriented ap-

proach i s that complications a r i se in the assignment of a correc t articula-

tory interpretation of flatness, a s pointed out by McCawley (1967, p. 112 ff).

Rather, i t seems reasonable to expect that all links in the speech com-

munication chain should contribute to phonological s t ructure, e. g. ar t icula-

tion favors and disfavors certain segments, while perception involves other

preferences and constraints. Consequently we can only hope to explain

phonological rules and phonological s t ructure if we succeed in constructing

adequate models of al l aspects of the speech communication process . F o r

a discussion of some concrete applications of phonetic theory to phonology,

see Jonasson (forthcoming), Lindblom (forthcoming), and McAllister (forth-

coming).

References: (The date of publication re fe r s to the editions used by the author. The orig- inal date of publication, when known, i s indicated in square brackets)

Bourciez, E. (1955): P r & c i s deJhon&tique -- -- --- francaise, Pa r i s .

Chomsky, N. and Halle, M. ( 1968): The - Sound - Pat tern of English, New York.

Coustenoble, H. N. (1945): La phon&tique du provencal moderne, Hertford. I

cont. next page

STL-QPSR 1/1971 41.

Fant, G. (1 969): "Distinctive Fea tures and Phonetic Dimensions", STL- QPSR 2-3/1969, pp. 1-18 (will a iso appear in the Proc . of the Second Congress of Applied Linguistics, Cambridge, England).

Fromkin, V. A. (1968): "The What and T1ihy of Intrinsicalness", UCLA, Working P a p e r s in Phonetics 10, pp. 156- 168.

Greenberg, J. H. (1966): Language Univer-, The Hague and Pa r i s .

Halle, M. (1964a)C 19581 : "On the Bases oi Phonology", in The Structure of Language (Eds. ) Fodor, J. A. and Katz, J . J. , Englewood Cliffs.

Halle, M. (1964b)[ 19621 : "Phonology in Generative Grammar", in The - Structure of Language (Eds. ) Fodor, J. A. and Katz, J. J'. , Englewood Cliffs.

Heinz, J. M. and Stevens, K. N. (196 1): "On the Proper t ies of Voiceless Fricat ive Consonants", J. Acoust. S o c Am. 33, pp. 589-596.

Jakobson, R. (1969)[: 19411: Kindersprache, Aphasie und allgemeine Lautgc setze, Stuttgart.

Jalcobson, R. (1962): "Principles de phonologie historique", Selected 1:Jritings I, Phonological Studies, The Hague, pp. 221-230.

Jakobson, R., Fant, G., and Halle, M. (1963)[1952]: Pre l iminar ies to Speech Analysis , Carnbridgc, Mass.

Jonas son, J. 0. (forthcoming): "Perceptual Fac to r s in Phonology", to be r ead a t the VIIth International Congress of Phonetic Sciences, Montreal 197 1. ,

King, R. D. (1969): Historical - L i n ~ i s t i c s - - - and - Generative -. - --- - - Grammar , .

Englewood Cliffs.

Ladefoged, P. (1967): "Linguistic Phonetics", UCLA, Working Paper s in Phone t ics 6 . .

Leopold, W. F. (1947): Speech Dgvelopment of a B i l i i s a l Child, 11, Evanston.

Lewis, M. M. (1936): - Infant -- Speech, London.

Liljencrants. J. and Lindblom, B. (forthcoming): "Numerical Simulation of Vowel Quality Systems: The Role of Perceptual ContrastM, to be sgbmitted for publication in Language.

Lindblom, B. E . F. (forthcoming): "Numerical Models in the Study of Spcech and Language", to be read a t the VIIth International Con-. g re s s of Phonetic Sciences, Montreal 197 1.

Luick, K. ( 1964) [ 19 143 : p- Historische - Grammatik - - - der englischen Sprache, I, Stuttgart and Oxford.

McAllister, R. (forthcoming): "The Nuclear S t r e s s Rule and the Descrip- tion of English Stress", to be read a t the VIIth International Congress of Phonetic Sciences, Montreal 197 1.

McCawley, J. D. (1967): " 3 2 ' ro le d' un syst6me de t r a i t s phonologique s dans une thhcmie du langage", Languages - 6, pp. 112 - 123.

I

Miller, G. A. and Nicely, P. E. ( 1955): "An Analysis of Perceptual Con- fusions Among Some English Consonantst1, J. Acoust. Soc. Am. 27, pp. 338-352. -

STL-QPSR 1/1971 42.

Mohr, B. and Wang, IV. S-Y. (1968): "Perceptual distance and the Specifica- tion of Phonological Features", Phonetica 18, pp. 3 1-45.

Moskowitz, A. (1970): "The Two-Year-Old Stage in the Acquisition of Eng- l i sh Phonology", Language - 46 (no. 2, p a r t I) , pp. 426-441.

Postal , P. (1968): Aspects of Phonological Theory, New York.

Principle s of the International Phonetic Association, (1963) [ 19473, London.

Pruzansky, S. (197 1): "Judgments of Similarit ies Among Initial Consonants Using an Auditory Sorting Apparatus", J. Acoust. Soc. Am. 2, p. 8 4 ( ~ ) .

Sivertsen, E . (1960): Cockney Phonology, Oslo.

Stevens, K. N. (1969): ttAcoustic Correlates of Place of Articulation fo r Stop and Fricat ive Consonantst1, M. I. T., QPR No. 89, pp. 199-205.

Sturtevant, E. H. (196 l)[ 19 171 : Linguistic Change, Chicago.

TIJang, W. S-Y. (1968): "The Basis of Speechtt, POLA, Second s e r i e s No. 4, University of Cal ifornia, Berkeley. I

Taleinreich, U., Labov, W., and Herzog, M. I. (1968): t tEmpir ical Founda- tions for a Theory of Language Change", in Directions for Historical Linguistics (Eds. ) Lehmann, W. P. and Malkiel, Y. , Austin, Texas.

TVeir, R. H. (1962): Language in the Crib, The Hague.

Woolley, D. E. (1968): "Some Observations on Stridency in English", P a p e r s f rom the Fourth Regional Meeting, Chicago Linguistic Society.

x

Zirmunskij, V.M. (1962)[1956]: Deutsche Mundartfcunde,Berlin (DDR).