On the role of perception in shaping phonological assimilation rules.

LANGUAGE AND SPEECH, 1992,35(1,2), 59 -72

59

ON THE ROLE OF PERCEPTION IN SHAPING PHONOLOGICAL ASSIMILATION RULES" SUSANL. HURA Uttiversity of Texas at Atistiit BJBRN LINDBLOM University of Texas at Atistin arid Stockliolni University and

L. DIEHL Uitiversity of Texas at Artstin RANDY

Assimilation of na+s to the place of articulation of following consonants is a common and natural process among the world's languages. Recent phonological theory attributes this naturalness t o the postulated geometry of articulatory features and the notion of sprading (hIcCarthi., 1988). Others view assimilation as a result of perception (Ohala, 1990), or as perceptually tolerated articulatory simplification (Kohler, 1990). Kohler notes that certain consonant classes (such as nasals and stops) are more likely than other classes (such as fricatives) to undergo place assimilation to a following consonant. To explain this pattern, he proposes that assimilation tends not to occur when the members of a consonant class are relatively distinctive perceptually, such that their articulatory reduction would be particularly d i e n t . This explanation, of course, presupposes that the stops and rnsals which undergo place assimilation are less distinctive than fricatives, which tend not to assimilate. We report experimental results that confirm Kohler's perceptual assumption: In the context of a following word initial stop, fricatives were less confusable than nasrlls or unreleased stops. We conclude, in agreement with Ohala and Kohler, that perceptual factors are likely to shape phonological assimilation rules.

K e y words: assimilation, feature geometry, nasalconsonants

INTRODUCTION The phottetic status of distinctive features In Preliminaries to Speech Analysis (PSA), Jakobson, Fant and Halle (1952) explicitly

*

This work was supported by a grant from the Advanced Research Program of the Texas Board of Coordination, a grant from the National Science Foundation (No. BNS-9011894), a grant from Humanistisk-Samhallsvetenskapliga Forskningsrldet, Sweden (No. F 149/91), and a grant from the National Institutes;of.*Health (No. DC00427). Address correspondence to Susan L. Hura, University of Texas, Department o f Linguistics, 501 Calhoun Hall, Austin, TX 78712-1 196; E-mail: LIFY471C autxvms.cc.utexas.edu.

Downloaded from las.sagepub.com at CMU Libraries - library.cmich.edu on November 4, 2015

Perceptioii aiid A ssiiiiilatioii

60

stated their belief in the primacy of the perceptual dimensions of distinctive features. For instance, they wrote:

“. . . The closer w e are in our

investigation to the destination of the mesmge (i.e. its perception by the receiver), the more accurately can we gauge the information conveyed by its sound shape. This determines the operational hierarchy of levels of decreasing pertinence: perceptual, aural, acoustical and articulatory (the latter carrying no direct information to the receiver) . ..” (p. 12).

Subsequently, in Tlie Sotifid Parteni of Eiigfisli (SPE), Chomsky and Halle (1968) proposed their alternative set of features, many of which are primarily articulatory in nature. For example, Chomsky and Halle substituted ailterior, coroiral, high, low, back, distributed, covered, and roirizd for Jakobson’s coiiipnct (a so-called ‘resonance’ feature), grave, flat, and sharp (‘tonality’ features). Since the appearance of SPE several proposals have aimed at improving its feature framework. Keating (1.987) reviews the major revisions, One striking similarity among all these revisions is that they retain the articulatory bias of the SPE feature system. The prevalence of articulatory phonetics is currently evident in the paradigm of ‘feature geometry’ (Clements, 1985, 1991; Sagey, 1986; Ladefoged and Halle, 1988; McCarthy, 1988). hlccarthy (1988) augmented the SPE features with: “a set of articulatory features for laryngeal state ([stiff vocal cords], Islack vocal cords], [spread glottis], [constricted glottis]) [Halle and Stevens, 19711 and the feature [labial] [Anderson, 19711” (p. 89).

A recent tutorial chapter by Halle on current feature choices is noteworthy for its emphasis on articulatory dimensions (Halle, 1990, Fig. 3.3; see also Clements, 1991). Chomsky and Halle (1968, pp. 306-308) offered several arguments in favor of the articulatory slant of SPE features. One clear advantage of SPE features derives from the treatment of assiiiiilatioiis Consider the case of a language in which consonants palatalize before the vowel /i/. Let us compare the way such a rule would be formulated in the perceptually-based feature framework of PSA and in SPE terms: PSA:

C

+

C [+sharp]

1-

[

V +diffuse -grave

]

Because the PSA system uses the separate feature [sharp] for palatalization,’ it fails to capture the assimilatory nature of the process. In the SPE treatment; however, the

’

Compare Jakobson’s analogous use of [flat] for labialization, velarization, o r pharyngealiza tion.


S. L. Hiira, B. Liridbloni, arid R. L. Diehl

61

features [high] and [back] are used for both consonants and vowels, so the SPE rule is able to reflect the increased articulatory similarity between the consonant undergoing the change and the vowel that induces it. This is one example showing that the use of articulatory dimensions can accurately describe the nature of assimilation. In fact, the ‘phonetic content’ of many phonological processes does indeed seem to be articulatory. It is therefore not surprising that variants of the articulatory (SPE) framework have been prominent in the recent work of phonologists. Although it appears noncontroversial t o suggest that, at the present stage, some phonological assimilation rules are easier t o describe in articulatory rather than perceptual terms, the absolute articulatory bias of current phonological theory is nevertheless puzzling. There are several questions we might ask about the state of the theory: (i) Was Jakobson really so misguided? Is it safe t o ignore his insights about the primacy of perceptual factors in phonology? (ii) Does the absence of perceptual factors in current phonological theory mean that all major problems associated with the phonetics of distinctive features have been solved? Can we assume that the only processes of phonological interest are articulatory? (iii) Are speakers more important than listeners? If we assume that the factors that shape the phonology of a language originate in the behavior of people producing arid perceiving it, then why should articulatory - but not perceptual - factors receive privileged treatment in our theory? What independent evidence is there to justify this rather lopsided explanatory approach? Below, we summarize three different interpretations of the phonetics of assimilations. The first is an example of current work on feature geometry, as proposed by hlcCarthy (1988). Its basis is articulatory, making no reference to perception. The other two (Ohala, 1990; Kolder, 1990) were developed from phonetic observations. Although differing in certain significant respects, Ohala and Kohler agree that reference t o perceptual information is crucial in a correct account of phonological assimilations. Next we present some experimental results which bear on this issue. Finally, in the light of those results, we return to a discussion of the role of perceptual factors in shaping assimilation rules in particular and sound patterns in general.

Tltree treatments of place assiriiilatiori of nasals Assirnilation arid feature geometry. A cross-linguistically conmon assinulation rule is one where a nasal consonant assimilates in place of articulation to a following consonant. This rule would be written as follows in terms of SPE features: [+nasal]

-.

[

(x anterior

py’;:;l

]

[

o( anterior

/ ; ; ; ; ;0

]

where a, p, y are variables over + and -. hlcCarthy (1988) pointed out a problem with the SPE formulation;.of the rule: The combination of the features [coronal], [anterior], and [back], which describes the very common process of nasal assimilation, “is predicted to be no more likely than an impossible one that assimilates any arbitrary set of three features, like [coronal], [nasal], and [sonorant] ” (p. 86). McCarthy suggested that feature geometry offers a neat solution


62

Perception a n d Assimilatioii

to this problem by creating a hierarchical arrangement of features from which likely combinations can be deduced. This is accomplished by representing place features as daughters of an abstract ‘place’ node, and describing assimilations as spreading of the ‘place’ node association lines, rather than as spreading of individual +/- place features. Feature geometry makes correct predictions in a large number of familiar cases. For instance, consider the phonetic behavior of the in- and ail- prefixes of Swedish verbs’ (Elert, 1970; Linell, Svensson and Ohman, 1971). The final /n/ here assimilates t o the place of articulation of following stops at each of six places of articulation: labial [im-1, [ z m - ] ; labio-dental [IW-], [aq-]; dental [IG-] , [a!-] ; retroflex [~n-] , [sen-]; palatal [~p-] , [sen-] ; and velar [lo-1, [zq-1. Similar patterns occur for English in-, zin-, and eii- prefixes: unbearable, intolerable, untimely, entail, incongruous, unconscious, encroach. In some cases the assimilation has become part of English orthography: impossible, empower. According to a feature geometry account, the reason assimilation is a natural and common process follows from the postulated geometric arrangement of features and nodes and the notion of spreading. From a methodological viewpoint, it is instructive to note how the hierarchical arrangement of the feature geometry is developed as a theoretical construct. It is chosen not on the basis of facts independent of the primary phonological data, but rather solely on the basis of such data. This is in keeping with “the general principle that the justification for the categories and principles proposed for any linguistic level must be supported entirely by evidence pertaining to that level” (Clements, 1985, p. 230). That is, feature geometries are shaped by the phonological data themselves, namely the naturalness of rules like the assimilation of nasals to the place of articulation of following consonants. Assiriiilatioii as perceptiially tolerated artkitlatory sii?iplification. Kohler (1990) reported data showing that apical consonants such as / t / , Id/, and /n/ are particularly prone to undergo place assimilation t o following stops in German casual speech: ailbringen [mb] , aiigeben [ q g ] , mitbringen -* [pb] , mitgehen -, [kg]. However, when apicals occur after labials and dorsals, they remain apicals. Labials and dorsals occurring before apicals also remain unchanged:. abiiehmen [pn] , abdrehen -* [pd] , zuriicknehmen [ k n ] , zuruckdrehen [kd] . Kohler further notes that fricatives d o not participate: Auffahrt -, [ff] , Ausfahrt -, [sf], illustrating the pattern for many /auf-/, /aus-/ pairs in German. The issue here is maintenance vs. neutralization of lexical contrast: For apical stops occurring before other stops, assimilation takes place, potentially neutralizing contrasts, while for non-apicals and fricatives, contrasts are maintained. Such assimilations occur in syllable-final, but not syllable-initial position. Kohler’s interpretation is as follows:

-

-

-+

2

-

-

Our examples here, and elsewhere in the text, are prefixes, but the assimilation effect is not limited to prefixes. It is a productive process, as shown, for instance, by Kohler’s ( 1 990) across word-boundary examples.


S.L. Hura, B. Liiidbloni, mid R.L. Diehl

“. -

63

the initial position in a word has high signalling value for a listener and thus demands greater articulatory precision from a speaker than the final position . . fricatives are not assimilated under any conditions, because they are acoustically and auditorily far more distinct than nasals and unrelwsed stops with regard to place cues so that their articulatory reduction would be too salient, and is, therefore, not tolerated” (p. 88). ~

.

As is well known, the spectrographic place correlates of nasals are rather subtle in comparison with those of stops and fricatives (Fant, 1968, p. 236-253). In an experiment where listeners were asked to judge pairs of speech sounds in terms of perceptual similarity, hlohr and Wang (1968) found the nasals [m] , [n], and [ I ) ] to be significantly more similar to each other than voiced [b], [d], and [g] are, which were in turn more tightly clustered than voiceless [p] , [t] ,and [k] .3 Viewing assimilatory place variations in the context of such acoustic and perceptual observations, Jonasson (1971) suggested that these variations often come about because the perceptual effects obtained by varying their place of articulation are too small to offer robust support for context-independent phonemic contrasts. Accordingly, articulatory simplifications (such as place assimilation in nasals) would be expected to occur only when special perceptual conditions prevail? According t o Kohler’s account, assimilations originate as articulatory simplifications, but their phonologization presupposes perceptual evaluation. An implication of this view for the phonetic status of distinctive features is that these features are neither exclusively articulatory nor exclusively perceptual, but a simultaneous blend of both (Lindblom, 1992). Assiriiilatiori iiidiiced by misperceptioii. Ohala (1984, 1990) reviewed a great deal of evidence indicating that CV transitions are richer in place cues than VC transitions (see also Kawasaki-Fukumori, this volume). He also cited several studies indicating a tendency for listeners to misperceive a VClC2V sequence as having only a single consonant at the place of articulation of C2 (Repp, 1978; Fujimura, hlacchi, and Streeter, 1978; Schouten and Pols, 1983). This effect is particularly pronounced when the duration of the intervocalic closure interval is decreased. As shown by Ohala’s (1990) Figure 1, listeners tend to report hearing a single consonant when the gap between

As a further illustration of the rather subtle nature of place cues in nasals, consider Carlson, Granstram, and Pauli (1972) who investigated the perceptual effects of exchanging the occlusion and the first 0-3 glottal pulses of the vowel in naturally spoken [la], [ n a ] , and [ m a ] syllables. Replacing the occlusion itself had minor effects, a result which is in agreement with previous findings reported by hlalkcot (1956). However, when the replacement consisted of the occlusion plus one or two glottal pulses from the vowel, responses tended to shift drastically in favor of the replacement. These authors concluded that important place and manner cues for nasal and lateral segments are located near the consonant-vowel boundary’ and within a rather narrow time window.

Such conditions could possibly involve creating segments that are in complementary distribution, that preserve perceptual similarity among themselves, but maintain their perceptual distances t o other sound classes relatively unchanged.


64

Perccptiori atid Assimilation

V c and CV transitions is short, whereas more consonant clusters are reported at longer durations. The critical duration at which listeners switch from single-consonant t o consonant cluster responses was found t o be about 95 msec for voiced stops and 150 msec for voiceless stops. These durations match rather closely the durations of stops in these two categories measured in laboratory speech. Ohala (1990) suggested that:

“. . . it

is experience with the natural structure of speech which guides the listener in evaluating and integrating the cues present in speech: since there is a richer, more reliable set of place cues in the CV transition than the VC transition, listeners weigh the former more heavily than the latter in deciding what they’ve heard. Since intervocalic clusters have longer closure durations than single consonants, a longer closure duration is necessary for listeners to hear clusters. Since voiceless stops have longer closure duration than voiced stops, a shorter duration is necessary for the latter to be heard” (pp. 264265).

He went on t o point out that certain assimilatory sound changes (e.g., Latin scripftr scritto (Italian); riocte > iiotte (Italian); priiwi teempus > priritenzps (French); ad+pZicare > applicare (Italian); etc.) “could have occurred due t o less experienced listeners lacking the perceptual ability t o integrate the weaker place cues in the VC transitions” (p. 265). According to Ohala’s account, assiniilations have perceptual roots. They originate as listener errors. Kohler and Ohala differ from McCarthy in attributing an important role to perception. They are in agreement that VC events are perceptually less robust and more vulnerable than CV events. They disagree, however, on the role played by production in the phonologization of assimilations. Whereas Kohler sees a need to appeal to both speaker- and listener-induced factors, Ohala explicitly plays down the role of the speaker in sound change.

>

SO!dE EXPERlhlENTAL RESULTS

The following experiment was designed to provide perceptual data relevant to the various accounts of assimilation described above, particularly that of Kohler (1990). If, as Kohler suggests, assimilation is less likely to occur for consonants that are more distinctive, then we should observe that the less assimilable consonant classes are also less confusable, In particular, English fricatives should be more accurately identified, in the context of a following stop consonant, than either English nasals or unreleased stops. We have no reason t o predict a difference in identification performance between nasals and unreleased stops.

Aierlrods arid procediires Recordings were made of eight native adult speakers of American English, four feniale and four male, reading a sct of nonsense names. The set consisted of nanies where nasals, unreleascd stops, and fricatives occur before a voice-lessstop.


S.L. Hrira, B. Liiidbloiii, mid R. L. Diehl Nasals

stops

Fricatives

Labial:

Shanim Perry

Shanip Perry

Shanif Perry

Alveolar: Posrerior:’

Shanin Terry

Shanit Terry

Shanis Terry

Shaning Kerry

Shanick Kerry

Shanisli Kerry

65

The names were pronounced as follows: /J&nIrn phdii/, /J&nrt’ t l I k i i / , IJ6nr.f chdii/. In order t o have a set of last names where the initial stop was free of the contextual effects of the preceding consonant, we also included a set of names where the last name was preceded by a vowel: Anna Perry, Anna Terry, Anna Kerry. To test Kohler’s claims about perceptual distinctiveness, the original recordings were digitized and waveformedited to create stimuli where each first name was followed by each last name. The original recordings were low-pass filtered (8 kHz cut-off frequency), digitized at a rate of 20 kHz, and split into first name and last name waveform files. First name files included the signal from the beginning of the first name up to, but not including, the burst of the initial consonant o f the last Last name files were created from neutral context ‘‘Anna’’ recordings, and included the waveform from a few milliseconds before the stop burst t o the end. The final versions of the stimuli were created by appending first name and last name files in all possible combinations. This produced 27 items per speaker, for a total of 216 stimuli. All stimuli were randomized to create a single listening test. The stimuli were presented in groups of 10 with 2.5 second inter-stimulus intervals and 5 seconds between blocks. There was a total of 230 items in the test (216 test items, plus 10 practice items, and four filler items), which lasted approximately 20 minutes. The subjects were 22 adult native speakers of American English, who reported no hearing problems. Their task was to identify the first name in each item, and they responded by marking their choice on a preprinted answer sheet.

Restilts Subjects’ responses were scored in terms of error percentages, which averaged 5.2%. Errors were broken down by consonant class and place of articulation and averaged

Note that although English /k/ is conventionally classified as a velar, phonetically the [k I of Kerry is palatal. Also note that if [-II-J] and [-rk] are assumed t o assimilate t o this palatal [ k ] , all consonant sequences are homorganic except [Jk] which is [palato-alveolar] plus [palatal]. For convenience, we have adopted the term “posterior” to refer to the set of sounds [n. k, J]. The first name files were created in such a way as to include the st,op gap as it was produced for a following homorganic stop. In the final concatenated-Cersion of our stimuli, this stop gap is combined with the two non-homorganic stops. The duration of stop gaps for /p, t , k/ are known t o differ in duration, so our stimuli may have been somewhat unnatural since they included the gap duration characteristic of .one stop, but the burst and other place cues of a different stop.


66

Perceptiort arid Assiinilatiori

across subjects. Nasals showed the highest error rate (6.9%),folloived by stops (5.6%) and fricatives (3.0%); o f the three place categories, posterior stiniuli showed the highest error rate (8.0%), followed b y labials (3.9%) and alveolars (3.7%). A three-way analysis of variance (subjects X consonant class X place of articulation) showed significant effects for b o t h consonant class (F (2,42) = 4.4, p < 0.02) and place of articulation (F (2,42) = 6.6, p < O.OI).’ Planned comparisons revealed a significant difference for nasal and stop consonants YS. fricatives (F (2,42) = 23.27, p < 0.01); n o significant difference was found between nasals and stops (F ( 2 , 4 2 ) = 2.68, NS). It appears t o b e easier t o distinguish among t h e fricative consonants than among nasals or unreleased stops in the same environment.8 These results are consistent with Koliler’s assumption that the less assimilable consonant classp tend t o be more distinctive, and thus the results support the idea that there is a perceptual basis for the differences in assimilatory behavior o f nasals, stops, and fricatives. An examination of the pat tern o f confusions showed that, unlike Ohala’s predictions, most errors in this test were non-assimilatory’ (see Table 1). That is, subjects’ incorrect responses did not seem t o be strongly influenced b y the place of articulation of the stop which followed t h e stimulus segment. Instead, t h e pattern o f responses is better characterized as the use of a “default” segment in many cases: 76.1% of all incorrect nasal responses were In/, 73% of all incorrect fricative responses were /J/, and 83% of all incorrect stop responses were I t / . If subjects were responding based on the place of articulation o f the following stop consonant, responses would be balanced among the three places o f articulation i n each consonant class.

’ A second ANOVA was carried out for speakers

X consonant class X place of articulation, which revealed no significant effect of consonant class or place. The failure of these factors t o reach significance in this case is due t o a large amount of individual variation between speakers, and the small number of speakers (n = 8). There appears t o be a tendency for the greatest number of errors for stimuli produced by female talkers t o be for stops, while nasal stimuli accounted for the greatest number of errors for male speakers.

There was only one cross-consonant-class confusion among all subjects for the entire test.

’ While most errors were non-assimilatory across all places of articulation, the propor-

tion of assimilatory and non-assimilatory errors is different for alveolar stimuli than for labial and posterior stimuli. This tendency could be due t o a response bias of our subjects, who, as speakers of English, expect alveolar consonants tp undergo assimilation more frequently than either labial or posterior segments. HkGever, this accounts only for the response pattern for alveolar stops; for alveolar nasals, assimilatory errors do not predominate. Because the overall number of errors made on alveolar stimuli was very small (3.7%), it would be unwise t o base an account of the results o n the proportion of assiinilatory and non-assimilatory errors found here.


S.L.Hirra, B. Liridblont, arid R.L. Diehl

67

TABLE1 Percentage of assimilatory (+A) and non-assimilatory (-A) errors for each manner class (at each place of articulation) Labial +A -A Nasals

16.7 15.7

Fricatives

2.1

stops

1.1 24.1

Average

4.3

11.1 22.2

Alveolar +A -A

Posterior +A -A

Average +A -A

6.5

16.7 41.7

12.0 21.3

36.2 38.3

8.5 10.6

15.6 17.7

5.7

21.8 39.1

10.3 22.9

15.8 17.5

10.9 22.4

2.8 8.0

DISCUSSION The assimilation of nasals to the place of articulation of following stop consonants is a very common process among the world’s languages (Ohala, 1990; McCarthy, 1988; Kohler, 1990). Stops assimilate as readily as nasals in this position, e.g., that boy [pb] , good-bye [bb], Red Cross [gk] (Bailey, 1985, p. 129; Knowles, 1987, p. 130). The assimilation of fricatives to the place of following consonants is not as widely attested. Kohler’s German data appear t o illustrate a general cross-linguistic pattern. Why is it that nasals and unreleased stops tend to assimilate to following stops in this position, but fricatives d o not? Our experiment was an attempt to shed some light on this question. We found that subjects made significantly more errors in the identification of nasals and stops than in the identification of fricatives. How d o these results bear on the three interpretations of phonological assimilations discussed at the beginning of this paper? Irriplications for feature geometry McCarthy’s feature geometry uses representations based upon a hierarchical arrangement of features and nodes. This approach is solely articulatorily-based and gives perception no role in accounting for the naturalness of processes like place assimilation. Our results show a distinct perceptual difference between nasal and stop consonants and fricatives - differences which parallel the likelihood of each class to undergo assimilation to a following stop. That is, we found more misperceptions for nasals and stops, which are more likely t o undergo place assimilation, and fewer :inisperceptions for fricatives, which are less likely to assimilate. Feature geometry is admittedly more successful than SPE notation in describing assimilatory processes. Recall the case of Swedish irz- and art- prefixes, and English irt-, irrt-, and en-, whe’re the final In/ assimilates to the place of articulation of a following


68

Perceptioiz arid Assimilation

consonant. These cases can be accurately described by the spreading of features under the place node in a feature geometry. However, consider English Inis- and dis- prefures, a case where we do not find assimilation. Although conditions are similar to those for in-, zin-, and en-. the final Is/ remains [s] in this case. Why do prefutes ending in /n/ - but not ones ending in Is/ - undergo assimilation? While feature geometry can tell us why nasals assimilate only a certain set of features from following stops (those which fall under the ‘place’ node), it fails to give a principled account of why nasals are more likely t o undergo assimilation than fricatives. It might be suggested that the specification for [nasal] triggers assimilation in this class of sounds. Thus may help descriptively, but it provides no satisfactory explanation, because, a priori feature geometries provide no reason to connect nasality with a tendency to assimilate. The same argument applies if one asserts that it is the specification for [continuant] that prevents fricatives from assimilating. Again, feature geometry offers no dataindependent motivation for linking this feature to the way fricatives behave in potentially assimilatory situations. The Cp-domiriance effect According t o Ohala (1990), it is unnecessary to assume that assimilation reflects an implicit strategy of talkers to simplify articulatory trajectories. Instead, he suggests that assimilation derives from misperception, Specifically, in a VC C2V sequence, the properties of C2 (e.g., place) tend to be perceived as also belonging to C1, particularly at short C1C2 durations. Thus, when listeners repeat the sequence, actual assimilation results. With a fairly high incidence of this kind of perceptual error, the assimilation can become phonologized. On this hypothesis, we might expect that most of the identification errors in our experiment would be based on perceptual assimilation to the place of articulation of the following consonant. In fact, errors of this type did not predominate. However, it would be a mistake to reject Ohala’s hypothesis on the basis of our results, because the VClC2V intervals used in our experiment appear to have been longer than the duration at which perceptual assimilation errors typically occur (Repp, 1978; Ohala, 1990). In other words, our experiment was not designed to provide a clear test of Ohala’s hypothesis. We may nevertheless ask whether our results are otherwise consistent with that hypothesis. Ohala (1990), unlike Kohler (1990), made no explicit predictions about the relative confusability of fricatives, stops, and nasals and their consequent susceptibility to place assimilation. Such differences in confusability, however, could well operate within the framework of Ohala’s hypothesis, t o yield differences among consonant manner classes in likelihood of undergoing assimilation. If (certain) fricatives, for example, are more salient and less confusable than nasals, then they should be less susceptible to perceptual assimilation t o a following consonant. Accordingly, Ohala’s hypothesis is not incompatible with different consonant classes eF$biting varying tendencies toward actual place assimilation.

On the perceptual properties of fricatives Kohler (1990) stated that “fricatives are not assimilated”, being auditorily “far more


S. L. Hiira, B. Liridblorii, arid R.L. Dielil

69

distinct than nasals and unreleased stops with regard t o place cues . . .” (p. 88). Our results are consistent with this claim, showing that it is significantly easier to discriminate among fricatives than among nasals or stops. We should keep in mind that Kohler’s conclusion was made with reference to German fricatives. Is his claim valid as a more general statement? Cross-linguistically, are fricatives more perceptually salient and distinctive than nasals and other consonants? While we cannot fully pursue that question here, we nevertheless offer a few brief remarks on the role of perceptual factors in the formation of phonological structure. Specifically, we are concerned with perceptual differences that seem to exist between sibilant and nonsibilant fricatives. While recognizable as a class, non-sibilants may possess less withinclass distinctiveness than sibilants. First, fricative inventories with a large number of non-sibilant segments are apparently rare in the languages of the world (however, see Austro-Tai and Sino-Tibetan languages in Maddieson, 1984). Second, non-sibilant fricatives seem to be less distinctive than sibilants. In English [co] syllables presented under relatively favorable conditions of filtering and noise (S/N = +12 dB in the frequency band of 200-5000 Hz), /f/ and / O / are misperceived more often than Is/ and /J/, and tend to be frequently confused with each other (Miller and Nicely, 1955, p. 344, Table MI).The perceptual similarity of /f/ and / o / is also reflected in dialect differences and speech development data. For example, “thirty” in RP is [oati], but [fa?i] in Cockney English (Sivertsen, 1960). Black English has a similar pattern where a word like “with” is pronounced [ w ~ f ] (Labov, 1972). And it is well-known that children acquiring English often substitute [f] for [o] (Lewis, 1936; hloskowitz, 1970; Ferguson, 1975). In view of the above considerations, it seems reasonable t o predict that non-sibilant fricatives may assimilate as readily as nasals and unreleased stops, since they d o not seem to po.ssess the perceptual salience and distinctiveness of sibilants. A perceptually-based predictiori

Ohala’s Cz-dominance effect and Kohler’s mechanism o f articulatory/perceptual interaction share the assumption that, in VClC2V sequences, the assimilation of C1 is causally linked to its perceptual weakness. It would seem to follow from both of these accounts that if, in a certain language, we did find fricatives that were no more distinct than nasals and unreleased stops, then we should expect them too to be subject to place assimilation. Is there such evidence? Let us present a few relevant observations. The Japanese /h/ phoneme (Jones, 1967) has three contextual variants, [c] , [h], and [Q] , dies%- distribution is predictable by the identity of a following vowel. While there are many languages that use more “fortis” versions of [ q ] , [ h ] , and [Q] phonemically, the Japanese /h/ variants all seem to be non-sibilant and are perceptually very similar. The phenomenon of “debuccalization” of IS/ should also be mentioned in this context. In South American varieties of Spanish, an [s] -like sound. occurs before vowels, but before other consonants there is a range of realizations including [h], [c], [XI, and devoiced [ F ] and [!] (Jones, 1967). As with Japanese /h/, the pre-consonantal allophones of South American Spanish IS/ may be perceptually indistinct non-sibilants. It is highly likely that perceptual factors are at work shaping such patterns, hence the


Perception arid Assiiirilation

70

extent to which fricatives undergo assimilation or other types of weakening merits further quantitative study. We interpret the present results as supporting the view that perceptual factors play a role in the on-line production of assimilations, as well as in their phonologization. If that conclusion is correct, it has implications for the further development of feature geometry models. Either they need t o be augmented with formal devices that can be justified empirically and, at least partly, on perceptual grounds, or it might be suggested that such a device is already available in the form of the stridency dimension. The feature [strident] does indeed refer to an acoustic and perceptual property, the intensity of fricative noise. This feature could possibly be redefined to reflect the perceptual salience and distinctiveness among segments. The specification for [strident] could then be used to reflect the general sorts of assimilatory patterns we discuss, perhaps by shunting [+strident] sounds away from assimilation rules. There is reason to believe that such an adjusted definition of stridency would fail to capture the real nature of the patterns of assimilation seen cross-linguistically. In our study, fricatives as a cIass were shown t o be the most distinctive segments. Distinctiveness should not be seen as an absolute property of an individual segment; rather, it is a “systemic” criterion. A distinctive segment is one that maintains perceptual contrast with all other segments in the system. A perceptually highly-valued system is one within which all possible contrasts reach high values. In the notation of Liljencrants and Lindblom (1972), it is a system for which, within the space of possible choices (in the universal phonetic space), the expression: n

i-1

i=2

j=1

2

Z

l/(dij)*

is minimized or reaches below a threshold value. (Here ti stands for the size of the system and dii represents degree of perceptual contrast associated with units i andj.) The point captured by the formula’s double summation signs is that the phonological selection of a given element i depends on that element’s perceptual distance from all other elements, just as the selection of any other element j depends on its relation t o the rest of the system. Distinctiveness is not a condition that is met by paradigm members individually. It is a round-robin operation applied painvise to all the logically possible contrasts and evaluated as a collective property of the system. If that is the nature of distinctiveness, we must conclude that classifying segments as [+strident] or [-strident] to block and induce assimilation, respectively, is not going t o be satisfactory. Such labels are segmentspecific and do not reflect the segment’s systemic status. We should also bear in mind that [+strident] segments apparently do become perceptually weak in certain contexts and accordingly may undergo assimilation (e.g., “nice shape” [naIJJci p] ;“does she” [dagJi] (Knowles, 1987)). Furthermore, [+strident] segments induce a phonological pattern whose effect is perceptual disambiguation. An example comes from the history of the plural and genitive ending of English which, in its present form, has a vowel [a] before strident consonants (Halle, 1990). The suffii [XI is the original form, but in hliddle English the schwa was dropped except after


S.L. Hiira, B. Liridbloni, arid R. L. Dielil

71

strident corisoriaiits. According to Kiparsky (1975, pp. 276-277), vowel deletion could still occur even after strident consonants in the time of Shakespeare (cf: Shakespeare: “my mistress’ eyes”). The blocking of schwa delection in present-day English has the effect of avoiding clusters such as [IS, 32, tJs, d p ] that are articulatorily complex but could no doubt also be perceptually ambiguous (Shakespeare: “his horse back”). These examples show that [+strident] segments also participate in perceptually motivated processes. In some cases, they weaken. They d o undergo assimilations. In other situations their perceptual properties cause a strengthening of a sound pattern and can, as just seen, constrain historical development by the retention of an older, and, in certain contexts, perceptually more functional form. Again note that “perceptually more functional” is not an attribute intrinsic to an individual segment or feature. It is a systemic concept. CONCLUSlONS

As shown by the present study, and as argued by Kohler and Ohala, perceptual factors are clearly at work in the formation of phonological assimilations. What phonology needs then is a means of representing these perceptual processes. Exploring ways of integrating the present-day predominantly articulatory distinctive feature frameworks with perceptual facts falls outside the scope of the present paper. However, our preceding analysis has indicated that the perceptual property that appears to play an important part in determining a segment’s phonological behavior is its distinctiveness. We also pointed out that distinctiveness must be seen as a systemic, rather than segment-intrinsic, attribute. Such considerations make it probable that, in the future, problems will not be solved by merely adding a list or a geometry of perceptual features t o existing articulatory dimensions. A formalism must be developed that recognizes the role of systemic forces.

REFERENCES ANDERSON, s. (1971). On the description of ‘apicalized’ consonants. Linguistic Inquiry, 2, 103-107. BAILEY, C.-J.N. (1985). English phonetic transcription. Summer Institute of Linguistics. Dalhs, TX. CARLSON, R., GRANSTROhl, B., and PAULI, S. (1972). Perceptive evaluation of segmental cues. In Quarterly Progress and SIUIUSReport (Department of Speech Communication, Royal Institute of Technology, Stockholm, Sweden), 1,206-209. CHOhlSKY, N., and HALLE, hI. (1968). The Sound Pattern O f Ef?g/iS/J.New York: Harper & Row. CLEhlENTS, C.N. (1985). The geometry of phonological features. Phofiology Yearbook, 2, 223-250. CLEhlENTS, G.N. (1991). Phce of articulation in consonants and vowels: A unified theory. IVorking Papers of the Cornell Plionetics Laboratory, 5, 71 -123. ELERT. C.C. (1970). Ljud oclz ord i svenskun. Stockholm: Almqvist and Wiksell. FANT, G . (1968). Analysis and synthesis of speech processcs. In B. hlalmberg (ed.), Munual of Phonetics (pp. 173-277). Amsterdam: North Holland. PERGUSON, C.A. (1,975). The fricatives in child language acquisition. in L. fkllman (ed.), Proceedings of the Eleventh International Congress of Linguists (pp. 647-663). Bologna: hiulino.


12

Perception aiid Assiiiiilatioii

PUJIMURA, o., hlACCHI, h l f . , and STREETER, L.A. (1978). Perception of stop consonants with conflicting transitional cues: A cross-linguistic study. Language and Speech, 21, 337-346. HALLE, hl. (1990). Phonology. In D. Osherson and €I. Lasnik (eds.), Language, Vol. Z (pp. 43-68). Cambridge, hlA: hlIT Press. IIALLE, hl., and STEVENS, K. (1971). A note o n laryngeal features. M T Research Laboratory of Electronics Quarterly Progress Report. 101, 198-21 3. JAKOBSON, R., FANT, G., and HALLE, hl. (1952). Prelimirtaries to Speech Analysis. Cambridge, MA: h1IT Press. JONASSON, J. (1971). Perceptual similarity and articulatory reinterpretation as a source of phonological innovation. Pilcts (Department of Linguistics, Stockholm University), 8, 30-42. JONES, D. (1967). The Phoneme, its Nature and Use. Cambridge, hlA: IIeffer and Sons. KEATING, P. (1987). A survey of distinctive feature systems. UCLA Working Papers in Phonetics, 66,124-150. KIPARSKY, P. (1975). Comment o n the role of phonology in language, In J.F. Kavanagh and J.E. Cutting (eds.), The Role of Speech in Language (pp. 271 -280). Cambridge, MA: hllT Press. KNOWLES, G. (1 987). Patterns of Spoken Ei~glidi.London: Longmans. KOHLER, K. (1990): Segmental reduction in connected speech: Phonological facts and phonetic explanations. In W.J. Hardcastle and A. hlarchal (eds.), Speech Production and Speech hfodehig (pp. 69-92). Dordrecht: Kluwer Publishers. LABOV, W. (1972). Laiigtcage in !lie Inner City: Studies in the Black English Ver~iacular.Philadelphia: University of Pennsylvania Press. LADEFOGED, P., and HALLE, hl. (1988). Some major features o f the International Phonetic Alphabet. Larigrage, 64,577-590. LEWIS, h l .hl. (1936). Infa/zt Spcecli: A Study of the Beginning of Language. London: Routledge and Kegan Paul. LILJENCRANTS, J., and LINDBLOhl, U. (1972). Numerical simulation of vowel quality systems: The role of perceptual contrast. Language, 48,839-862. LINDBLOh1, B. (1992). Phonological units as adaptive cmergents of lexical development. In C.A. Ferguson, L. hlenn, and C. StoeI-Gammon (eds.), Plionological Development (pp. 000400). Parkton, hlD: York Press. LINELL, P., SVENSSON, B., and OHMAN, S. (1971). Iajitdstrirktitr. Lund: Gleerups. MCCARTHY, J.J. (1988). Feature geometry and dependency: A revicw. P/ionetica, 43,84-108. hlADFIESON, 1. (1984). Patterns ofsound. Cambridge, U.K.: Carnbridsc University Press. MALECOT, A. (1956). Acoustic cues for nasal consonants. Iangirage. 32.274-284. MILLER, G.A., and NICELY, P.E. (1955). An analysis of perccptual confusions among some English consonants. Journal of the Acoustical Societ.)]of America. 27, 338-352. , hlOHK, B., and WANG, W.S.-Y. (1968). Perceptual distance and t h c specification of phonological features. Pfionetica. 18, 31 -45. hlOSKOWITZ, A. (1 970). The two-year+ld stage in the acquisition of English phonology. language, 46,426-441. OHALA, J.J. (1984). Prosodic phonology and phonetics. Plionology Yearbook, 1,113-127. OHALA, J.J. (1990). The phonetics and phonology of aspects of assimilation. In J. Kingston and hl.E. Bccknian (cds.), Papers in Laboratory Pfionolog-vI: Befitvert tlie Graniniar and Plzysics of Speech (pp. 258-275). Cambridge: Cambridge University Press. REPP, B.H. (1978). Perceptual integration and differentiation of spectral cues for intervocalic stop consonants. Perception & Psycl~ophysics,24.471 -484. SAGEY, E. (1986). The representation of features and relations in non-linear phonology. Doctoral dissertation, hllT. SCHOIJTEN, h1.E.H.. and POLS, L.C.W. (1983). Perception of plosive consonants. In hl. van den Broecke, V. van Ileuven, and \V. Zonneveld (eds.), Sound Structures: Studies for Aiitonie Cohen (pp. 227-243). Dordrecht: Foris. SIVEKTSEN, E. (1960). Cockncy Phonology. Oslo: Oslo University Press.


Phonological reduplication in sign language: Rules rule.

Rules for Shaping Neural Connections in the Developing Brain.

Learning general phonological rules from distributional information: a computational model.

Orthographic and phonological neighborhood effects in handwritten word perception.

Lightness perception in simple images: testing the anchoring rules.

Assimilation and contrast in motion perception: explorations in cooperativity.

Role of actin in shaping autophagosomes.

Electrophysiological registration of phonological perception in the subthalamic nucleus of patients with Parkinson's Disease.

The role of nickel in urea assimilation by algae.

The Role of the Microbiota in Shaping Infectious Immunity.

The role of language in shaping international migration.

The role of social context in shaping student-athlete opinions.

The role of maternal care in shaping CNS function.

The potential role of genetic assimilation during maize domestication.

Processing of phonological variation in children with hearing loss: compensation for English place assimilation in connected speech.

Shaping visual space perception through bodily sensations: Testing the impact of nociceptive stimuli on visual perception in peripersonal space with temporal order judgments.

The Role of Phonological versus Morphological Skills in the Development of Arabic Spelling: An Intervention Study.

The shaping of social perception by stimulus and knowledge cues to human animacy.

On the role of the supramarginal gyrus in phonological processing and verbal working memory: evidence from rTMS studies.

Evolution of the perception of the doctor's role in China.

Neuroimaging sheds new light on the phonological deficit in dyslexia.

Relation between Phonological Processing, Auditory Processing and Speech Perception among Bilingual Poor Readers.

Gender hierarchy in the space: the role of gender status in shaping the spatial agency bias.

Phonological and lexical influences on phonological awareness in children with specific language impairment and dyslexia.