Enhanced discriminability at the phonetic boundaries for the voicing feature in macaques

Transcription

1 Pereption &Psyhophysis 1982,32 (6), Enhaned disriminability at the phoneti boundaries for the voiing feature in maaques PATRICIA K. KUHL and DENISE M. PADDEN University ojwahln,ton, Salttk, Wahln,ton DiarimiDation of apeeh IOUDda from three omputer-pnerated ontinua that ranged from voled to voieleessyllables (/be.pai, Ida tai, andip bi) was tested with three maaques. The stimuli on.h ontinuum varied in voie-onset time (VOT). Pain of stimuli that were equally different in VOT were hosen suh that they were either withln-eategory pairs (syllables given the same phoneti label by human Hsteners) or between-eategory pairs (syllables given dif ferent phoneti labels by human Hsteners). Results demonstrated that disrimination performane was always best for between-eategory pain of stimuli, thus rephating the "phoneme boundary effet" seen in adulthatenen and in human infantsas youngas 1 month of age. The flndinp are disussed in terms of their speifi impat on aounts of voiing pereption in human Hsteners andinterms of theirimpatondisussions of the evolution of language. Studies of human infants as young as 1 month ofage have demonstrated that their disrimination of sounds from a ontinuum ranging from one phoneti unit to another (e.g., from /ba/ to /pa/) is typially enhaned the region of the adult-defined ''boundary'' between ategories and is relatively poor within ategories(eimas, 1974a; Eimas, 1975; Eimas, Siqueland, Juszyk, & Vigorito, 1971). This differential disriminability is onduive to the phoneti ategorization of speeh sounds and has been interpreted as supporting the idea that innate fators play an important role in the development of speeh pereption (Eimas, 1974b; Eimas & Tartter, 1979; Juszyk, 1981; Kuhl, 1979a; Morse, 1974). Although a number of authors agree that the infant's abilities are probably innately determined, the preise nature and origins of the infant's predispositions are not known. It has been suggested that these effets reflet the operation of a mehanism speifially designed to detet the aousti properties of speeh sounds (Eimas, 1974b). The possibility has also been raised, however, that these speifi "phoneme-boundary effets" might This researh was supported by a grant to P. K. Kuhl from the National Siene Foundation (BNS ). We wish to thank K. L. Morgan-Landan and M. Kulik for assistane in running animals during the training phase of the experiment. We gratefully aknowledge J. M. Miller, Department of.otolaryngology, for donating spae for the ondut of these experiments, the ooperation of the Haskins Laboratories (NIH Contrat NIH ) in providing the syntheti stimuli, and the Regional Primate Researh Center (NIH Grant RR 00166) for the use of ore failities and the help of ore staff throughout the ondut of this researh. The authors thank A. N. Meltzoff for ritial omments on an earlier version of this manusript. The authors are affiliated with the Department of Speeh and Hearing Sienes and the Regional Primate Researh Center, University ofwashington, Seattle, Washington be attributable to the infant's more general auditory pereptual abilities (Kuhl, 1978, 1979b). Experiments examining the nature and origins of these effets have entered on tests of their speiflityto speeh as opposed to nonspeeh (Miller,Wier,Pastore, Kelly, & Dooling, 1976;Pisoni, 1977), and to humans as opposed to animals (Kuhl, 1981; Kuhl & Miller, 1975; Morse & Snowdon, 1975; Sinnott, Beeher, Moody, & Stebbins, 1976; Waters & Wilson, J976). A partiular advantage of omparative experiments is that if animals demonstrate the phoneme-boundary effet, one an argue onviningly that the effet does not neessitate speeh-speifi mehanisms. This in turn redues our need, in the absene of other data, to argue that suh mehanisms exist. Moreover, animal data ontribute to disussions examining the potential role of auditory onstraints in the evolution of language. Demonstrations of the phoneme-boundary effet in nonhuman speies raise the possibility that the phoneti inventory reflets ertain psyhoaousti onstraints (Kuhl, 1979b; Kuhl & Miller, 1975; Stevens, 1972). There are five published experiments that ompare the pereption of speeh sounds by human and animal listeners. Three of the five studies (Kuhl & Miller, 1975, 1978; Sinnott et al., 1976; Waters& Wilson, 1976) were aimed at determining whether the loation of a pereptual boundary for animal listeners oinided with the loation of the phoneti boundary in human listeners. These studies did not diretly assess disriminability along the ontinuum, but provided support for the notion that ertain nonhuman speies pereptually partition speeh ontinua in the phoneti boundary region as defined by human listeners (see Kuhl, 1979b, for a review). For example, Kuhl and Miller (1975) obtained Identifiation funtions for humans and hinhillasusing three stimulus sets, a bilabial (/ba/ to fpa/), an alveolar Copyright 1983 Psyhonomi Soiety, In /83/ $01.15/0

2 ANIMAL DISCRIMINATION OF ACOUSTIC CUES FOR VOICING 543 (fdal to Ita!), and a velar (fgal to Ikaf) ontinuum. The stimuli on eah ontinuum varied in voie-onset time (VOT), the time in milliseonds between the release of the artiulatory onstrition and the onset oflaryngeal voiing.the animal subjets were trained in a onditionedavoidane paradigm. They learned to respond differentially to good syntheti exemplars of Idal (0 mse VOT) and Ital (+80 mse VOT), taken from the alveolar ontinuum. When performane on these endpoint stimuli was onsistently above 95% orret, the stimuli between 0 VOT and +80 VOT (in lo-mse steps) were presented as generalization stimuli. The results demonstrated that the pereptual boundary for animal and human subjets did not differ signifiantly. In addition, studies using the other two stimulus sets showed that the exat loation of the pereptual boundary depended upon the plae of artiulation (bilabial, alveolar, or velar) of the voied-voieless pair, just as it does for human listeners (Kuhl & Miller, 1978). Two studies diretly addressed disriminability of stimulus pairs from speeh-sound ontinua with animals. Morse and Snowdon (1975) used a heart rate proedure to examine rhesus monkeys' disrimination of stimuli from a syntheti Ibae-dae-gael ontinuum. The proedure involved 20 presentations of one stimulus followed by 20 presentations of a seond stimulus. Heart rate typially habituates to the first stimulus and dishabituates when the stimulus is hanged. The results demonstrated that the rhesus monkeys disriminated both the within-ategory and between-ategory pairs. Both groups demonstrated signifiantly more dishabituation than did the ontrol group, whih was presented with a single stimulus repeated 40 times. However, the degree of dishabituation was signifiantly greater for those animals presented with betweenategory omparisons-that is, with stimuli falling on opposite sides of the human boundary-than for those subjets presented with within-ategory pairs, thus suggesting that the phoneme boundary effet may exist in animals. The seond experiment (Kuhl, 1981) assessed differential disriminability along a speeh ontinuum ranging from Idal to Ital in the hinhilla using proedures typially assoiated with auity studies in psyhophysis. This study provided a diret estimate of the just notieable differene in VOT (~VOT) that ould be deteted by the animal at various VOTs along the ontinuum. The rationale for suh an experiment derived from the argument that sensitivity to a stimulus hange should be maximum at the loation of the phoneti boundary and minimum near the enter of the stimulus ategory. The results of the study showed that the smallest ~VOT values, indiating greatest sensitivity, ourred at the VOT values nearest the phoneti boundary, whereas ~VOT was maximum at points most distant from the boundary region. The present experiment was designed to extend these disrimination data to different voied-voieless ontinua (bilabial, alveolar, and velar series). The testing tehnique was one that has been used in experiments on adult listeners (e.g., Wood, 1976), a same-different disrimination tehnique. METHOD Subjets Three (two male, one female) juvenile Japanese monkeys (Maaa fusatat served as subjets. They were between 1 and 3 years of age at the onset of training. Eah of the animals was housed in an individual age at the University of Washington's Regional Primate Researh Center. They had aess to water in their home ages at all times and were fed one daily at the ompletion of the experimental session. Stimuli The voied and voielessstop onsonants were omputer synthesized at the Haskin's Laboratories (New Haven, Connetiut) on the parallel resonane synthesizer aording to the parameter speifiations desribed by Abramson and Lisker (1970) for bilabial (fba-pa/), alveolar (fda-taj), and velar (fga-ka/) ontrasts. Naturally produed voied and voieless stop onsonants are distinguished by the timing of the onset of laryngeal vibration (voiing) relative to the release of the onstrition in the supralaryngeal musulature. In voied stops (/ba/, /da/, /gaj), the onset of voiing preedes the release of the artiulatory onstrition by some to 5 to 40 rnse, whereas in voielessstops, the release of the artiulatory onstrition preedes the onset of voiing by more than 2540 mse. The preise differene in the timing of these two artiulatory events depends upon the plae of artiulation of the stop onsonant (Lisker & Abramson, 1964). There are a number of aousti features that differentiate synthetially generated voied and voieless syllables; the timing differene desribed above is evidened in the relative onsets of periodiity (voiing) and the burst of energy that ours when the artiulatory onstrition is released. Other aousti fators, suh as the presene or absene of low-frequeny energy in the first formant at the onset of voiing (Liberman, Delattre, & Cooper, 1958; Lisker, 1975; Stevens & Klatt, 1974; Summerfield & Haggard, 1977), the presene or absene of aspiration noise, and the aperiodi signal that results when air rushes through the open glottis (Fisher-Jorgenson, 1954), as well as its intensity (Repp, 1979), are also known to be important aousti ues for the pereption of the voiing feature. In the synthetially generated stimuli reated by Abramson and Lisker (1970) and used in this as well as in many other speeh experiments reported in the literature, these ues ovary. While the stimuli are said to vary in VOT, a name that empha-: sizes the timing ue, the additional aousti fators just mentioned also vary in the stimuli. Abramson and Lisker's (1970) data showed that when adult listeners are asked to label syntheti stimuli that vary in VOT along a ontinuum ranging from voied to voieless stimuli, the loations of the phoneti boundaries (the 50% point on the identifiation funtion) are at approximately +22 mse VOT for the bilabial stimuli, +35 rnse VOT for the alveolar stimuli, and +42 mse VOT for the velar stimuli. Using standard identifiation proedures, we verified that adults tested in our situation produed similar results. In this study, three pairs of stimuli were tested from eah ontinuum, two within-ategory pairs (one voied and one voieless) and one between-ategory pair. Eah stimulus pair differed by 20 mse VOT. The between-ategory pair was hosen suh that it straddled the phoneti boundary on eah ontinuum, and the two within-ategory pairs were hosen to be immediately adjaent on both sides. Tablel lists the nine pairs tested. These values were hosen to oinide with those used by investigators in infant studies (Eilers, Gavin, & Wilson, 1979;

3 5M KUHL AND PADDEN Within Voied Between Voied-Voieless Within Voieless Note-All VOT in milliseonds. Table 1 The Nine Stimulus Pairs Tested in the Experiment Bilabial VOT ovs vs, vs. +60 Alveolar VOT +5 vs vs vs. +65 Velar VOT +10 vs, vs vs. +70 Eimas et al., 1971; Lasky, Syrdal-Lasky, & Klein, 1975; Streeter, 1976). Spetrograms of stimuli from these three series have been published (Kuhl & Miller, 1978) along with more omplete aousti analyses of the signals. Eah syllable was 434 mse in duration and had a fundamental frequeny that was onstant at 114 Hz until the last 100 mse, during whih the fundamental fell to 70 Hz. Apparatus The experiment was onduted in a double-walled, soundproof lac booth. During testing, the animals were restrained in primate hairs. Audio signals were delivered by a two-hannel tape dek (TEAC, Model A-2300S) through a single earphone (TDH-49 with MX-41/AR ushion) to the animal's right ear. A response key was loated diretly in front of the hair, and a green light was mounted at eye level 1 ft in front of the animal. A red light was adjaent to the green light. An automati feeder under omputer ontrol delivered 2 of applesaue through a rubber tube loated near the animal's mouth. A small laboratory omputer (Raytheon 706) ontrolled the delivery of sound and all of the appropriate ontingenies during the experiment. A programmable attenuator (Grason-Stadler, Mode11284) was used to adjust the intensity levels of the signalsduring the training phase of the experiment.: After training, all stimuli were presented at a onstant level (65 db SPL). Information onerning eah trial was printed on an eletroni data terminal (Texas Instruments, Model 700). Proedure A positive-reinforement proedure was employed. The animal initiated trials by depressing the response key when the green light was blinking. As soon as the animal depressed the key, the light stopped blinking and was on steadily. If the animal held the key for the duration of a variable foreperiod (VFP), whih ranged from.01 to 1.2 se, a trial was presented. If the animal released the key before the end of the VFP, a time-out period (TO) ourred, during whih the green light was turned off and the red light was turned on for 7 se and keypressing responses failed to initiate trials. Animals were tested for 1 h eah day. Two kinds of trials, same (S) and different (D), were run with equal probability. During S trials, four idential stimuli were presented at l-se intervals measured onset to onset (e.g., AAAA). During D trials, the first two stimuli were idential to the stimuli presented during S trials, but the last two stimuli were different (e.g., AABB).I In order to be reinfored, the animal was required to ontinue to depress the key for the full duration of the S trials (1.7 se timed from the onset of the third stimulus), produing a "orret rejetion," and to release the key during the 1.7-se trial interval (also timed from the third stimulus) on D trials, produing a "hit" response. If the animal inorretly released the key during the 1.7-se trial interval on an S trial and thus produed a "false-positive" response or failed to release the key during the 1.7-se trial interval on a D trial and thus produed a "miss" response, no food reinforement was delivered and a 7-se TO period ourred. A TO period also ourred if the animal released the response key during the presentation of the first two stimuli on either S or D trials (an "early-release" response). At the ompletion of eah trial the green light was turned off and kept off until the animal released the key for.5 se; after this time interval, the light again began to blink, indiating to the animal that a trial ould be initiated. Trial Struture The nine stimulus pairs were presented in a randomizedblok design using repeated measures. The animals were tested on eah stimulus pair in random order for a 1.5-min period (approximately 20 trials when the animal was working steadily); during that time, S trials (AA pairs) and D trials (AB pairs) for that stimulus pair ourred with equal probability. Eah 1.5-min trial blok was separated by a 5-se pause. In a typial 50-min session, eah stimulus pair was tested three times to provide approximately 60 trials per day per stimulus pair. Preliminary Training The basi proedures used to train the animals were similar to those desribed by Sinnott et al. (1976). Briefly, the animal was plaed in a primate hair eah day and trained, using standard shaping proedures, to press and release the response key for food reinforement. 'fhe animal was gradually trained to depress the key until a sound (the eventual B stimulus) was presented and then to release the key for reinforement. The interval prior to the presentation of the B stimulus (VFP) was slowly lengthened, but ontinued to be varied from trial to trial to prevent the animal from timing his release response rather than listening for the stimulus. When the VFP was approximately 3 se in duration and the animal onsistently held the key down until the stimulus had been presented and released the key as soon as the sound had been presented, a seond stimulus (the A stimulus), attenuated by 50 db, was introdued prior to the B stimulus. The animal ontinued to be reinfored for releasing the bar when B was presented and was given a TO period for releasing to A, as the intensity of A was systematially inreased until it equaled the intensity of B. After the animal had sueeded at this stage in training, S trials (AAAA) and D trials (AABB) were run with equal probability and with all the ontingenies previously desribed in effet. The last step remaining in the pretraining period was the blok-to-blok variation in the stimulus pair being tested. The stimulus pair used during training onsisted of a vowel ontrast (fal vs. Iii), and the stimulus pairs used to adapt the animal to the randomized-blok design onsisted of additional vowel ontrasts (fal vs, 10/), pairs of idential vowels differing in pith ontour (rise vs, fall), and syllable pairs differing in the initial onsonant (fsal vs. IIa/; Ivai vs. lsa/). When performane on these training stimuli was onsistently above 80% orret, disrimination testing began. The training period ranged from 3 to 9 months for individual animals. RESULTS The disrimination data for eah animal were organized in separate 2 X 2 stimulus-response matries like

4 ANIMAL DISCRIMINATION OF ACOUSTIC CUES FOR VOICING 545 Table 2 Stimulus-Response Matrix Computed for Eah Stimulus Pair Response Different Same Different Hit P(D/D) MissP(S/D) Stimulus Same False-Positive pedis) Corret-Rejetion P(S/S) those shown in Table 2. As indiated, the onditional probability of a "hit" is the probability of responding "different," that is, of releasing the response key, when the members ofthe stimulus pair were different [P(D/D)]. Similarly, the onditional probability ofa "false-positive" is the probability of responding "different" when the members of the stimulus pair were atually the same [P(D/S)]. Conditional probabilities for "miss" responses [P(S/D)] and "orret-rejetion" responses [P(S/S)] are simply 1 - P(D/D) and 1 - P(D/S), respetively. The matries for eah animal were based on approximately 120 trials (six bloks) for eah stimulus pair. Only data from the first two sessions for eah animal were used so we ould assess performane in the absene of training; studies on adults have shown that protrated training with feedbak an produe eiling effets that obsure potential peaks in disriminability (e.g., Carney, Widen, & Viemeister, 1977). A number of analyses were onduted using these stimulus-response matries. The simplest was a perentorret measure, alulated by adding the probabilities ofhits and orret rejetions, dividing by two, and multiplying by 100. Thus, a sore of 50% orret represents hane. This measure takes into aount the animals' responses on both S and D trials. The mean perentorret sores are plotted in Figure 1. As Figure 1 shows, the animals performed better, for all three test series, on the between-ategory pair than on either of the two within-ategory pairs. This group trend was shown for eah ofthe three individual animals. In no instane did an animal perform better on the within-ategory ontrasts than on the between-ategory ontrast for a given test series. A three-way ANOVA examining the main effets of stimulus ontrast (within vs. between), plae of artiulation (bilabial, alveolar, or velar), and trial blok (1-6) revealed signifiant effets for both stimulus ontrast [F(1,2) = 35.5, p <.03] and plae of artiulation [F(2,4) = 10.8, p <.02] but not for trial blok [F(5,1O) = 1.5, p <.25]. Neither the two-way interations nor the three-way interation approahed signifiane (p <.20 in all instanes). Examination of the perent-orret sores also revealed that performane on the within-ategory pairs tended to differ, with performane on the withinvoieless pair exeeding performane on the withinvoied pair for the bilabial and alveolar ontrasts. However, this effet was shown to be due to a hange in reo sponse bias rather than to a true differene in disrirninability for within-ategory pairs. Reall that the animals were trained to disriminate S (AAAA) trials from D (AABB) trials and that the higher VOT value in a given pair served as the B stimulus.! Beause of this, all animals tended to release the key more frequently in response to pairs ontaining a stimulus with a higher VOT value, even on an S trial. This ould be seen in a variety of results: first, the animals tended to produe a greater number of "hit" responses to pairs of stimuli with high VOT values, but also greater numbers of "false-positive" responses. Seond, animals tended to produe greater numbers of "early-release" responses (releasing before the end of the first two stimulus presentations and therefore before the atual start of the trial) when the A stimulus had a higher VOT value. To separate potential effets of response bias from those assoiated with true hanges in disriminability, two sets of disriminability/response-bias measures were alulated using the data from the 2 X 2 stimulusresponse matries. The two measures ofdisriminability were the d' parameter of signal-detetion theory (Green & Swets, 1966), whih assumes normal distributions and equal variane, and -In 1], a distribution-free index of disriminability desribed by Lue (1963). The two measures of response bias were the {3 of signal-detetion theory (Green & Swets, 1966) and In b, a distributionfree index ofresponse bias. BILABIAL ALVEOLAR VELAR Q 100 Q 100 Q 100 OJ ;;; OJ E 80 E 80 "5 ~ 80 <J> <J> s u ~ 40 (; '? 20 '" Q;, 0 o w m ~ W W 00 ro 00 VOT (ms) ~ u ~ 40 (; '-? 20 Q> ~ Q> l>. 0 o ~ m m W W 00 ro 00 VOT (ms) o w m ~ W W 00 ro 00 VOT (ms) Figure 1. Average perent-orret sores-[(probability of hit + orret-rejetion responses)/2 X loo]-for the nine stimulus pairs tested in the experiment. The hathed area shows the range of performane obtained for eah pair.

5 546 KUHL AND PADDEN The disriminability index -In 17 is desribed by the formula -In 17 = ~In[P(D/D)P(S/S)/p(S/D)p(D/S)]. Its value is zero at hane and inreaseswith the auray of performane. Figure 2 plots -In 17 as a funtion of the stimulus pair. For eah plae of artiulation, the -In 17 index is greater for the between-ategory pair than it is for either of the within-ategory pairs, indiating greater sensitivity. The d' analysis revealed an idential pattern of results. This inrease in disriminability for between-ategory pairs is similar to that shown by Wood (1976), who tested human listeners on pairs of stimuli differing by 20 mse on a bilabial VOT ontinuum and reported his data in terms ofthe -In 17 index of disriminability. Wood's data showed omparable within-ategory disriminability, but slightly greater between-ategory disriminability, when ompared with the data obtained here. The response-bias parameter Inb is desribed by the formula Inb =~ln [P(S/S)P(S/D)/p(D/S)P(D/D)]. When there is no response bias, Inb is equal to zero; it beomes inreasingly positive with inreasing bias toward S responses (holding the key) and inreasingly negative with inreasing bias toward D responses (releasing the key). The Inb index for eah stimulus pair is provided in Figure 3. The measure of response bias (fj) of signal-detetion theory produed a pattern of similar results. The data indiate that the animals demonstrated a general tendeny toward S responses, regard. less of the pair being tested. While "hit" and "orretrejetion" responses were equally reinfored, this ten deny toward holding the key was probably due to the fat that only three of the nine pairs were easily disriminable, plus the fat that half of all trials presented were S trials, whih required a holding response. The density of reinforement, therefore, was atually greater for holding responses than for lifting responses. This would tend to ause animals to refrain from lifting the response key unless they were quite sure that the members ofthe stimulus pair were different. In addition to the overall tendeny toward "same" responses, the animals demonstrated a systemati hange in response bias with inreasing VOT. Sine lifting the key was assoiated with reinforement more frequently in the presene of signals with higher VOT values, animals tended to release the key more frequently when the stimulus pair ontained a stimulus with a higher VOT value. Reall, however, that only for the betweenategory pairs was this tendeny assoiated with greater disriminability. The improved disriminability for within-voieless pairs seen in the perent-orret measure an therefore be attributed to response bias rather than to a true inrease in disriminability. In ontrast, the peak in disriminability for pairs straddling the boundary represents a true inrease in disriminability. These response-bias data an be ompared with those obtained on human listeners by Wood (1976), who also used the Inb index. He found a signifiant shift toward "same" responses for within-ategory ontrasts and a signifiant shift toward "different" responses for -CO +-' Q) C I --- : CO E... o en o o -.5 Bilabial Alveolar Velar ~ WithIn Between WIthin Ibl Ib-pl Ipl Wlthm Between Wlth,n 101 lo-ti Itl Within Between Within Igl Ig-kl Ikl Figure 2. Average data for a distribution-free index of disriminability (-In TI) for the nine stimulus pairs. Higher numbers indiate greater sensitivity (see text for additional details).

6 ANIMAL DISCRIMINATION OF ACOUSTIC CUES FOR VOICING C1:l.- Q).0 -- (f) C1:l : a Q) (f) -1 o 0- (f) -2 Q) a: -3 Bilabial Alveolar Velar Wlth,n Between Within Ibl Ib-pl Ipl Wlth,n Between Within Idl Id-II III Stimulus pair Within Between Within Igl Ig-kl Ikl Figure 3. Average data for a distribution-free index of response bias (In b) for the nine stimulus pairs. Positive numbers indiate a bias toward "same" responses (holding the key), and negative numbers indiate a bias toward "different" responses (releasing the key). between-ategory ontrasts. We obtained a similar shift in response bias from "same" responses to "different" responses when omparing the voied within-ategory pairs with the between-ategory pairs (Figure 3), but not for the voieless within-ategory pairs. We attribute this differene to our speifi trial struture.' DISCUSSION In the present study, we trained monkeys to respond on a same-different task and then tested them with pairs of stimuli from a physial ontinuum that ranged pereptually from voied to voieless sounds. The pairs of stimuli were hosen suh that they were separated by an equal physial differene in VOT on eah of the three ontinua tested and suh that some were pereived to be phonetially idential by adult human listeners while others were pereived to be phonetially different. Measures of disriminability demonstrated that monkeys disriminated sounds that were phonetially different (i.e., straddled the phoneti boundary) signifiantly better than they disriminated sounds that were phonetially idential (i.e., fell on one side of the boundary). This was true for all three speeh ontinua studied. The fat that animal listeners demonstrate relatively good disriminability at the boundaries between phoneti ategories and relatively poor disriminability within ategories, just as human adults and infants do, demonstrates that the phoneme-boundary effet is not exlusive to human listeners. The data raise two important theoretial issues: (1) the relevane of animal data to the interpretation of human data, both adult and infant, and (2) the role played by auditory onstraints in the evolution of language. Regarding the first issue, the relevane of animal data to interpretations of human adult and infant data, we argue that systemati omparisons among adult, infant, and animal studies will aid in developing strong theories onerning the nature and origins of the mehanisms underlying phoneti pereption. Comparisons between human adults and infants demonstrate the degree to whih the infant demonstrates an initial apaity to partition an aousti ontinuum in a phonetially appropriate way. Comparisons between humans and animals suggest the degree to whih effets should be attributed to general auditory pereptual mehanisms rather than to mehanisms evolved speifially for proessing speeh information. The issue of whether phoneti pereption involves mehanisms that are speeh-speifi will not be resolved with a single omparison. And, given that the initial omparisons between humans and animals have revealed many striking similarities (Kuhl, 1981; Kuhl & Miller, 1975, 1978; Morse & Snowdon, 1975; Waters & Wilson, 1976), as well as some differenes (Sinnott et al., 1976), the answer to the speeh-speifiity issue will not be a simple yes or no. Rather, it will be a determination of the level at whih speial mehanisms must be invoked to aount for the data. The omparisons of interest form. a hierarhy. To date we have examined whether animals tested in labeling tasks pereptually partition speeh ontinua at the phoneti boundaries and whether any peaks in disriminability are onsistent with the loa-

7 548 KUHL AND PADDEN tions of phoneti boundaries. The results of these studies on the voied-voieless distintion onfirmed the existene of appropriate boundaries for the hinhilla (Kuhl & Miller, 1978) and the monkey (Waters & Wilson, 1976), and showed that hinhillas demonstrate differential disriminability for stimuli along a Ida-tal ontinuum, with best performane shown at the loation of the phoneti boundary (Kuhl, 1981). The present data extend the fmding of differential disriminability to all three voied-voieless ontinua in a nonhuman primate. Given that these initial omparisons show similarities between the human and animal data, then omparative tests involving more omplex examples provided by the adult literature an be made. In partiular, future studies should examine the role of partiular aousti ues for voiing that have been shown to affet the loation of the phoneti boundary in adults. An example is the determination of the effet of the first-formant transition on the boundary loation in maaques, sine studies have shown that the boundary systematially shifts as the frequeny of the first formant is lowered in adult listeners(lisker, 1975; Summerfield& Haggard, 1977). Reently, experiments have been undertaken with infants that address the potential interation between these two kinds of aousti ues (first-formant duration and VOT) for voiing pereption (Miller& Eimas, Note 1). The results suggested that the infant's disrimination of speeh sounds was influened by both aousti ues, as it is for adults. It will now be important to determine if animals show similareffets. The ontinued omparison of adult, infant, and animal data usingthe same stimuli and omparable methods should eventually identify the preise examples for whih mehanisms speifi to speeh must be invoked to aount for the data, and the extent to whih those mehanisms are funtional at birth. Adult experiments that isolate the role of individual aousti ues and speify the extent to whih they govern the boundary loations, alone or in ombination, will be helpful. Until the set of rules for ombining the ues for voiing pereption are determined and the experiments that are defmitive tests for the use of those rules are identified, the most powerful omparative and developmental experiments armot be run. As defmitive examplesare tested, we willknow exatly how far we an push the argument that the adult and animal data are omparable. It is possible that a full aount will suggest that animals use simpler rules for pereptually grouping stimuli, separating them on the basis of an aousti priniple suh as, for the voiing ontrast, the relative timing of two aousti events, whereas human listeners employ a more omplex set of rules. More omplex rules might involve taking into aount the.values of other aousti ues. It is also possible, however, that effets as omplex as the reently observed "trading relations" (Best et al., 1981) derive from general rules about the pereptual grouping of auditory stimuli, and are inherent in the funtional harateristis of the auditory system. Pushed to its limits, this latter aount holds that speeh sounds form "natural lasses." This notion, whih has been developed by Rosh (1973) for ertain visual ategories, has also been modified for appliation to speeh (see Kuhl, in press, and Stevens, 1981, for disussion). The seond major point of this disussion, the role played by auditory onstraints in the evolution of language, is intrinsially tied to the first. That is, our understanding of the role of auditory onstraints in the evolution of language will depend upon what eventually turns out to be ommon, and what divergent, in human and animal. If the data eventually show that animalsuse simpler rules in forming auditory ategories for speeh, sounds while humans use a more intriate ontextdependent set of rules, then we would onlude that the onstraints imposed by the auditory system provided a set of broad guidelines that served to initially struture the aoustis of language but did not solely determine them. These onstraints ould have taken the form of a set of natural psyhophysial boundaries (Kuhl & Miller, 1975) whose inherent harateristis inluded poor disriminability among stimuli falling on one side of suh boundaries but good disriminability for stimuli straddling them. Given that these natural psyhophysial boundaries were determined by the mammalian auditory system, it would have been natural for the aoustis of language to reflet these onstraints (Kuhl, 1979b; Stevens, 1981). But even if one admits to the existene of natural psyhophysial boundaries and their role in the evolution of speehsound ategories, the question of how omplete an explanation this provides for the pereption of speehsound ategories in humans still remains. Sine speeh ategories are represented by diverse aousti events, a omplete aount based solely on auditory onstraints would require one to argue that not only boundaries, but also ategory enters, are determined by the funtional harateristis of the auditory system (see Kuhl, in press, Kuhl & Padden, 1983, and Stevens, 1981, for disussion). In summary, we have shown that animals display the tendeny to partition ontinua in ways that are onduive to the phoneti disrimination of voied and voieless stimuli. This was shown in an identifiation task using stimuli from a voied-voieless ontinuum; the animals behaved as though they pereived an abrupt hange in the quality of the stimulus at preisely the point at whih many languages separate the ategories (Kuhl & Miller, 1975, 1978; Waters & Wilson, 1976). Also, animals demonstrate poor disriminability for withinategory aousti variants and good disriminability for between-ategory aousti variants. This was seen in previous studies (e.g., Kuhl, 1981) and in this experiment. Further studies will be required to determine

8 ANIMAL DISCRIMINATION OF ACOUSTIC CUES FOR VOICING 549 exatly 'how far the analogy extends. Their outomes have important impliations for models of speeh proessing and for understanding the evolution of language. RD'EIlENCE NOTE 1. Miller, J. L., Elmu, P. D. Contextual pereption of vote Ing by III/ants. P.per presented.t the meetina of the Soiety for Researh In Child Development, Boston, April REfERENCES ABRA"SON, A., & LISKER, L. Disriminability alolll the voilna ontinuum: Cross l.nau.ae tests. Proftdlngs ofth~ Sixth International Congress of Phoneti Siene, Pragu~, Prague: Aademl., CARNEY, A. E., WIDIN, G. P., & VIB..BISTJ:R. N. F. Nonategorial pereption of stop onsonants differina in VOT. Journal of the Aoustial Soiety of Ameria, 1977, 62, EILERS, R. E., GAVIN. W., & WILSON, W. R. Linaulsti experiene and phonemi pereption in infany: A ross-linauisti study. Child Development, 1979, 50, EI..AI, P. D. Auditory and Unaulstl proessln, of ues for plae of artiul.tion by Infants. hreptlon 4 Psyhophysis, 1974,16, (.) EI..AI, P. D. Linaulstl proesslna of speeh by youn, infants. In R. Shlefelbush L. Lloyd (Bds.), Languag~ and perspetlva-aquisltion, retardation,and Intervention. Baltimore: University Park Press, (b) EI..AI, P. D. Auditory and phoneti odina of the ues for speeh: Disriminationofthe /r-ii distintion by youna infants. hreption4 hyhophysis, 1975,11, ~ EI..AI, P. D., SIQUBLAND, E. R.,JUICZYK, P., & VIGORITO, J. Speeh pereptionin Infants. Selina, 1971, 171, EI..AI, P. D., & TARTT&R, V. C. On the development of speeh pereption: Meh.nisms and.natoaies.ln H. W. Reese. L. P. Llpsltt (Bds.), Advanes in 1llld development and behavior (Vol. 13).New York: AademiPress, FIIHBR.JORGBNION, E. Aousti analysis of stop onsonants. MisellaneaP1IoMtla, 1954,2, GREEN, D. M., & SWBTI, J. A. Signal detetion t1leory and psyhophysis. New York: Wiley,1966. JUICIYK, P. W. Infant speeh pereption: A ritial.ppral... In P. D. Elmu J. L. Miller (Eds.), hl'8petiws on tm studyofspeeh. Hillsdale, N.J: Erlb.um, KUHL, P. K. Predispositions for the pereption of speeh-sound ateaorles: A speies speifi phenomenon? In F. D. Minifle L. L. Lloyd (Eds.), Communlati~and ogniti~ abilitid Early be1lavioral assessment. B.ltlmore: University P.rk Press, KUHL. P. K. Speeh pereption In early Infany: Pereptual onstany for spetrally diulmijar vowel ateaorles. Jouf1Ul1 oftm Aoustitll Sollty ofam~ritl, 1979, 'I, (.) KUHL, P. K. Models and mehanlstns In speeh pereption: Speies omparisons provide further ontributions. B",in, JH. haviorandevolution, 1979, 16, (b) KUHL, P. K. Disrimination of speeh by nonhuman animals: B.si.udltory sensitivities ondulve to the pereption of speeh sound ateaorles. Jouf1Ul1 of tm Aoustitll SoWty of Ammtl, 1981, 71, 340-~9. KUHL, P. K. Cateaorlzation of speeh by infants. In J. Mehler R. Fox (Bds.), Neonate ognition: Beyond tm blooming buuingo'lfuslon. Hilladale, N.J: Erlt.um, In press. KUHL, P. K., & MILLER, J. D. Speeh pereption by the hinhlll.: Voied-voieless distintion In alveolar ploslve eonson.nts. Selina, 1975, 1,., KUHL, P. K MILLEa, J. D. Speeh pereption by the hinhilla: Identifiation funtions for syntheti VOT stimuli. Journal oftmaoustitllsoewty ofameritl, 1978,63, KUHL, P, K., PADDU, D. M. Enhaned disrimlnabillty at the phoneti boundaries for the plae feature in maaques. Journal ofthe AoustltllSollty ofamerltl, 1983, in press. LASKY, R. E. SYRDAL-LAIKY, A., & KLEIN, R. E. VOT disrimmation by four to six and a half month old infants from Spanish environments. Jouf1Ul1 of.experimental Psy1lology, 1975, 20, LIBEa..AN, A. M.,DELLATRE, P. C., &CooPEa, F. S. Some ues for the distintion between voied and voieless stops in initial position. LaniuagundSpeeh, 1'958, 1, LIIKICR, L. Is It VOT or a firat-formant transition detetor? Journal of the Aoustial Soit!ty. of Ameria, 1975, 57, LIIKBR, L., & ABRA"SON, A. S. A ross-ianauaae study of voleina in initial stops: Aoustial meuurements. Word, 1964, 20, LuCI:, R. D. Detetion and reoanition. In R. D. Lue, R. R. Bush, E. Oalanter (Bds.), Handbook ofmatmmlltitll Psyhology. New York: Wlley, MILLaR, J. D. WI.a, C. C., PAITOU, R. E., KELLY, W. J., DooLING, R. J. DIsrlmlnation and labellna of noise-buzz sequenes with vary1na noise-lead times: An example of ate,orl.1 pereption. Journal of the Aoustial Soiety of Ameritl, , MouB, P. A. Infant speeh pereption: A preliminary model and review of the literature. In R. Shiefelbush L. Lloyd (Bds.). Langullge pel'8petiva-aquisition, retllrdation, lind intervention. Baltimore: University ParkPress, Moua, P. A., " SNOWDON, C. T. An Investiptlon of ate,orlal speeh disrimlnation by rhesus monkeys. Pereption 4 Psyhophysis, 1975,17,9-16. PIIONI, D. B. Identifiation and disrimlnatlon of the relative onset time of two omponent tones: Impliations for voilna pereption in stops. Journal of the AoUltial Soiety of AmerlCtl, 1977,61,13' RBPP, B. H. Relative amplitude of upiratlon noise u a vollna ue for syllable-initial stop onsonants. Language and Spee1l, 1979,22, RoaH. E. Natural ateaorles. Cogniti~ pqltology, 1973, 4, SINNOT, J. M., BCH.R, M. D., MOODY. D. B. " hrrinl, W. C. Speeh sound disrimination by monkeys and humans. Journal of the Aoustial Soiety of Ameria, 1976, 60, SnVl:N8. K. N. The quantai nature of speeh: Evidene from' artiulatory-aousti data. In E. E. David, Jr., P. B. Denes (Eds.), Hu"",,, ommuniation: A unlfjed v~w. New York: MCOraw-Hill, SnVl:N8, K. N. Constraints Imposed by the auditory system on the properties used to lualfy speeh sounds: Evidene from phonololy, aoustis, and JlIYCh~s.In T. Myers, J. Laver, a: J. Anderson (Eds.), Advanes in pq1lology: T1w o,nit/~,..,..ntation ofspe«ll (Vol. 7). Atnsterdam: North Hoiland, SnVl:N8, K. N. " KLATT, D. H. Role of formant transitions In the voied-voieless distintlon for stopa. Jouf1Ul1 of tm AoustialSowty ofamem, 1974,55, STu.Tl:R, L. Lanauqe pereption of two-month-old infants shows effets of both innate mehanlstns and experiene. Nature, 1976, Ut, SU...EU'I.LD. Q.. HAGGARD. M. On the dissoiation of spetral and temporal ues to the vollna distintion In Initial stop onsonants. Journal of the AoUltitll Soiety of Amerltl, 1977,

9 550 KUHLANDPADDEN WATERS, R. S., & WILSON, W. A., JR. Speeh pereption by. rhesus monkeys: The voiing distintion in synthesized labial and velar stop onsonants. Pereption II Psyhophysis, 1976, 19, WOOD, C. C. Dlsriminability, response blu, and phoneme ateliories In disrimination of voie onset time. Journal of the Aoustial Soiety ofameria, 1976,60, NOTE 1. In typial same-different formats with human listeners, S trials onsist of both AA and BB trials, and D trials onsist of both AB and BA trials. We have not been able to train our animals to do the latter kind of task with more than a single stimulus pair, and sine the design involved the olletion of data from eah animal on all nine stimulus pairs (i.e., repeated measures), we hose the restrited format desribed above, in whih S trials onsist of AA pairs and D trials onsist of AB pairs. The B stimulus in any given pair was the stimulus with the higher VOT value. This format tended to inrease the animal's response bias, but the analysis proedure allowed the separation of response bias and disrirninability. (Manusript reeived Marh 25, 1982; revision aepted for publiation September 23, 1982.)