UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Adult and infant perception of an English phonetic distinction Pegg, Judith E. 1995

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-ubc_1995-060403.pdf [ 2.15MB ]
JSON: 831-1.0088348.json
JSON-LD: 831-1.0088348-ld.json
RDF/XML (Pretty): 831-1.0088348-rdf.xml
RDF/JSON: 831-1.0088348-rdf.json
Turtle: 831-1.0088348-turtle.txt
N-Triples: 831-1.0088348-rdf-ntriples.txt
Original Record: 831-1.0088348-source.json
Full Text

Full Text

ADULT AND INFANT PERCEPTION OF AN ENGLISH PHONETIC DISTINCTIONByJUDITH E. PEGGB.A., The University of British Columbia, 1986M.A., The University of British Columbia, 1989A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHYinTHE FACULTY OF GRADUATE STUDIESDepartment of PsychologyWe accept this thesis as conformingthe required standardTHE UNIVERSITY OF BRITISH COLUMBIASeptember 1995© Judith B. Pegg, 1995In presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.Department of 59f4L 4ô4’/The University of British ColumbiaVancouver, CanadaDate(Signature)DE-6 (2188)itAbstractPrevious research has revealed that very young infants discriminate most speechcontrasts with which they are presented whether the contrasts are native or non-native whileadults have difficulty discriminating non-native speech contrasts but easily discriminatethose contrasts holding meaningful (phonemic) status in their native language. Severalstudies have shown that this reorganization in phonetic perception from language-generalperception to language-specific perception occurs at about 10 to 12 months: infants this ageattend only to native phonemic contrasts. It is of interest to determine if exposure to aphonetic variant plays an important role in influencing perception. We know from previousresearch that absence of exposure does not always lead to a lack of discrimination. Thisthesis was designed to determine if exposure per se maintains discriminability. To this endEnglish-speaking adults and infants were tested using a phonetic distinction that does nothold phonemic status in English but does occur in English. This distinction involves thephonetic variants [da] and the stop produced following Is! transcribed as [ta].When tested in an identification procedure, English-speaking adults identify both[da] and (s)[ta] as members of one English phonemic category (i.e. [da]). When tested in adiscrimination procedure and a category change procedure, adults discriminate (s) [ta] from[da] (albeit not as well as would be expected for a native phonemic contrast). With respectto infants, 6- to 8-month-olds discriminate this distinction revealing further support forbroad-based phonetic perception at this age. However, 10- to 12-month-old infants do notdiscriminate, suggesting that the native phonemic status of the contrast (but not exposure)is the important factor in the reorganization. Discussion centers around how these resultsadd to the existing literature and why infants of 10- to 12-months would fail to discriminatea native phonetic distinction.11)TABLE OF CONTENTSABSTRACT iiTABLE OF CONTENTS iiiLIST OF TABLES vLIST OF FIGURES viACKNOWLEDGMENTS viiGENERAL INTRODUCTION 1History of Research Investigating the Influence of Experience on Speech PerceptionEvidence with respect to consonants 4Evidence with respect to vowels 10Approaches and the Need for Further Empirical Research 16Perceptual Magnet Hypothesis 18Assimilation Hypothesis 19The Four-Factor Model 21The Present Research 25EXPERIMENT 1The Category Goodness Task 26Method 31Subjects 31Stimuli 31Apparatus and Procedure 33Results 35Individual exemplars 36Discussion 38EXPERIMENT 2The AX Discrimination Procedure 40Method 41Subjects 41Stimuli 42Apparatus and Procedure 42Results 43Discussion 47ivEXPERiMENT 3The Category Change Procedure with Adults 50Method 51Subjects 51Stimuli, Apparatus, and Procedure 51Results 53Discussion 56EXPERIMENT 4Conditioned Head-Turn Procedure with Infants 59Method 61Subjects 61Stimuli 62Apparatus and Procedure 62Results 66Discussion 69GENERAL DISCUSSION 71REFERENCES 84APPENDIX 1 91APPENDIX 2 94VLIST OF TABLESTABLE 1.Acoustical analyses for each of the 15 exemplars 32TABLE 2Experimental design (Study 1) 34TABLE 3Experimental design (Study 2) 43TABLE 4Experimental design (Study 4) 65viLIST OF FIGURESFIGURE 1Graphic representation of Kuhi’ s perceptual magnet effect 11FIGURE 2Graphic representation of Best et al.’ s assimilation model 20FIGURE 3Main effect for page heading 35FIGURE 4Ratings for exemplars when “sda” and “da” (D) are page headings. . .37FIGURE 5A-Prime values for English listeners tested in an AX procedure 44FIGURE 6Interaction between pair type and order of pairings 47FIGURE 7A-Prime means for adults in the category change procedure 54FIGURE 8Percent of SAME responses for (s)[taj versus [daJ referents 56FIGURE 9Proportion of adults and infants reaching criterion 67FIGURE 10A-Primes for adults, young infants and old infants 68vi,ACKNOWLEDGMENTSI would like to thank all the adults who participated in this research and the parentswho brought their babies to our lab. Without the cooperation of willing parents, thisproject would not have been possible. As well, I want to thank Diane Hobday forrecruiting parents and generally helping to make the lab a friendly place for parents to visitand a wonderful place for us to work . Many thanks to Tracey Smilie and CarmenSwanson for the assistance in collecting data.I would also like to thank my thesis committee: Richard Tees for his erudite adviceand support throughout graduate school and his calming words prior to and during thevarious examinations; Lawrence Ward for his inquiring critique of this endeavour and mylogic; and my advisor, Janet Werker, for her academic support during the process ofdesigning, conducting and writing of this thesis as well as for support during the entireprocess of graduate studies. Janet not only provided an environment conducive toachieving academic excellence, but also taught me invaluable lessons about graciousness,kindness, motherhood, and general friendship.Last but not least, to my family of origin, I give thanks for the vote of confidencethat I would complete my formal education and do well. I’m sorry that my mother andfather cannot be here to share in the finished product. To my family of procreation, Iapologize for not always being there when you needed or wanted me. And I thank you allfor understanding how important this goal was to me and for helping to make it apossibility.This research was supported in part by a doctoral fellowship from the NationalResearch Council of Canada and in part from a Graduate Student Fellowship from theMedical Research Council of Canada.1GENERAL INTRODUCTIONIn the process of becoming a speaker of a language, one must acquire a tacitunderstanding of the phonological principles1 inherent in the native language.Phonological principles specify among other things, variations in speech sounds that signaldifferences in meaning (phonemes) and differences in productions that occur within alanguage but do not specify meaningful differences (phones and allophones). Duringseveral years of research, adult and infant phonetic perception has been examined todetermine how understanding of phonological principles develops. It is now wellestablished that during the first year of life, phonetic perception is affected by the linguisticenvironment (Best, LaFleur, McRoberts, & Silver-Isenstadt, in press; Kuhi, Williams,Lacerda, Stevens, & Lindblom, 1992; Polka & Werker, 1994; Werker & Tees, 1984a;Werker, 1989). For example, whereas 6- to 8-month-old infants discriminate non-nativeand native consonant contrasts with equal ease, 10- to 12-month-old infants and adults arebest at discriminating only those consonant contrasts that hold meaningful or phonemicstatus2 in their native language (Werker & Tees, 1984a). Such evidence suggests that thephonological system begins to develop in the first year of life.Research also indicates, however, that the changes in speech perception across thefirst year of life do not involve a permanent loss in auditory abilities. In some testingsituations adults attend to phonetic differences. For example, under some conditions,adults discriminate non-native contrasts (e.g., Werker & Logan, 1985) and can evenimprove their ability to discriminate if given training (e.g., Pisoni, Logan, & Lively, 1994;Tees & Werker, 1984). Furthermore, adults can and do use some phonetic detail (non-phonemic) to disambiguate utterances (Dutton, 1992). Finally, discrimination of some1 The phonology of a language refers to rule-governed patterning at the level of sound (See Appendix 1 formore complete definitions).2 Meaningful or phonemic contrasts are those which differentiate meaning in a minimal pair. An exampleof a phonemic contrast in English is [dJ versus [t] because this contrast signals a difference in meaningbetween the minimal pair “dot” versus “tot’.2non-native contrasts remains high if those contrasts are not assimilated to native phonemiccategories (Best, McRoberts, & Sithole, 1988). Thus, the change in speech perceptionabilities is more appropriately termed a developmental reorganization in speech perception.The present research was designed to further extend our understanding of thefactors responsible for the developmental reorganization in phonetic perception, inparticular, to test the hypothesis that by 10 to 12 months of age infants discriminate onlythose contrasts that hold phonemic status in their native language. Experiments comparingdiscrimination of native versus non-native contrasts have, to date, typically defined a non-native contrast as a pair of consonants or vowels that hold phonemic status in a non-nativelanguage but not in the native language. However, exposure to the members of the pair hasnot been controlled in most experiments (but see Polka, 1991). That is, listeners may havebeen linguistically exposed to neither, one, or both members of the pair in the non-nativecontrast. In some cases, neither member of the non-native contrast occurs in the nativelanguage (e.g., a Zulu click contrast, see below). In other cases, one of the members ofthe non-native contrast may be a phoneme in the native language (or at least very similar toa native phoneme) and the other member may never occur in the native language. In stillother cases, both members may occur in the native language, one as a phoneme and theother as an allophone (see Appendix 1) or both may be allophones3. Thus, in mostprevious studies there is a confound between the role of exposure and the role of non-native status. This makes it impossible to determine if the reorganization in phoneticperception is due to the phonemic status of a contrast or the simple lack of linguisticexposure.One way to test the possibility that simple lack of exposure leads to a change innon-native phonemic perception is to present listeners with a speech contrast that they neverhear. Best et al. (1988) conducted such a study by presenting English speaking4adults andOf course, the final possibility is that both members occur as phonemes in the native language. Thiswould mean that the pair is a native contrast.Hearafter, “English adults” refers to English-speaking adults.310- to 12-month-old infants with a Zulu click contrast: a contrast which does not occur inEnglish and which is not similar to any English phone. English adults and 10- to 12-month-old infants easily discriminated this non-native contrast. This shows that lack oflinguistic exposure does not necessarily lead to reduced discrimination.Although Best et al.’s research (1988) reveals that lack of linguistic exposure maynot always lead to a reduced ability to discriminate, we do not yet know if linguisticexposure per se is sufficient to maintain an ability to discriminate independent of phonemicstatus. To conduct such a study, it is necessary to present listeners with a pair of phones towhich they are routinely and regularly exposed (i.e., native) but one that does not holdphonemic status in their native language. One distinction which fits these criteria is thecontext-dependent allophone. Context-dependent allophones are phonetic (non-phonemic)variants that occur in complementary distribution. In other words, the phonetic alternationof the phoneme occurs in a different and specific context. This kind of allophone isspecified by the phonology of the language.One English context-dependent allophonic distinction is the alternation of thephoneme /tJ: namely [th] which occurs in initial position in a word or syllable and [t] whichoccurs following [sj. The English phonological rule states that alveolar stops5 in initialposition are either a voiceless aspirated [ti’] or a voiced unaspirated [dJ. Following [s],only one alveolar stop is allowed and this is a voiceless unaspirated [t]6.The purpose of this research is to examine adult and infant perception of thelanguage-specific phonetic distinction, [d] versus [t]7. By using this distinction, linguisticexposure is maintained because English uses both of these phones but the pairing is notused to contrast meaning. Thus, we can address the question of whether linguisticexposure lacking in phonemic status is sufficient to allow maintenance of discrimination.Alveolar stops are consonants produced by putting the tongue against the alveolar ridge on the roof of themouth.6 The same rule applies to all stops and thus includes the phonemes fb, p/ and Ig, k/.The reason for selecting this pair and not the allophonic distinction [t] versus [th] is explained in detail atthe end of the introduction.4There are several additional theoretical reasons why it is of interest to assess infantand adult perception of this phonetic distinction. To understand these issues it is necessaryfirst to review evidence of the effect of experience on speech perception for bothconsonants and vowels, and second, to summarize what we know and what remains to berevealed, and third, to briefly discuss how assessing adult and infant perception of thisallophonic distinction will help to formulate theory.History of Research Investigating the Influence ofExperience on Speech PerceptionEvidence With Respect to ConsonantsUnderstanding the effect of experience on phonetic perception began with earlystudies investigating adult perception of consonants. Lisker and Abramson (1970)assessed adult perception of a synthetic continuum of speech sounds which varied in voiceonset time (VOT8)for three stop categories; labial, apical9 and velar. Tests of English,Thai, and Spanish speaking adults revealed that category boundaries vary as a function ofthe phonology of the native language. There are only two voicing categories in Spanishand English and both Spanish and English speakers identified only two categories in eachof the continua (ip, hi, Id, t/, and /g, ki). Importantly, adults from each language grouphad their boundaries in different locations. In accordance with Thai phonology, however,Thai speakers identified three categories for the labial and apical continua (/b, p, hi and Id,t, thi) and two for the velar continuum (1k, kh/). These results show that the linguisticenvironment has an effect on adult speech perception such that categorical boundaries, aswell as the number of categories perceived, differ as a function of the phonology of the8 VOT is defined as the timing of the onset of voicing relative to the burst, or release of the closure, instop consonants (Lisker & Abranison, 1964). However, several other acoustic and phonetic cues covarywith VOT (for details see Lisker & Abramson, 1967; McKain & Stern, 1982.)9 The term “apical’ includes both alveolar and dental place of articulations. It refers to phones producedby raising the tip of the tongue to the central or frontal region of the oral cavity. Since contrasts evaluatedby Lisker and Abramson (1970) include both dental and alveolar, apical is a more accurate term.5listener’s language’°. In addition, this research demonstrates that VOT is a sufficient cuefor native listeners to differentiate between voicing distinctions that are phonemic in theirnative language.Several experiments extended this research by using contrasts involving place andmanner distinctions. These studies confirm that adults discriminate those contrasts whichhold phonemic status in their native language but, under most circumstances, havedifficulty discriminating non-native phonemes (e.g., Abramson & Lisker, 1970; McKain,Best, & Strange, 1981; Miyawaki, Strange, Verbrugge, Liberman, Jenkins, & Fujimura,1975; Polka, 1991; Trehub, 1976; Werker, Gilbert, Humphries, & Tees, 1981; Werker& Tees, 1 984b). For example, without training Japanese adults fail to discriminate theEnglish apical, lateral distinction, Ir, 1/ (Miyawaki et al., 1975) whereas English adultsshow a categorical boundary effect for this contrast. On the other hand, English adultshave difficulty discriminating the Hindi retroflex versus dental contrast, IT! versus It!,whereas Hindi adults easily discriminate it (Werker et al., 1981).Eimas, Siqueland, Jusczyk, and Vigorito (1971) were the first to extend theresearch with VOT continua to infants. They tested 4-month-old English-learning infants11in a high-amplitude sucking paradigm (HAS) using a synthesized VOT continuum ofbilabial stop consonants. Infants showed an increase in sucking rate when the stops werefrom either side of the English phonemic boundary (e.g., +20 ms /baJ versus +40 ms /pal)but not when the stops were from the same phonetic category (either -20 ms and 0 ms or+60 ms and +80 ms). This research was the first to reveal that infants perceive speech in alinguistically relevant manner.Several subsequent studies confirmed that very young infants can discriminate non-native phonetic contrasts that differ in VOT and many studies revealed a similar pattern for10 Categorical perception is revealed when identification of speech sounds (labelling) predictsdiscrimination. It can be argued that a category boundary effect is a consequence of categorical perception.The boundary effect is defined as enhanced discrimination between exemplars that cross a category boundary(between category) over those exemplars from within the same category even though the exemplars differ inequal steps along the continuum.11 Hereafter, “English infants” refers to English-learning infants.6place-of-articulation (e.g., Aslin, Pisoni, Hennessy, & Perey, 1981; Best et a!., 1988;Eimas, 1975; Streeter, 1976; Werker et al., 1981; Werker & Tees, 1984a; Werker &Lalonde, 1988). These remarkable abilities extend to contrasts that young infants have notheard before. For example, in an early study, Laskey, Syrdal-Laskey, and Klein (1975)reported that 4- to 6.5-month-old infants can discriminate between exemplars from differentVOT categories even if they are not phonemic in their ambient language but are in otherlanguages. Guatemalan infants discriminate the Guatemalan prevoiced (-60 ms) from thevoiced (-20 ms) phones and discriminate the English voiced (+20 ms) from the voiceless(+60 ms) phones, but do not discriminate a contrast that is not phonemic in any language—the two voiced phones (-20 and +20 ms).Trehub (1976) conducted the first study to assess perception of a phonetic contrastin both infants and adults. Five- to 17-week-old infants from an English-speakingenvironment were tested in the HAS procedure with two naturally produced contrasts: aFrench contrast (IpaJ versus /päl) and a Czech contrast (IzaI versus i$’aJ). The resultsrevealed that infants discriminate both of these non-native contrasts. English adults werepresented with the same Czech contrast and an English contrast, In! versus hi!, but theprocedure differed making direct comparison between adults and infants impossible. Theadult task was to decide if during a presentation of 10 speech sounds, the stimuluschanged. Although English adults correctly identified change and control trials for theEnglish contrast, they failed to do so for the Czech contrast thus evidencing only nativelanguage phonemic perception.Taken together, the results from these early studies led to the suggestion that veryyoung infants are endowed with a broad-based perceptual ability that allows them todiscriminate between most phonemes used to contrast meaning in the world’s languages’2.On the other hand, linguistic experience serves to attenuate adults’ ability to discriminate12 Although there is some evidence that young infants may fail to discriminate some contrasts (e.g.,Eilers, Gavin, & OIler, 1982; Eilers & Minifie, 1975), the results are difficult to interpret due tomethodological problems. As well, other researchers have failed to replicate these findings (e.g., Aslin,Pisoni, Hennessy & Perey, 1981).7non-native phonemes. What these studies failed to identify is the age at which linguisticexposure has an effect on discrimination of non-native phonemes.An investigation of when the language environment begins to have an effect oninfant speech perception came with the research of Werker, Tees, and colleagues. Toexplore this question, Werker, Gilbert, Humphries, and Tees (1981) initially sought toreplicate previous evidence of young infants’ discrimination and adults’ lack ofdiscrimination of non-native contrasts. Unlike Trehub (1976), however, Werker andcolleagues used an identical procedure for both infants and adults. Both infants and adultswere tested in the Conditioned Head Turn Procedure (e.g., Kuhl, 1985). In thisprocedure, infants are rewarded for a correct head-turn (turning when the stimuluschanges) by computer activated stuffed animals and by verbal praise from an experimenterseated in the presentation room with the infant. Adults tested in this procedure press abutton or raise their hand when they hear a change in the stimulus and are informed of acorrect response during change trials by a flashing light. Because the procedure isfunctionally identical for both adults and infants, it is possible to directly compare resultsfrom the two groups.Werker et al. (1981) tested English and Hindi speaking adults as well as 6- to 8-month-old English infants using speech contrasts that are phonemic in Hindi but not inEnglish. The consonants were in initial position and differed either on place-of-articulation(a retroflex /TaJ versus a dental /ta/) or voicing (an unaspirated voiceless /tha/ versus abreathy voiced /dhal).The results from this experiment revealed that both 6- to 8-month-old Englishinfants and Hindi speaking adults discriminate between these speech contrasts. On theother hand, English speaking adults do not reach criterion when tested with exemplars fromeither of the non-native (Hindi) contrasts although they easily discriminate betweenexemplars from a native contrast, Ida! versus Thai. Thus these studies providedconfirmatory evidence of broad-based phonetic perception in young infants and of8phonemic perception in adults. This finding was replicated using a glottalized velar /ki/versus a glottalized uvular 1q11, a contrast that is phonemic in Nthlakampx (a BritishColumbia Interior Salish language) but not in English (Werker & Tees, 1984a).The question of when the reorganization occurs was addressed in a series of studieswith adults, children, and infants. All three age groups were tested in the ConditionedHead Turn Procedure using contrasts that either are or are not phonemic in their ambientlanguage (native versus non-native contrasts). Children aged 12-, 8-, and 4-years weretested on the Hindi contrasts and the results indicated that the reorganization in speechperception had already occurred even in the youngest children (Werker & Tees, 1983).That is, children’s perception of consonants resembled the language-specific phonemicperception of English adults. Moving down in age, Werker and Tees tested English,Hindi, and Nthlakampx infants on one Hindi contrast and one Nthlakampx contrast.Results indicated that babies as young as 10- to 12-months of age no longer easilydiscriminate non-native contrasts showing that a reorganization in speech perception occursat this early age (Werker & Tees, 1984a). Replications using different contrasts (Werker &Lalonde, 1988) and using a habituation-dishabituation procedure (Best & McRoberts,1989; Best et al., in press) confirmed the results that infants under 10-months of age easilydiscriminate non-native phonemic contrasts but infants over 10-months of age and adultsonly reach criterion on native phonemic contrasts.One possible explanation of why the reorganization in cross-language speechperception occurs at this age is because infants have learned which phones in the nativelanguage are used to contrast meaning. If this is the case, infants should be able to usethese newly formed categories to discriminate between words. Werker, Cohen, and Lloyd(submitted) tested infants in a habituation-dishabituation task in which two phoneticallydissimilar nonsense words were paired with two novel objects during the habituation phaseand the pairings were switched during the dishabituation phase. Female infants 14 monthsof age can learn these phonetically dissimilar pairings and can easily detect the switch in the9word-object pairing. The results differ, however, when the words are phonetically similar.When Stager and Werker (1995) tested infants in the same procedure but used thephonetically similar nonsense words /bi/ vs. Idi/, infants of 14 months no longer learnedthat two different words were paired with two different objects and they did not detect theswitch. This suggests that infants cannot use native language minimal pairs to distinguishmeaning until at least 14 months of age.There are three codicils to this evidence which must be mentioned. First, thedevelopmental reorganization occurring at the end of the first year does not necessarilypreclude further development in categorical perception. Burnham, Earnshaw, and Clark(1991) tested 9- to 11-month-old infants, 2- and 6-year-old children, and adults in amodified Conditioned Head Turn Procedure in which subjects could respond by turning ineither direction13. The results from Burnham et al. (1991) suggest that there is someimprovement in categorization of native speech sounds with age.Second, as mentioned earlier, adults can discriminate non-native phonemes undersome circumstances. When tested with 1500 ms ISI’s, adults discriminate only nativelanguage phonemic contrasts but when tested using ISI’s shorter than 1500 ms, adultsdiscriminate non-native phonemes which reveals evidence of phonetic processing and/oradults discriminate exemplars from the same phonetic category (i.e., two different /dal’s)which reveals evidence of acoustic processing (Jamieson & Morosan, 1986; Pisoni, Aslin,Perey, & Hennessy, 1982; Werker & Logan, 1985). Thus as noted above, adultperceptual biases are not due to a loss in auditory sensory abilities but are better describedas a developmental reorganization in speech perception.Third, adults and 10- to 12-month-old infants may maintain their ability todiscriminate some non-native contrasts. In reminder, Best and colleagues (1988) revealedthat adults and 10- to 12-month-old infants easily discriminate a non-native Zulu click13 Although other researchers have been unable to train infants in this procedure, older children and adultshave been tested using this two-choice procedure yielding reliable and replicable results.10contrast, a contrast to which they are not exposed. Best et al. hypothesize that the extent towhich a phone is discriminated depends on an interaction between the native languagephonology and the characteristics of the non-native phone. In their view, non-nativephonemes are reorganized as a function of how assimilable they are to native languagephonemic categories.14Evidence With Respect To VowelsRecent evidence indicates that the developmental reorganization in phoneticperception extends to vowels, although there are some interesting differences betweenconsonant and vowel perception. Perceptual differences may occur because vowels differfrom consonants in several characteristics. Vowels are uttered more slowly and have alonger steady state portion than consonants. Vowels carry prosodic information, cues tospeaker identity, evidence about the emotional message, and information about linguisticstress. As well, the characteristics of vowels may be exaggerated in speech directedtowards infants (e.g., Fernald & Mazzie, 1991; Papousek, Bornstein, Nuzzo, Papousek,& Symes, 1990; Papousek, Papousek, & Symes, 1991) possibly making vowels moresalient to young infants.Early studies revealed that adult and infant vowel perception differ. Adults showmore continuous (less categorical) perception of vowels than consonants (e.g., Fry,Abramson, Eimas, & Liberman, 1962; Schouten & van Hessen, 1992). A similar patternhas been found for 2-month-old infants when the vowel is relatively long (240 ms). Whenthe vowel duration is shortened, however, infants show better between categorydiscrimination than within (Swoboda, Morse, & Leavitt, 1976).Most early research with infants focused on the structure of infant vowel categoriesaddressing such questions as whether infants ignore speaker differences across samecategory vowel productions (e.g., Kuhl, 1979, 1983). More recent investigations of infant14 This model is outlined in greater detail in a later section (see p. 19).11vowel perception have assessed the perceptual structure within vowel categories (Grieser &Kuhi, 1989; Kuhl, 1991; Kuhi et a!., 1992). This more recent work was designed toassess whether within category discrimination differs as a function of vowel quality.Grieser and Kuhi synthesized a group of speech sounds that fall within the vowelspace of one category as defined by the first and second formants15 (Fl, F2). The centerof the vowel space was determined by collecting goodness ratings from adults. Followingthis, a set of stimuli was constructed with the best exemplar located at the center of the F 1,F2 vowel space while the other exemplars differed from the central vowel in either Fl, F2,or both. Goodness ratings were taken again and adults judged the exemplars that weremost distinct from the best exemplar as less good members of the same vowel category’6.Kuhl and colleagues hypothesized that the most central (the best) exemplar of a vowelcategory might act as a perceptual attractor, thus reducing perceptual space between it and apoorer example of the category. More specifically, Kuhi and colleagues predicted that inthe Conditioned Head Turn Procedure, adults will find it more difficult to discriminate inone order than another: When the central exemplar is the repeating background stimulus(the referent) and the less central exemplar is the change stimulus, discrimination will bemore difficult than the reverse (Figure 1).Figure 1: Graphic representation of Kuhi’s perceptual magnet effect.Discrimination attenuatedBackground TestBest PoorExemplar ExemplarTest I BackgroundDiscrimination enhanced15 Formants are defined as a group of overtones corresponding to a resonating frequency of the air in thevocal tract (Ladefoged, 1982). Simply stated, formants provide acoustic information about a speechsegment.16 But see Sussman & Lauckner-Morano (1995) for evidence suggesting that the vowels used in thesestudies might not be judged by all native speakers as members of the same vowel category.12To test this prediction, adult and infant perception was assessed using theConditioned Head Turn Procedure varying order of presentation of the stimuli. Resultsrevealed that although both adults and infants easily discriminate among vowels from thesame phonemic category, discrimination differs as a function of order of presentation.Specifically, adults detect a change in the vowel stimulus 90% of the time when thebackground is a peripheral exemplar but only 78% of the time when the background is thecentral exemplar. Infants of 6 months do less well overall than adults, but still demonstrate73% accuracy in the first case and 61% in the second. This effect is termed the “perceptualmagnet” effect17.Although this evidence shows that vowels judged as better exemplars can influencediscrimination, the direction of the effect is inconsistent with evidence from other areas ofperception (see also Warren, 1985). For example, in visual perception, horizontal andvertical lines are considered to be “cardinal” lines and are more perceptually salient.Because they are more perceptually salient, they may be considered prototypical lines.However, when presented with a background of vertical lines, subjects can easily detect aline 15 degrees off vertical whereas the reverse is more difficult (e.g., Treisman, 1986).Thus, cardinal lines do not act as perceptual magnets but instead, facilitate discriminability.Similar examples can be found in auditory perception. Trehub, Thorpe, and Trainor(1990) assessed infants’ ability to detect a one-semitone change embedded in Westernmelodies. If “good” melodies can be considered more prototypical than “bad” melodies, aperceptual magnet effect would predict that discrimination of a one note change would beattenuated in the good melody and enhanced in the bad melody. Instead, the evidencesuggests that 7- to 10-month-old infants detect the semitone change only in the context ofthe “good” Western melody again showing a reverse pattern of discrimination to thatreported by Kuhi. Finally, within the realm of speech, Sawusch and Jusczyk (1981)17 This effect was initially termed the “prototype magnet effect” (Kuhl, 1991) but in more recent writingsit has been renamed the “perceptual magnet” effect. This new term avoids the implication that therepresentational model of speech perception involves prototypes. It also more accurately depicts the notionthat perceptual space seems to be affected by order of presentation.13demonstrated that adaptors can have an effect on perception opposite to that of the magneteffect. Specifically, in the second study, /spal led listeners to label an exemplar from themiddle of a fba-pal continuum (an ambiguous exemplar) as IbI while a Thai adaptor ledlisteners to label the same exemplar as a /pI. That is, in this procedure, the adaptors appearto push the ambiguous item to the opposing category, not pull it toward the same category.Thus, evidence of a perceptual magnet effect in vowel perception is indeed interesting.Further evidence provided by Kuhl and colleagues suggests that the linguisticenvironment influences the perceptual structure of vowel categories (Kuhi et al., 1992; butsee Polka & Bonn, 1994; Sussman & Lauckner-Morano, 1995). Six- to 7-month-oldinfants from Swedish and American-English environments were tested with a series ofexemplars surrounding either the English vowel, hI, or the front-rounded Swedish vowel,lyl. American infants showed a perceptual magnet effect when listening to exemplars of theEnglish vowel but not when listening to exemplars of the Swedish vowel. On the otherhand, Swedish infants demonstrated the perceptual magnet effect only for the Swedishvowel (Kuhl et al., 1992). Thus, the evidence suggests that the ambient languageinfluences the internal structure of native language vowel categories by 6-months of age.Polka and Werker (1994; Polka, 1995; Werker & Polka, 1993a) investigated thequestions of whether the ambient language influences between category cross-linguisticvowel perception, and if so at what age. This research differed from Kuhl’ s in that it wasdesigned to investigate perception of two non-native vowel categories rather than theinternal structure of a single vowel category. In a series of studies, adults and infants weretested with two German vowel contrasts both involving a front rounded versus a backrounded distinction. One pair of vowels comprised a tense vowel contrast (back roundedIu:/ versus front rounded Iy:I) and the other a lax vowel contrast (back rounded JUl versusfront rounded /Y/). All vowels were embedded in the same CVC context (d_t).Initially, native English-speaking adults were tested in an AXB discriminationprocedure (Polka, 1995; Werker & Polka, 1993b). English adults easily discriminated14both the tense and the lax vowel contrasts. All the English adults reached native Germanlevels of performance’8on the tense contrast while 2 out of 10 did so for the lax contrast.Thus, although English adults find the lax contrast more difficult than the tense, theydiscriminate both of the German vowel contrasts well above chance.In the second task, both German and English adults were tested in a wordidentification and rating task (Polka, 1995; Werker & Polka, 1993b). German adults wereasked to compare each vowel to 14 different written vowels in the same context (dVt) andto rate the quality of the match. German adults had no difficulty correctly assigning thevowels to their appropriate categories and they rated them all as very good examples of theGerman categories. In the same procedure, English speakers were asked to compareGerman vowels to the vowels in English dVt syllables and rate the quality of the match.All four German vowels were consistently selected by English adults as members of eitherthe English vowel /uJ as in “ooze” or /21 as in “hook”. Ratings scores did reveal somesensitivity to differences among German vowels, however, as the back rounded vowels(both the tense and the lax) were judged as better exemplars than the front rounded vowels.Since English uses only back rounded vowels, it is not surprising that English listenersjudged the back rounded German vowels as better exemplars of English vowels.Polka and Werker then investigated the question of when the linguistic environmentbegins to have an effect on two-category vowel perception. German and English adultsand 10- to 12- and 6- to 8-month-old English learning infants’ were tested in theConditioned Head Turn Procedure with the two German vowel contrasts (Polka & Werker,1994). Because the ratings of quality provided by English speakers tested in the wordmatching task were considered to be similar to Kuhl’s judgments of category goodness, itwas hypothesized that the German vowels should engender effects similar to thoseproduced by Kuhl’s prototypical vowels. To allow for this possibility, subjects heardeither the front rounded vowel as the background stimulus and the back rounded vowel as18 A cut-off of at least 90% correct was used as German speakers native-like criterion.15the test stimulus or the reverse. A magnet effect would be evident if subjects presentedwith the back rounded vowel (the better exemplar) as a referent performed worse than thosepresented with the front rounded vowel as a referent.The results revealed a significant difference between adults and infants. Virtuallyall the adults, both German and English discriminated these German vowel contrasts: 10out of 10 German adults reached criterion on both contrasts and 10 out of 10 English adultsreached criterion on the tense contrast with 8 out of 10 reaching it for the lax.In contrast to the evidence from adults, infants in general had more difficultydiscriminating the German vowels. However, significantly more infants from the 6- to 8-month age group discriminated these contrasts: 14 of the 40 younger infants reachedcriterion but only 3 of the 28 older infants did. This was the first evidence that thelinguistic environment modifies infant discrimination of a between-category non-nativevowel contrast. The evidence also suggests that the effect of the language environment onvowel perception may begin by 6 to 8 months but is not complete until about 10 months.Interestingly, there was evidence of a perceptual magnet effect in the youngerinfants. Those 6- to 8-month-old infants presented with the front rounded vowel as thereferent were more likely to reach criterion: 80% reached criterion on the lax and 60% onthe tense contrast. When presented with the back rounded vowel as the referent, however,only 10% (1 of 10 infants) reached criterion on the lax and none did so on the tense (Polka& Werker, 1994). On the other hand, older infants did not demonstrate a magnet effect as1 of 14 infants aged 10 to 12 months reached criterion on the tense contrast and 2 of 14 didso on the lax contrast19. These results are consistent with Kuhl’s evidence of a magnet19 To determine if the effects of the native language may be first evident as a perceptual magnet effect at6-months, Polka and Werker (1994) extended the research to infants under 6 months. Four- and 6-month-old infants were tested in a habituation-dishabituation procedure with the same German vowel contrasts.Only the 4-month old infants discriminated the German vowel contrasts and order of presentation did notaffect discrimination in either 4- or 6-month-old infants. Thus, these results suggest that the linguisticenvironment begins to have an effect on non-native vowel perception by 6 months of age although theperceptual magnet effect may only be evident when infants are tested in a head-turn procedure. In anotherstudy using these same German vowels, German and English infants were subject to the same order effectssuggesting that this developmental shift may be due to markedness of vowels (Polka & Bohn, 1994).16effect in 6- to 8-month-old infants. However, whereas KuhFs results reveal a perceptualmagnet effect on discrimination of exemplars within one native language vowel category,Polka and Werke?s evidence reveals an effect on between category discrimination of non-native vowels (that may be assimilated to a single native language vowel category).In summary, the development of cross-linguistic vowel perception follows adifferent trajectory than the development of consonant perception. First, the evidencesuggests that the effect of the ambient language on vowels and consonants may begin atdifferent ages, albeit both in the first year of life. Although consonant perception showsevidence of reorganizing at 10 to 12 months, the reorganization in vowel perception maybegin about 6 months of age as evidenced by a perceptual magnet effect. Still, vowelperception may only show evidence of full reorganization by 10 months of age. Second, itappears that adult judgments of goodness play an important role in the internal structure ofvowel categories as well as in perception of non-native vowels but there is no evidence todate that consonant categories are subject to a perceptual magnet effect. Finally, the mannerin which the ambient language influences vowel and consonant perception may differ.Consonant categories appear to become more adult-like while vowel categories appear tobecome narrower but not necessarily adult-like because English adults easily discriminateboth German vowel contrasts. This difference still needs to be investigated.Approaches and the Need for Further Empirical ResearchDespite the extensive research investigating the developmental reorganization inphonetic perception, there remain gaps in the literature. By addressing some of theseremaining questions we can further elaborate the factors underlying the developmentalreorganization. First, a brief summary of what we do know: Adults attend to nativelanguage phonemic information in consonants, discriminate between both native and nonnative vowel contrasts, and in some sensitive testing situations, discriminate non-native17phonemic consonant contrasts and acoustic differences. In contrast, 10- to 12-month-oldinfants discriminate only native phonemic contrasts irrespective of whether they are vowelsor consonants. Six- to 8-month-old infants discriminate non-native consonant contrastsand are subject to a perceptual magnet effect when tested with non-native vowel contrasts.Finally, we know that lack of exposure is not the only factor in the reorganization as olderinfants and adults continue to discriminate Zulu clicks that do not occur in the nativelanguage.The present research addresses several unknown factors and each of these relate toembryonic models proposed to describe and/or explain the developmental changes in non-native phonetic perception. This includes Kuhl’s perceptual magnet hypothesis outlinedpreviously (Kuhl, 1991: Kuhl et al., 1992), Best’s assimilation model (1988, 1994), andWerker and Pegg’s four-factor model (1992). Although other theoretical models have beenproposed to explain the more general overall tuning of speech perception to various featuresof the native language (e.g., Jusczyk, 1993; Jusczyk, Hohne, & Mandel, in press), there isno comprehensive theory of developmental changes in phonetic perception20. Thus,further research is needed to help inform theory.Before discussing the theoretical impetus for this research, it is necessary to brieflydiscuss the selection of the phonetic distinction to be used in this study. An Englishcontext-dependent allophonic distinction is the variation in alveolar stop consonants.English allows two alveolar stops in initial position—the voiced unaspirated alveolar stop,[d] and the voiceless aspirated alveolar stop, [tl9. A third variant occurs following Is/—thevoiceless unaspirated alveolar stop [t] (hereafter transcribed as (s)[tJ or (s)[ta] to clearlydifferentiate it from [thj)21. To assess discrimination of a native language phoneticdistinction, it is necessary to remove one of the context-dependent variants from its20 Several hypotheses have been generated to explain the developmental reorganizatin. They includenotions of perceptual development, articulatory mediation, dynamical systems, cognitive development, andphonological development (see Werker & Pegg, 1992, for a discussion of each view).21 This phone may be produced by some speakers as flap.18standard context. I decided to remove the voiceless unaspirated stop (s) [t] from the contextof [s] and present it in initial position. Thus, the first member of the pair to be assessed isthe variant (s)[taj.The second member of the pair will be [da]. Linguistically speaking, [thi and (s)[t]are allophonic variants of the phoneme It! while [da] and (s)[taj are not allophonic variantsof a single phoneme but rather come from two different underlying phonemic categories.Nevertheless, [da] was selected as the second member of the pair because (s) [ta] and [da]are acoustically and perceptually similar and they still constitute a language-specificphonetic distinction. Evidence of acoustic similarity between (s)[ta] and [da] is apparent inTable 1 (Experiment 1). As can be seen in this table, the only feature consistentlydifferentiating (s)[ta] from [da] is the F2 onset (a consequence of coarticulation). Evidenceof perceptual similarity involves the fact that English listeners rely on V0T22 to distinguishbetween stop consonants (Lisker & Abramson, 1970; Lotz, Abramson, Gerstman,Ingemann, & Nemser, 1960). Because VOT does not differ between (s)[ta] and [da], thesetwo phones should be perceived as very similar.Perceptual Magnet HypothesisPerceptual magnet effects have been revealed for both within- and between-categoryvowel perception. Recall, Polka and Werker (1994) extended Kuhl et al.’s (1992)evidence of a magnet effect for within category vowel perception to between categoryvowel perception. Thus, it is possible that perceptual magnet effects could arise in thepresent research investigating between category perception for a phonetic distinction. It isnot known, however, whether consonant perception is subject to a perceptual magneteffect. Thus, there are two main differences between the current research and Kuhl’sresearch. First, I am attempting to replicate Polka and Werker’s (1994) evidence that themagnet effect generalizes to between category discrimination (as opposed to within22 In reminder, VOT is voice-onset-time.19category discrimination). Second, I am assessing the possibility that the perceptual magneteffect generalizes to consonant perception (as opposed to vowel perception).Language-specific phonetic variants provide an ideal test of this possibility,particularly if adults judge both phones as members of the same phonemic category but saythat they differ in quality of match to that phoneme. That is, I predict that exemplars ofboth [da] and (s)[ta] will be considered members of the phonemic category Id!. It ispossible, however, that adults might judge [da] as a more central exemplar of the phonemiccategory Id! than (s)[ta]. Thus, [da] could act as a perceptual magnet. To evaluate thesepredictions, I will initially assess adult goodness judgments of the two English phones inExperiment 1 and then determine if there are perceptual magnet effects when adults aretested in two discrimination procedures (Experiment 2 and 3).Assimilation HypothesisThe assimilation model was proposed by Best, McRoberts, and Sithole in 1988 toexplain variations in discriminability of non-native phonemes (expanded in Best, 1994).Best describes four types of comparisons between non-native speech sounds and nativelanguage phonology that will predict different degrees of discrimination. (1) The easiestnon-native speech sound to discriminate is one that has no native language category towhich it can be assimilated23.An example of a non-assimilable contrast is the Zulu clickcontrast <ca> versus <xa>. English speaking adults and 10- to 12-month-old infants fromEnglish were presented with this contrast (Best et al., 1988). No English speech soundsare similar to either of these Zulu clicks. The results indicated that both adults and 10- to12-month-old infants easily discriminate this contrast. (2) When the non-native phones areassimilated to two different native language phonemic categories, discrimination is slightlymore difficult, but still possible. As well, adults are amenable to discrimination training for23 In current writings (e.g. Best, 1994), assimilability is discussed in terms of the overlap in articulatorygestures rather than in phonological categories. As well, Best has included other possibilities such assounds perceived as non-speech. Neither of these changes affect the present study because Best’s predictionsdo not change and the speech contrast in question falls within the realm of speech.20these two categoly contrasts. An example is the Hindi contrast /tl1a1 versus Idhal (Werkeret al., 1981). (3) More difficult discrimination is apparent when the non-native phones areboth assimilated to the same category but differ in category goodness (the Nthlakampxcontrast /ki/ versus 1q11; Werker & Tees, 1984a). (4) Finally, the most difficult non-nativecontrasts to discriminate are those which are assimilated to a single categoiy in the nativelanguage and do not differ in category goodness such as the Hindi retroflex versus dentalcontrast, Ida! versus /Da! (Werker & Tees, 1984a). These contrasts are expected to be theleast amenable to training.Figure 2 depicts the 1988 version of Best’s model. The least assimilable contrast(easiest to discriminate) is on the outer edges of the continuum and the most assimilable isin the middle (the most difficult to discriminate).Figure 2: Graphic representation of Best et al.’s assimilation model.Of particular importance here, is the fact that both adults and infants discriminate thenon-native Zulu click contrast despite lack of linguistic exposure to this speech sound (Bestet al., 1988). The evidence suggests that lack of exposure cannot be the only factor in thedevelopmental reorganization. Rather, the results support the notion that the importantfactor in discriminability is the presence or absence of native language phonemic categoriesto which the non-native phonemic contrast can be assimilated. That is, if there is no native/id!NONNATiVENATiVEIda! /Da!/dhaJ4iiItaJIda]/da]21language phonemic category sufficiently similar to members of the non-native contrast, theability to discriminate the non-native contrast remains.This model directly relates to the present research. First, we do not know whethersimple linguistic exposure to a phonetic form (short of phonemic status) will maintaindiscrimination. Although absence of exposure is not the only factor influencing infantspeech perception (Best et al., 1988), presence of exposure may affect discrimination. Inaddition, although there is evidence to suggest that adults can use allophonic information tounderstand the meaning of ambiguous utterances (Dutton, 1992), we do not know whetheradults will discriminate two native phones that are acoustically and perceptually similar.Furthermore, if adults do discriminate language-specific phonetic information, we do notknow under what testing conditions discrimination will be evident. By assessingperception of a language-specific phonetic distinction, we will determine if linguisticexposure alone is sufficient to maintain discrimination.A second issue arises from the Best et al. model. They suggest that if two non-native exemplars are both perceived as equally good members of the same phonemiccategory, discrimination should be more difficult than if one phone is judged as less goodthan another (Best, 1994). Such differences in goodness might be evident as a perceptualmagnet effect. Thus, if when tested with a native phonetic distinction, adults revealdifferences in judgments of quality of the phones (s)[ta] and [da], discrimination should bemoderate. However, if there are no differences in judgments of quality, discriminationshould be difficult.The Four Factor ModelThe Four Factor Model, proposed by Werker and Pegg (1992), is an attempt toprovide a descriptive framework for the developmental changes in non-native phoneticperception outlined in the literature review. The four-factors in this model are, acoustic,broad-basedphonetic, language-specific phonetic, and phonemic. Briefly, evidence of the22acoustic factor is revealed in adults when they show discrimination of within categoryexemplars—in other words, discrimination of two exemplars that are not used in thephonemic inventory of any known language (e.g. Pisoni & Tash, 1974; Carney, Widin, &Viemeister, 1977). The broad-based phonetic factor describes the ability of 6- to 8-month-old infants to discriminate non-native phones that correspond to phonemes used in one ofthe world’s languages other than their own. Evidence of this ability has been reported inmany studies (e.g., Aslin et al., 1981; Best et al., 1988; Eimas et al., 1971; Werker &Tees, 1984a). Broad-based phonetic also describes the ability of adults to discriminatenon-native phonemes in some testing situations (e.g., Pisoni et al., 1982; Werker &Logan, 1985).The term language-specific phonetic is intended to describe the fact that although10- to 12-month-old infants show evidence of discriminating only native phonemiccontrasts, they do not yet use these categories to distinguish meaning (Stager & Werker,1995; Werker & Baldwin, 1991; Werker & Pegg, 1992). Finally, the phonemic factor ismeant to describe the evidence that at some point in development the perceptual phoneticcategories are functionally phonemic and adults clearly show this pattern of discrimination.There is no evidence to date differentiating acoustic and broad-based phonetic in infantperception and no definitive evidence distinguishing language-specific phonetic fromphonemic in adults. Nevertheless, a developmental progression is implied by these termsas it is expected that an individual will move from broad-based phonetic (or acoustic) tolanguage-specific phonetic and then to phonemic. This does not mean that adults cannotperform at previous levels, only that infants do not perform at higher levels and the mostreadily accessible level for adults is the phonemic.Of specific interest to the present thesis are the conceptualizations of broad-basedphonetic and language-specific phonetic factors. Broad-based phonetic was meant to depictthe ability of 6- to 8-month-old infants to discriminate most non-native contrasts. We donot know, however, if this broad-based phonetic discrimination extends to only those23phonetic distinctions which have been used in the phonemic inventory of one of theworld’s languages or whether it extends also to native language phonetic variation. Giventhe evidence of broad-based phonetic perception in 6- to 8-month-old infants, it is likelythat infants will discriminate language-specific phonetic variants. Additional support forthis prediction arises from the evidence that infants of 2 months discriminate allophonicinformation in medial position (Hohne & Jusczyk, 1994). This indicates that very younginfants can attend to native allophonic information that might be used to segment the speechstream which suggests in turn, that non-random phonetic variation may be not only usefulbut also perceptually salient to young infants.The possibility also exists, however, that if the phonetic variation distinguishingtwo phones is subtle, does not correspond precisely to that used as a phonemic contrast inany of the world’s languages, and is presented without supporting contextual information,young infants may not be able to discriminate it. Recall, young infants often fail todiscriminate phonetic variation from within a phonetic category . This has led researchersand theorists to hypothesize that young infants may be endowed with sensitivities thatconstrain them to detect only that phonetic variation that conforms to a phonemic distinctionin one of the world’s languages (e.g. Eimas, 1975). If this is correct and the language-specific phonetic distinction is not used phonemically in any language, young infants couldfail to discriminate. Thus, it is essential to test 6- to 8-month-old infants on their ability todiscriminate non-phonemic, systematic phonetic variation, that occurs in their nativelanguage. This will allow us to determine if broad-based perception includes onlysensitivity to the world’s phonemes or if it includes sensitivity to any systematic phoneticvariation.Finally, the ability of infants to discriminate native phonetic (non-phonemic)differences may or may not differ as a function of the age of the infant. That is, we do notknow if the perceptual biases of 10- to 12-month-old infants are always in accordance withnative language phonemic categories or whether infants this age will also attend to native24phonetic variants. There are two different predictions about how 10- to 12-month-oldinfants will perceive native phonetic information arising from two differentconceptualizations of language-specific phonetic processing (Werker & Pegg, 1992).According to the first prediction, since context-dependent allophones are phonologicallyrelevant to the infant’s native language (albeit not phonemic), infants may attend to them.Thus, when one variant is removed from its context and presented in the context of anacoustically similar phone, infants may discriminate such a pair. If so, perception in 10- to12-month-old infants would not differ from perception in 6- to 8-month-old infants as bothage groups would show discrimination. Such evidence would indicate that thereorganization in cross-language speech perception is not necessarily related to thephonemic status of speech sounds but may be related to actual linguistic exposure.A second prediction regarding the meaning of language-specific phoneticprocessing involves the notion that the 10- to 12-month-old infant’s language-specificcategories are equivalent to phonemic categories but are not yet used to contrast meaning.According to this view, the reorganization in infant speech perception is a fundamental stepin creating language-specific categories that will eventually be used to differentiatemeaning. These new perceptual categories may be recruited at some later point indevelopment to allow acquisition of lexical terms (e.g., Stager & Werker, 1995; Werker &Baldwin, 1991; Werker & Pegg, 1992). If this view accurately portrays older infants’abilities, 10- to 12-month-old infants should fail to attend to native-language phoneticvariants. Such evidence would provide further support for the hypothesis that thedevelopmental reorganization is based on phonemic principles. Thus, this experiment willallow us to determine which conceptualization of the language-specific phonetic factor in10- to 12-month-old infants is more accurate.In summary, all of these questions can be addressed by assessing perception of anative language phonetic distinction. The results from each experiment will enable us tofurther elaborate the embryonic model of Best Ct al. (1988), to examine the generalizability25of the perceptual magnet effect (KuhI et al., 1992), and to further investigate the underlyingstructure of both broad-based and language-specific phonetic factors (Werker & Pegg,1992).The Present ResearchThe present experiments will determine first, if when both phones are presented ininitial position, English adults treat the English phone (s)[ta] as perceptually equivalent tothe phone [da]. In the first experiment, adults were given the opportunity to judgeexemplars as good or less good examples of a written character. In the second experiment,adults were tested using the AX (same or different) procedure to determine if adults are ableto discriminate between the phonetic variants when tested in a sensitive procedure. Thefinal experiment with adults was designed to ascertain if they are able to discriminate thelanguage-specific phonetic distinction in a procedure that typically elicits native languagephonemic perception. In this experiment, adults are tested in a category change paradigmthat also allows direct comparison between adult and infant discrimination. Finally, infantsof both 6 to 8 and 10 to 12 months of age were tested in the Conditioned Head TurnProcedure (a category change paradigm but with minor modifications) to determine if thereis a developmental change in perception of these phones between older and younger infantsand to determine if there is a difference between adult and infant discrimination of the nativephonetic distinction.26EXPERIMENT 1The Category Goodness TaskAs reviewed above, previous research has shown that in most testing situations,adults categorize consonants in accordance with native-language phonemic categories (e.g.,Abramson & Lisker, 1970; McKain et al., 1981; Miyawaki et aL, 1975; Trehub, 1976;Werker, 1991; Werker & Pegg, 1992; Werker & Tees, 1984a). That is, adultsdiscriminate well among phonetic variations that signal a difference in meaning in theirnative language but show poorer discrimination perfonnance on phonetic variations withinnative phonemic categories (e.g., Repp, 1984; Werker & Logan 1985; Werker & Tees,1984b). It would seem that such a strategy is functional in that it allows listeners to attendto the meaning of utterances while not being confused by irrelevant phonetic differences.Even within the above pattern, however, several studies show that non-nativecontrasts differ in their discriminability (e.g., Best et al., 1988; Polka, 1991; Werker etal., 1981). For example, in their initial work, Werker et al. reported that English adultsfind the Hindi retroflex/dental distinction Ida! versus Ira! more difficult than the Hindivoicing distinction Ida! versus /dhal.Previous researchers have attempted to systematically explore the parameters thatcontribute to discrimination of non-native contrasts. As reviewed above, Best et al. (1988)hypothesized that differences in ease of discrimination can be explained by the extent towhich non-native contrasts can be assimilated to the native phonology. Another approachtaken by Polka (1991) was to hypothesize that discrimination differences may be accountedfor by whether the non-native contrast includes members that exist as allophones in thenative language. To test this hypothesis, she used four non-native contrasts that differedwith respect to their English allophonic status: either one, both or none of the members27were similar to English allophones24.The results did not support the hypothesis: althoughthere were differences in ease of discrimination among the four non-native contrasts,performance could not be predicted by the native allophonic status. Instead, a post-hocanalysis revealed that the pattern of results were best predicted by Best’s assimilationmodel.A recent experiment was designed to assess the impact of native allophonicinformation on perception (Dutton, 1992). Adults were presented with several two wordutterances that differed in the location of the word boundary such as “great tie”, “greateye”, and “gray tie”. In each series, acoustic correlates of allophonic cues were variedsystematically to determine if changes in these cues lead to the percept of differentutterances. In the above utterances, the acoustic cues are prior vowel duration, closureduration, and aspiration duration. Evidence that adults use this phonetic information arisesfrom the fact that a change in the acoustic cues changes the percept. It appears, therefore,that adults can attend to allophonic variation when it is presented in context and when suchvariation can be used to disambiguate the meaning.Another study suggests that adults fail to discriminate even native languagephonemic contrasts if the contrast is not presented in its standard context. Repp and Lin(1989) constructed a synthetic [ba][pha] continuum varying VOT in 10 ms steps. Theypresented these stimuli to English adults in an AXB discrimination task. Thediscrimination function showed a sharp peak at about 24 ms of VOT revealing evidence ofphonemic perception of this continuum. However, when the same continuum waspreceded by Is!, the English adults failed to show this category boundary effect. That is,the characteristic peak in discrimination between [ba] and [pha] was not evident when themay be important to point out that Polka used non-native contrasts which included membershypothesized to be similar to native allophones. However, transcriptions of non-native phonemes andnative allophones may be identical but productions of each may differ in important acoustic or phoneticcharacteristics. In addition, not all allophones occur systematically, some occur in free variation. Forexample, the English medial position flap [T], as in some people’s productions of “little,” is an optionalallophonic variant of It?. The allophones used by Polka were of this type.28syllables were preceded by /5• Since differences between bilabial stops following Is! donot differentiate meaning in English26,this evidence supports the notion that adults processspeech on a phonemic level. As Repp and Lin state, “To the unsophisticated listener, both[spa] and [spha] are Ispal.” (Repp & Lin, 1989, p. 324)Repp and Lin’s research may be considered a study of allophonic perception in thatit examined the effect of placing two different phonemes in a context where only a singlephoneme is allowed in English omitting the allophonic variant that typically occurs in thiscontext: the naturally produced allophone [p1 as in “spin.” In the research presented here,I examine adult perception of two phonetic variants in a context where only a singlephoneme is allowed. The phonetic variant, (s) [ta], is presented in a context in which itdoes not regularly occur. In other words, I assess whether both (s)[ta] without the Is! and[da] are perceived as Ida!.Sawush and Jusczyk (1981) also investigated the effect of a preceding is/ onperception. They investigated the adaptation effects of synthetically produced [spha],[spa], [sba], and [sla] on a [ba] to [pha] continuum to determine if these adaptorsdifferentially influenced the location of the perceived boundary between fb/ and ipi. In thesecond experiment, Sawush and Jusczyk examined adult identification of two bilabialstops—a 0 ms VOT exemplar and a 10 ms VOT exemplar—both from the fbi end of theVOT continuum. When these two bilabial stops were presented in initial position (inisolation), subjects identified the 0 ms VOT exemplar as “B” a mean of 94% of the time andthe 10 ms VOT exemplar as a “B” 87% of the time. Yet when presented with the samesignals preceded by an is!, English listeners identified the 0 ms VOT exemplar as a “B”only 40% of time and the 10 ms VOT exemplar as a “B” only 16% of the time. Thisevidence reinforces the notion that the percept of a bilabial stop preceded by 1sf is usually25 Recall, phones are transcribed in square brackets and phonemes are transcribed in slanted brackets.26 In English phonology, bilabial stops follow the same rules as apical stops: the phonemes are thevoiced unaspirated stop [ba] and the unvoiced aspirated stop [pha] but when an Is! occurs before a bilabialstop, the allophomc variant is the unvoiced unaspirated stop [pa].29labelled as a “P’ even when the VOT falls well into the range perceived as “B” when no isIis present.Both the study by Sawush and Jusczyk (1981) and the one by Repp and Lin (1989)examined perception of synthetically produced stimuli preceded by Is!. Only one earlystudy examined perception of naturally produced stop consonants of this type (Lotz et al.,1960). Lotz and colleagues recorded an American English speaker’s productions of stopconsonants for each place of articulation (labial, alveolar, and velar) either in initial positionor following Is/. They then removed the Is?. Individuals from several languageenvironments were asked to identify the stops. While American English listeners identifiedthe consonants spliced from Is/ as voiced segments [b, d, g,] (apparently relying onabsence of aspiration), Spanish and Hungarian listeners identified them as unvoicedaspirated [ph, th, kh] (apparently relying on absence of voicing) and Thai listenersidentified them as unvoiced, unaspirated segments [p, t. k] (thus attending to both voicingand aspiration).On the basis of all the above data, we expected that English listeners would hear a[dl when presented with (s) [ta]. That is, although (s) [ta] and [da] are derived from twodifferent underlying phonemes in English, when they are presented in initial position, theydo not constitute a phonemic contrast. The acoustic cues to a phonemic distinction are notpresent. Thus, English listeners presented with (s)[ta] and [da] exemplars should hearvariants of a single English phoneme.Although English listeners are predicted to hear variants of a single Englishphoneme, it is not known whether adults will hear differences in quality between thephonetic categories. Acoustic differences between these phones do exist (see Table 1) andgiven the data from Spanish and Hungarian listeners (Lotz et al., 1960), it is apparent thatphonetic detail is still remaining. This detail could allow at least some listeners todiscriminate [da] from (s) [ta].30This first experiment assesses adult perception using a category goodnessprocedure to determine if adults detect differences between (s)[ta] and [da] exemplars.Differences in quality will be evident if adults rate exemplars from one category as betterexamples of initial position alveolar stops than the other. Since [da] does occur in initialposition and (s)[ta] never does, it is possible that when adults are asked to rate categorygoodness of the exemplars, they will judge [da] to be a better example of the category ofinitial position alveolar stops than (s)[ta]. If this is the case, Kuhl’s work (1991) wouldpredict that when adults are tested in discrimination tasks (Experiments 2 and 3), they willshow an asymmetry consistent with the perceptual magnet effect.It is also possible that both (s)[ta] and [da] will be assimilated to the Englishphonemic category Ida! and that adults will consider both (s)[ta] and [da] to be equally goodmembers of that category. That is, since (s)[ta] never occurs in initial position and since ithas a short VOT, English listeners may hear a [da] and not notice any phonetic differencesbetween the exemplars. If that is the case, Best’s model (Best et al., 1988) would predictthat adults would have difficulty discriminating this contrast when tested in the categorychange procedure (Experiment 3).Although the primary purpose of this experiment is to see whether (s)[ta] and [da]are perceived as equally good members of the same phonemic category, this study alsoprovides important information as to whether any of the 10 individual exemplars in thestimulus set are perceived as consistently different in quality. If one of the exemplars isconsistently judged as either a good or a poor member of the category, it might also bemore easily discriminated from other exemplars on that basis, and would be excluded fromthe stimulus set in the subsequent studies.In summary, by collecting category goodness ratings from English adults, Idetermined whether adults label these phonetic variants as from the same or differentcategories and determined whether adults detect subtle differences in quality of the various31exemplars. The results from this initial study have implications for predictions insubsequent studies with regard to both discrimination and order effects.MethodSubjects: Thirty English university students between the ages of 18 and 30 yearsparticipated in this study. All subjects were monolingual speakers of English and had noknowledge of any other language before the age of 8 years27. They were recruited in oneof two ways. (1) If students were interested in volunteering as research participants andreceiving an extra credit to apply to a psychology course of their choice, they filled out acard with their name and phone number. (2) A sign up sheet was posted on the bulletinboard in the Psychology Department with a brief description of the procedure. In bothcases, respondents were later contacted by phone, given more details about the task, and ifinterested, appointments were made. Following the procedure, they were given a verbaldescription of the study and a two-page summary explaining the theoretical basis of theresearch project.Stimuli: The speech samples used in this study were produced by a male, native Englishspeaker. Several exemplars of [daj and [sta] were digitized directly into a Macintosh II FXcomputer using the Signalize speech analysis program28 via a GW Instruments analog-to-digital board (Model GWI-AMP). In addition, several exemplars of [thai were recorded.These [thai exemplars are not part of the stimulus set but were recorded to allowcomparison of acoustic characteristics. All signals with extreme values on non-phoneticcues such as fundamental frequency (FO), amplitude, intonation contour, and durationaldifferences were eliminated. Fifteen exemplars were selected for further analysis on the27 One person had been briefly exposed to Greek at 4 years of age.28 The Signalize Program was created by Eric Keller:32basis of similarity in each of these acoustic (non-phonetic) features - five for each category.Finally, the selected [sta] exemplars were modified by deleting the initial [s], taking carenot to remove any of the characteristics of the burst release of the alveolar stop consonant.Five exemplars for each phonetic category of [da], (s)[taj, and [tha] were importedinto the Mac Speech analysis program where spectra and spectrograms were generated toidentify criterial phonetic characteristics which might differentiate each of the categories.As can be seen in Table 1, the phonemes [dai and [tha] differ in VOT, burst duration, FOonset, Fl onset (the first formant), and F2 onset (acoustic cues differentiating [thai from[da] are italicized). They have overlapping values in syllable duration.Table 1: Acoustical analyses for each of the 15 exemplars.ITEM VOT (ms) Burst Syllable FO Onset Fl Onset F2 OnsetDuration Duration (Hz) (Hz) (Hz)(ms) (ms)thai 17.45 23.60 336 106 675 1479th 41.00 22.90 397 105 726 1366tha3 35.75 20.45 389 103 726 1400tha4 46.50 34.50 407 102 718 1298t1¼5 36.10 33.75 415 105 675 1358thi 6.30 15.20 350 96 449 1600da2 5.30 12.95 374 101 476 1600da3 3.50 14.60 353 98 458 1643da4 3.35 14.05 403 98 432 1617da5 6.20 13.05 369 100 458 1505(s)tal 7.55 14.85 347 101 441 1410(s)ta2 2.95 15.60 338 100 458 1418(s)ta3 1.70 15.25 353 100 459 1410(s)ta4 6.35 17.45 339 100 484 1436(s)ta5 3.75 15.55 362 101 484 140133Of importance, the allophones [da] and (s)[ta] differ only in F2 onset and have overlappingvalues in each of the other measures29. The main acoustic cue for differentiating between(s)[ta] and [da] is therefore, a higher F2 onset for [da] than (s)[ta] (in bold). Thisdifference is likely due to the coarticulation effects of producing an alveolar stop followingIs? leading to a decrease in the F2 value as explained by source-filter theory (e.g., Kent &Read, 1992).The 10 digitized stimuli to be used in the stimulus set (5 (s)[ta]’s and 5 [daj’s) wereimported into the Bliss System30. This program controlled the presentation of the stimuli.Apparatus and Procedure: Identification of exemplars is often assessed by giving thesubject a choice of key words or syllables in written form, to which a presented exemplar iscompared. Adults select a key word from a set of words or syllables which best representsthe category of the presented exemplar (e.g., Best et al., 1988; Kuhi et al., 1992; Polka,1995; Werker & Logan, 1985). In the present study, however, it was necessary to designa new identification procedure. Since English spelling does not differentiate betweenallophones, listeners would be trying to match the exemplars to letters. Thus, requiringlisteners to select one match from among a set of written English syllables could result in abias of listeners to search for phonemic, not phonetic, differences in exemplars. The newprocedure was designed to try to elicit differential ratings of quality among the phoneticexemplars. Subjects were tested in one of three conditions (10 per condition): They weregiven a response sheet with either “sta”, “da”, or “sda” printed at the top and the “d” or “t”were in bold print. (An example of a response sheet is in Appendix 2.)31 Subjects were29 It should be noted that it has been reported in the literature that the mean FO onset of (s)[ta] and [da]differ. Indeed, lower mean FO onset for [da] is reported to be one of the consistent acoustic cuesdifferentiating these allophonic variants (Mary Beckman and Doug Whalen, personal communication).Although there is a hint of this difference in the stimuli used in the current study, the two categories cannotbe distinguished on this basis since there is overlap in FO onset between the two categories.30 The Bliss Program for conducting experimental tests was designed by John Mertus of BrownUniversity.31 We did not include a condition with “ta” page heading because the set of stimuli did not include [thaiand our pilot observation indicated that English subjects simply do not perceive these stimuli as “t”. Thus,we were likely to have a floor effect in quality ratings.34presented with the 10 individual exemplars and their task was to rate the goodness of eachof the exemplars in relation to the referent, the one printed in bold at the top of their page.Each individual exemplar was followed by a 5 second response interval during which theywrote a number between 1 (poor) and 7 (excellent) on the response sheet.Subjects were seated in a sound-attenuated chamber in groups of no more thanthree. They were seated with their back to each other to ensure that no subject saw theother’s response sheet. Before being given the response sheet, subjects were told not toutter the syllable on the top of the page and were given instructions to rate the quality ofeach presented signal relative to the referent shown in bold print at the top of their page.They were also told that although the task was very difficult for most people, they shouldnevertheless try to listen carefully for differences between the sounds. It was hoped thatsuch explicit instructions would facilitate discrimination. Exemplars were presented freefield by a Compaq 286 computer using the Bliss Experimental Control System and werepresented at a comfortable listening level of 65 dB.Subjects were tested in two sessions with a 3-5 minute break between sessions.During each session, they were asked to rate 60 exemplars (see Table 2). Each of the 10individual exemplars was presented 6 times per session. Thus, in total there were 60presentations of [da] and 60 of (s)[ta] for a total of 120 responses and 12 judgments foreach individual exemplar. The order of presentation was randomized for each group ofsubjects and for each session.Table 2: Experimental Design_________________SESSION 1 SESSION 2Stimuli—>Group 1. Page Heading staGroup 2. Page Heading sdaGroup 3. Page Heading da[da] (s)[taj [da] js)[ta] II• 30 30 30 30 12030 30 30 30 12030 30 30 30 12035ResultsTo determine if judgments of category goodness differed as a function of thecategory of the allophone, a mixed-model Analysis of Variance (ANOVA) was conductedin which the group variable was page heading (“sta”, “sda”, or “da”) and the two repeatedmeasures were session (session 1 versus session 2) and phonetic category ([daJ versus(s)[ta]). The dependent variable was the mean rating value.There was a main effect for page heading, F(2,27)=5.47, p<.Oi, and an interactionbetween session and page heading, F(2,27)=3.24, p<.05. No other main effects orinteractions were significant. The main effect for page heading was analyzed by comparingmeans using Fisher’s Protected LSD. The results indicated that adults judged all exemplarsas less good when “sta” was the page heading than when either “da”, p<.005 or “sda”,p<.02, were the headings (See Figure 3). Ratings were not significantly different when“sda” versus “da” were page headings indicating that adults judged all exemplars as betterexamples of “D” than “T” even when the “D” follows “s”.Figure 3: Main effect foi page headingMeanRatings(Max=7)sta sda daPage headings36The interaction between session and page heading was analyzed by comparingsession for each page heading. When “sta” was the page heading, session was notsignificant. Likewise, when “da” was the page heading session was not significantalthough the means were slightly higher during the second session (First Session M=4.7 1,SD=.92; Second Session M=4.85, SD=l.OO). In contrast, when “sda” was the pageheading, there was a trend for higher goodness ratings during the first session,F(l,9)=3.87, p<.O8 (First Session M=4.74, SD=.42; Second Session M=4.35, SD=.77).Stifi, during the second session, signals were not necessarily considered poor examples ofthe category, simply less good.Individual Exemplars: To determine if some exemplars should be omitted from subsequentstudies, a second ANOVA was conducted to see if listeners judged some exemplars asconsistently different from the rest. Although session interacted with page heading in thefirst analyses, none of the individual effects reach a .05 level of significance. Thus, Idecided to collapse across session in this analysis. In contrast, I did not collapse acrosspage heading because of the a-priori prediction that this factor would significantly affectperception of quality. In this mixed-model ANOVA, therefore, the group variable waspage heading (“sta”, “sda”, and “da”) and the repeated measure was exemplar ([da]l,[da]2, [da]3, [da]4, [da]5, (s)[ta]l, (s)[taJ2, (s)[ta]3, (s)[ta]4, and (s)[ta]5)32. Thedependent variable remained mean rating score.All factors were significant. There was a main effect for page heading, F(2,27)=5.80, p<.008, showing the same effect as described above: All exemplars were ratedhigher for “sda” and “da” page headings than for “sta” page headings. There was also atrend for exemplar, F,(9, 243)=2.53, p<.O6 (Greenhouse-Geiser adjustment), and aninteraction between page heading and exemplar, F(18, 243)=2.75 p<.02 (Greenhouse-32 Note that it was not valid to include allophonic category in this analysis because exemplars overlappedwith category. Furthermore, a nested design would compare [dali with (s)[taj 1, [daJ2 with (s)[ta}2, etc.,which is also an invalid comparison.37Geiser adjustment). Overall, [da]3 was considered the best exemplar and [da]4 wasconsidered the worst but the interaction revealed more complex results.Follow-up analyses of the interaction involved examining “sta” page heading aloneand combining across “sda” and “da” page headings together (recall “da11 and “sda” pageheadings were not significantly different in the initial analysis). When “sda” and “da” werepage headings, page heading was not significant, as expected, and there was no interactionbetween exemplar and page heading. However, exemplar was significant, F(9, 162)=6.52p<.0002 (Greenhouse Geiser adjustment). A follow-up comparison of means for eachexemplar revealed that [daJ3 was judged the best exemplar and differed from (s)[ta]5 andlower, F( 1 )=8 .81, p<.O3 (Greenhouse-Geiser) while (s) [taJ4 was the worst and differedfrom (s)[taJ2 and higher, F(1)=9.37, p.c.O2 (Greenhouse Geiser). The ordering ofexemplars for these page headings is displayed in Figure 4 in which non-significantdifferences between exemplars are shown by underlining.Figure 4: Ratings for exemplars when “sda” and “da” (D) are page headings7654Ratings321038When “sta’ was the page heading, exemplar was not significant. All exemplarswere perceived as less good instances of the category (Range=3. 12 to 4.06). Indeed, thehighest rated exemplar for the “T” heading is lower than the lowest rated exemplar for thecombined “D” headings. This reveals that page heading has a strong effect on qualityratings but no exemplar stands out as consistently different. Thus all exemplars are used insubsequent experiments.DiscussionEnglish listeners in this study rated all exemplars as good examples of the categoryof “D” whether the “D” was in initial position or follows “s”. Interestingly, even thoughEnglish spelling rules do not allow “sd”, listeners rated the signals as good examples of the“sd” referent. They did not, however, rate the exemplars as good examples of the “st’treferent even though English spelling rules allow “st”. Indeed, inspection of the means forthe individual exemplars shows that even the highest rated exemplar for “st” is less than thelowest rated exemplar for either “d” or “sd”. Thus, the most important fact arising fromthese results is that both (s)[taj and [daj appear to be assimilated to the English category ofBest et al. (1988; Best, 1994) hypothesized that if two non-native phonemes areassimilated to the same native phonemic category and are judged to be equally goodexamples of the phoneme, discrimination will be more difficult than if the non-nativephonemes are judged to be different in quality. Since both allophonic variants in thepresent study are assimilated to “D” and they are judged as equally good members of “D”,discrimination in the subsequent experiments is expected to be poor.In addition, the lack of differences in goodness ratings suggests that adults will notshow evidence of an order effect in subsequent experiments. Recall, order effects wouldbe support for a perceptual magnet hypothesis if the exemplar judged as better acted as a39magnet by reducing perceptual distance among exemplars and resulting in decreased abilityto discriminate (Grieser & KuhI, 1989; KuhI, 1991; KuhI et al., 1992). If adults hadconsistently judged all (s)[tafs as worse examples of the category than all [da]s, I couldhave made the prediction that discrimination would be more difficult when [da] is thereferent and easier when (s)[ta] is the referent. However, with the high overlap in categoryratings and the non-significant difference between phonetic categories, discriminationshould not be better in one condition over another.This procedure did differ from the procedure used by Kuhl to obtain goodnessratings, however. Not only was Kuhl using vowels (and within category discrimination iseasier for vowels than for consonants), but she also had adults rate various within-vowelexemplars relative to a selection of written English vowels. In the present research, adultswere asked to rate various exemplars of consonants relative to one written consonant. It ispossible that even though this procedure was designed to increase sensitivity to withincategory phonetic differences, it still may not have been sensitive enough to elicitdifferential ratings. Thus, although these data do not predict a perceptual magnet effect, thenecessity of using a different procedure to obtain goodness ratings does raise the possibilitythat goodness differences are there, just not evident in this procedure. For this reason Iprobed for a perceptual magnet effect in the following experiments.The final reason for testing adults in this procedure was to ascertain if anyindividual signal was judged consistently different from other exemplars. This procedureallowed listeners to hear differences in quality between the exemplars when “sda” and “da”were page headings (albeit the differences did not coincide with categories). As mentionedabove, however, no single exemplar was judged consistently different for all pageheadings. In addition, the ratings of quality suggest that all exemplars fall within anacceptable range even when “sta” is the page heading and all exemplars can be used insubsequent experiments.40EXPERIMENT 2The AX Discrimination ProcedureThe second step in assessing adult perception of a language-specific phoneticdistinction is to determine if adults can discriminate [da] from (s)[ta] under relativelysensitive testing conditions. A task was chosen which is known to elicit within categoryphonetic perception. The AX (same-different) procedure has been shown to facilitatediscrimination of phonetic variation that is not phonemic (Carney et aL, 1977). In thisprocedure, adults are presented with a pair of speech sounds and are asked to say whetherthe two speech sounds are the same or different. As mentioned previously, ease ofdiscrimination can differ as a function of the ISI between the pair of stimuli. For example,when speech contrasts are presented with long 1Sfs (1500 ms), adults tend to discriminateonly if the contrast is phonemic. If the ISI’s are shorter, however, adults can discriminatewithin category phonetic variation (e.g., Pisoni & Tash, 1974; Werker & Logan, 1985).Thus, to facilitate discrimination of the allophonic distinction, adults were tested in the AXprocedure using a 500 ms ISI.The results from Experiment 1 suggest that for English adults the phone (s) [ta] isassimilated to the category “D” and that exemplars of both (s)[ta] and [da] are equally goodmembers of “D”. Since category goodness ratings for the two categories of phones did notdiffer and in fact exemplar ratings overlapped, it seems unlikely that order of presentationof stimuli will affect discrimination and thus no perceptual magnet effect is predicted.However, as mentioned in the discussion of Experiment 1, it is stifi of interest to probe fora order effects.It should be noted that the original work revealing a perceptual magnet effect (Kuhi,1991) was done using a category change procedure not a same/different discriminationtask. However the effect was partially replicated in recent research by Sussman and41Lauckner-Morano (1995) using an AAXX33 discrimination procedure with a 500 ms 1ST.Adults presented with vowels similar to Kuhl’s showed better within-categorydiscrimination when the non-prototype was presented in the “A” position, the referentposition. The similarity between an AX and an AAXX procedure would suggest that theAX procedure may also reveal an order effect if one is present. Thus, order was controlledto allow investigation of this possibility.A third question addressed in this study was whether the ability to discriminate [da]versus (s)[ta] improves or diminishes with practice. Previous evidence suggests thatrepeated exposure even without feedback may facilitate phonetic discrimination (e.g.,Werker & Logan, 1985) and this suggests that discrimination of native-language phoneswill increase across trials. To determine if there are practice effects on discrimination,adults were tested in two sessions.In summary, three questions were addressed in this experiment. First, are theEnglish phones [da] and (s)[ta] discriminable when adults are tested in a sensitiveprocedure? Second, is there evidence of a perceptual magnet effect for this consonantcontrast? Third, does performance change as a consequence of practice?MethodSubjects: Eighteen subjects between 18 and 25 years of age participated in this study.They were recruited as described in Experiment 1 and the selection criteria were identical.Two subjects’ data were excluded (one due to a self-reported hearing loss in one ear andone due to equipment error), resulting in a sample size of 16.This procedure involves presentation of two identical stimuli to be used as a referent and then twoidentical stimuli to be compared to the first.42Stimuli: The ten stimuli used in this study were the five (s) [ta] ‘s and the five [da] ‘5described in Experiment 1. The set of pairings included each individual exemplar pairedwith every other exemplar. This resulted in five different (DIFF) pairs for each exemplar(e.g., dal with (s)tal, (s)ta2, (s)ta3, (s)ta4, or (s)ta5) and 4 same (SAME) pairs (e.g., dalwith da2, da3, da4, or da5). Because physically identical pairings may alter the responsebias of a subject (Best, personal communication; Werker, 1993), no exemplar was pairedwith itself. Thus, one additional SAME contrast was randomly selected for each exemplarto equalize the numbers of SAME and DIFF pairs.Apparatus and Procedure: Presentation of contrasts was controlled by a 286 Compaqcomputer using the Bliss Program. Pairs were presented free field in an IAC soundattenuated chamber over a Bose speaker with a 500 ms 1ST and a 2000 ms responseinterval. During the response interval, subjects pressed buttons marked SAME or DllF ona mouse connected to the computer. The computer recorded all responses.Subjects were tested individually in three stages: a familiarization stage, and twotesting stages. To familiarize subjects with the procedure, they were presented with 16pairings alternating between two SAME trials and then two DIFF trials (in the same orderfor all subjects). Each subject was given a list of the correct responses and was told torespond on each trial by pressing the appropriate button. After this brief familiarization,testing began.Two rounds of testing were conducted with 100 trials in each set. Within each set,50 pairs began with a (s)[ta] and 50 began with a [da]. Of those 50, 25 were SAME pairsand 25 were DIFF (see Table 3). Each exemplar occurred 10 times in the “A” position. Toensure that no subject heard the pairs in the same order twice or in the same order asanother subject, the order of presentation of the 100 contrasts was randomized for eachround and for each subject.43Table 3: Experimental DesignSession 1 Session 2[da] first (s)[ta] first [da] first (s)[ta] firstSAME I DIFF SAME I DIFF SAME I DWF SAME I DIFF TOTAL25 25 25 25 25 25 25 25 200pairs pairs pairs pairs pairs pairs pairs pairs pairsTwo dependent variables were calculated for each sibject. A-Prime scores weregenerated by calculating the relationship between hits (correctly identifying a “different”pairing) and false alarms (incorrectly responding “different” to a “same” pairing). Theformula for A-Prime is .5+(H-FA)(l+H-FA)/[4H(1-FA)] where H=proportion of hits andFA=proportion of False Alarms (from Grier, 197 For each subject the computer alsocalculated overall percent of correct responses, percent correct when [da] was in firstposition, and percent correct when (s)[ta] was in first position.ResultsTo determine whether English listeners could discriminate the phones, a repeatedmeasures ANOVA was conducted using A-Prime scores. The repeated measures weresession of testing (first versus second) and order of pairings ([da] first versus (s)[taj first).There was a main effect for order, F91,15)=36.87, p<.000i, and an interaction betweenorder and session, F(l , 1 5)=7.24, p<.02. The main effect for order indicates that listenershad higher A-Prime scores when (s)[taj was in first position (M=.75, SD=. 12) than when[da] was in first position (M=.56, SD=. 15).34 A-Prime controls for the response bias of the subject by taking into consideration the frequency of falsealanns relative to the number of hits. It is a non-parametric statistic similar to D-Prime but unlike DPrime, the values are constrained to vary between 1 (perfect performance) and 0.5 (chance performance). Avalue below 0.5 is mathematically possible but suggests that the subject is systematically using some cueor strategy different than that required for the task.44Follow-up analyses of the interaction between order and session revealed asignificant difference between the first and the second session only when [da] was in firstposition, F(1,15)=8.68, p<.Ol (see Figure 5), but the magnitude of the order effect wasstill apparent even when taking into account the interaction. That is, althoughdiscrimination between [da] first pairings improved from the first to the second session,discrimination for [daj first pairings during the second session did not reach the levels ofaccuracy seen when (s)[taj was in first position.A-Prime scores in each of the conditions were compared to chance. Importantly,when (s) [ta] is in first position, the A-Prime scores are well above chance: Session 1,t(15)=10.44, p<.0001; Session 2, t(15)=7.01, p<.0001. Although adults tested with anative-language phonemic contrast would be expected to have close to 100% accuracy, thefact that A-Primes are over .74 when (s) [ta] is in first position is interesting indeed. Theresults differ dramatically when [da] is in initial position. In this condition, the A-Primescores are greater than chance only during the second session, t(15)=3.078, p<.O08(M=.60). Thus it appears that when [da] is in first position, discrimination is attenuated.Figure 5: A-Prime values for English listeners tested in an AX procedure.87D Session 1A-Prime Session 2ScoresChance 5(s)[ta] first [da] first45The direction of the order effect is consistent with predictions from the perceptualmagnet hypothesis: Discrimination is attenuated when the (putatively) better exemplar is infirst position. This result was surprising since adults in Experiment 1 perceived nodifferences in category goodness between the exemplars. When trying to understand thisinconsistency, it occurred to me that the order effect may be due to something other than aperceptual magnet effect so I decided to investigate the order effect further.In her original work, Kuhi used several vowels from within a single category (butsee Sussman & Lauckner-Morano, 1995, for evidence that the exemplars may not bejudged as coming from the same category). She suggested that the mechanism underlyingthe perceptual magnet effect is as follows: When a “good” exemplar from within a categoryis the referent, discrimination is attenuated because perceptual distance is reduced betweenthe good example and less good examples. Thus, stimuli compared to the good exemplarshould be judged the SAME more often than stimuli compared to the less good exemplar35.In the present case, since [da] first pairings showed reduced discriminability (appeared toact as the perceptual magnet), adults should be responding SAME more often to pairs with[da] first than with (s) [ta] first.Because this research used categories and not single exemplars, two kinds ofSAME responses are considered. The first type is when exemplars from within onecategory are compared to other exemplars from within that same category. In this case,SAME is the correct response (a correct rejection). The second kind is when exemplarsfrom within one category are compared to exemplars from another category (a different pairtype). In this case, SAME is an incorrect response (a miss). If a perceptual magnet effect35 Recall, Polka and Werker (1994) provided evidence that the perceptual magnet effect generalizes tobetween-category vowel perception (as opposed to within-category vowel perception). Polka and Werker didnot, however, assess whether the effect was due to a greater proportion of SAME responses when the“good” category was the referent.46is operating, this type should also have more SAME than DIFF responses (more missesthan hits) when a “good” exemplar (e.g. one of the [da] exemplars) is the background.Thus, it is important to control for the type of pairing in the next analysis.A repeated measures ANOVA was conducted comparing percent of SAMEresponses as a function of session (first versus second), order of pairings ([da] first versus(s)[ta] first) and pair type (same versus different). The results revealed a main effect forsession, F(l,15)=12.34, p<.003, a main effect for order F(1,15)=15.15, p<OO1 and amain effect for pair type, F(1,15)=33.40, p<.000i. There was a two-way interactionbetween order and pair type, F(l,15)=26.02, p<.000i. No other interactions reachedsignificance.The main effects are described first. The main effect for session indicates that thenumber of SAME responses declined from the first to the second session suggesting thatadults are detecting more differences among exemplars during the second session. Themain effect for pair type indicates that subjects responded SAME more to same pair types(M=59.38, SD=19.32) than to different pair types (M=36.63, SD=13.09). Thus adultscorrectly respond SAME more often than they incorrectly respond SAME. Finally, themain effect for order shows that subjects are significantly more likely to respond SAME to(s)[ta] first pairings (M=53.19, SD=22.61) than to [da] first pairings (M=42.81,SD=15.54). This effect is opposite to that predicted by the perceptual magnet effect.Furthennore, the percent of SAME responses is not significantly different from chance for(s)[ta] first pairings but is significantly less than chance for [da] first pairings.The two-way interaction between order and pair type also reveals a pattern of errorsinconsistent with those predicted by the perceptual magnet effect when it is extended todiscrimination of categories (see Figure 6). In reminder, the results from the A-Primeanalysis suggested that [da] acts as a perceptual magnet and therefore, when [da] is in firstposition, there should be greater SAME than DIFF responses in both same and differentpair types. In both cases, however, there were more DIFF responses. On the other hand,47when (s)[taj is in first position, SAIvIE responses are well above chance in same pair typesshowing the pattern that should occur for the “good” category. Finally, SAME responsesare below chance when (s)[taj is in first position and the pair types are different.Figure 6: Interaction between pair type and order of pairings.80 E] Same Pair types60Different pair typesChancePercent ofSAME 40_______responsesDiscussionThe results from this experiment suggest that English adults can discriminate (s)[ta]from [da]. When (s)[ta] is in first position, adults correctly discriminate well above chancelevels. That this is not an easy or straightforward task is evidenced by the fact that when[da] is in first position, performance is severely attenuated particularly during the first blockof 100 trials. Accuracy improves during the second session for those pairings with [da]first and reaches greater than chance levels of accuracy. Thus, discrimination is better thanchance in three of four conditions. These results provide evidence that when adults aretested in a sensitive procedure, they can discriminate (non-phonemic) native languagephones, and can do so even when one of the phones is presented in a non-standard context.This is a clear demonstration of language-specific phonetic processing in adults.20[1:I:ii(s) [ta] first pairs [da] first pairs48There is strong evidence of an order effect. When [da] is in the initial position,performance is attenuated and even though there is improvement during the second session,accuracy does not reach the levels of accuracy seen when (s)[ta] is in first position. Thedirection of the order effect is, at first blush, consistent with predictions from the perceptualmagnet effect (Kuhl et al., 1992). In contrast to predictions from the perceptual magnethypothesis, however, adults are less likely to respond SAME to [da] first pairings. Suchevidence suggests that the perceptual magnet effect cannot explain the order effects in thisstudy and some other mechanism must be responsible for the results.The results from the third analysis also suggest that (s) [ta] may form a moreinternally cohesive category than [da]. That is, when (s) [ta] is in first position and pairtype is the same, SAME responses (the correct response) are well above chance. On theother hand, when [da] is in first position and pair type is the same, SAME responses (thecorrect response) are not different from chance. This evidence suggests that discriminationof within category differences is difficult for (s) [ta] first pairings but not [da] first pairings:Adults do not differentiate between the individual exemplars of (s)[ta] but are differentiatingbetween exemplars of [da]. Thus, it appears that the (s)[ta] category has internalcohesiveness.One consequence of having a cohesive category may be that it makes discriminationof (s)[ta] from [da] easier when the cohesive category is the referent. Indeed, when (s)[ta]is first and pair type is different, SAME responses (the incorrect response) are belowchance: subjects are more likely to respond DIFF which is the correct response. On theother hand, when [da] is in first position and pair type is different, SAME responses (theincorrect response) are not different from chance. Thus, it is possible that adultdiscrimination is poor for [da] first pairings because the category is less cohesive making itdifficult to determine if there has been an exemplar change or a category change. Incontrast, when (s) [ta] is first, within category exemplar changes are not detected possiblymaking between category changes easier to detect. In summary, it is possible that (s)[taj49forms a more cohesive category than [da136 and therefore between category discriminationis easier for (s)[ta] first pairings than for [da] first pairings.36 The acoustic analyses reveals that (s)[taJ is less variable in FO than [da] even though there was overlapin the values.50EXPERIMENT 3The Category Change Procedure with AdultsThe first two experiments provide evidence that adults can discriminate thelanguage-specific phonetic distinction (s) [ta] versus [da] when tested in a sensitive AXprocedure. The next task is to test adults in a procedure that will allow comparison of adultdata and infant data, since this thesis is designed to investigate why 10- to 12-month-oldinfants fail to discriminate non-native contrasts. In their seminal work revealing the firstevidence of a developmental reorganization in non-native phonetic perception during thefirst year of life, Werker and Tees (1 984a) used a single procedure for testing both infantsand adults. Adults were tested in the category change procedure, a modified version of theconditioned head-turn procedure. Because the procedures differ in only minor ways, directcomparison of adult and infant data was possible. To enable similar comparison in thisthesis, adult perception of (s)[ta] and [da] was assessed using the category changeprocedure.By using a different procedure, I was also able to continue the investigation of themain question addressed in this thesis: Does exposure to native-language phoneticinformation maintain discrimination abilities, and if so, under what conditions doeslanguage-specific phonetic processing occur’? Previous research has shown that whenadults are tested in a category change procedure, they typically discriminate only nativelanguage phonemic contrasts. On the other hand, adults tested in an AX task with a shortISI demonstrate within category discrimination and even discrimination of some non-nativecontrasts. Researchers have concluded from this evidence that the category change task isless sensitive than an AX task.It is unclear whether the category change procedure elicits only native-languagephonemic perception, however. As mentioned in the Introduction, most previous researchhas confounded exposure and phonemic status. The stimuli used in the present research51were selected specifically to separate phonemic status from exposure to see if exposure issufficient for maintaining discriminability. If the phonemic status of a contrast is importantin adult speech perception and the category change procedure elicits only phonemicperception, adults should not discriminate (s)[ta] from [da] in this procedure. On the otherhand, if phonologically relevant phonetic information is in some way important in adultspeech perception, adults should discriminate these phones even when tested in thecategory change task.The category change procedure is also the procedure used by Kuhi et al. (1992) andPolka and Werker (1994) in which they showed evidence of a perceptual magnet effect forvowel perception. Thus, this procedure allows further investigation of whether the ordereffect apparent in Experiment 2 generalizes to a different procedure. Lack of an order effectin this procedure would provide even stronger evidence that the order effect seen in theExperiment 2 is due to a mechanism other than a perceptual magnet effect.MethodSubjects: Twelve English monolingual adults between 18 and 22 years of age participatedin the study. Recruitment was identical to Experiment 1 and selection criteria wereidentical. Two additional subjects were also tested in this procedure but their data arereported separately because they had both received extensive phonetics training prior toparticipating in the study.Stimuli. Apparatus. and Procedure: The stimuli were the ten exemplars describedpreviously, the 5 (s)[ta]’s and 5 [da]’s. The stimuli were presented on-line using a DataTranslation 2801A board. Phones were presented free field over a Bose speaker at 65 dB52with a 1500 ms ISI in the IAC sound attenuated booth. The Conditioned Head TurnProcedure was controlled by a 286 Compaq computer using custom software37.In the adult version of this task, subjects are repeatedly presented with variousexemplars from one phonetic category (called the referent or background). At randomintervals, three to five exemplars from a different phonetic category are presented andadults are asked to respond by raising their hand when they detect a change from thereferent to a different category.Subjects were tested individually. They were seated in the sound attenuatedchamber with a reinforcer box to their left. Subjects were presented with either (s)[ta] asthe background and [da] as the change stimuli or the reverse. The experimenter stoodoutside the chamber watching the subject through a one-way mirror. To initiate theprocedure, the experimenter pushed a button connected to the computer via a custom board.Presentation of the referent phones then began. When the experimenter decided that thesubject was ready (sitting quietly), the experimenter pushed another button to initiate a trial.During testing, the computer randomly selected change (experimental) or no-change(control) trials. Control trials occurred randomly on approximately 40% of the trials withthe constraint that no more than three consecutive control or experimental trials wouldoccur. Whenever a subject raised a hand to indicate perception of a change trial, theexperimenter pushed a response button and if a change trial had occurred, the computeractivated lights on the reinforcer box indicating to the subject that s/he was correct. A hitwas recorded by the computer. If there had been a no-change trial, no reinforcementoccurred and a false alarm was recorded. Other possible responses are a miss (when asubject fails to raise a hand during a change trial) and a correct rejection (when a subjectdoes not raise a hand on a control trial). There was no reinforcement for either a miss, acorrect rejection, or a false alarm although the absence of reinforcement for a false alarmcould indicate an error to the subject.‘ The custom software was designed by Kurt Golthardt, Nirvonics.53There were two phases to the testing procedure. During the training phase, subjectswere presented with 10 change trials to familiarize them with the procedure. During thisphase, only one exemplar from each category was presented. If subjects raised their handduring a change trial, the reinforcer was activated. If a subject failed to respond to a changetrial (a miss), the reinforcer was turned on automatically after the third presentation ofexemplars from the change category and two more presentations of the different exemplaroccurred while the light remained on. Then the background exemplar was presented again.At the end of 10 thals, the subjects were told that testing was about to begin. The intercomwas turned off so that the experimenter was unaware of whether a trial involved a changeor control trial.During testing, all five exemplars from each category were included in the set ofstimuli. Thus, subjects had to ignore within category variation and attend to betweencategory differences to be correct. The testing session involved 25 trials38. If a subjectfailed to identify three successive change trials, the computer notified the experimenter bybeeping and subjects were told that they had missed three change trials in a row and werenow going to be presented with three re-training trials. Retraining involved three changetrials identical to those described for the first phase, the initial training phase. Data from theretraining trials were not included in the analyses.Indicators of performance included percent of correct responses, A-Prime scores(see Experiment 2) and a preset criterion. The criterion was set as 7 out of 8 consecutivecorrect responses (see Experiment 4).ResultsTwo analyses were conducted to assess discrimination. They were both one wayANOVA’ s with the group variable being order of presentation ([da] referent versus (s) [ta]38 If a subject was within 2 trials of reaching criterion, a maximum additional 5 trials were presented.54referent). The dependent variables were either A-Prime scores or percent correct. Orderwas not significant in either analysis: the A-Primes were greater than chance when both[da] and (s)[ta] were the referents (M=.70, SD=.14: M=.74, SD=.12 respectively; seeFigure 7) and percent correct scores were better than chance in both orders ([da]M=63.653, SD=lO.66; (s)[ta] M=64.74, SD=8.92). It appears, therefore, that adults areable to discriminate (s) [ta] from [da] but that they do not show an order effect.Figure 7: A-Prime means for adults in the category change procedure.0.80.7A-PrimeScores0.60.5 Chance(s)[ta] referentThe two additional subjects who had received extensive training in phonetics,performed extremely well. One was correct on 22 out of the 25 trials and the other wascorrect on all of the trials. Such evidence suggests that there is a systematic and detectablephonetic difference between the English voiced unaspirated alveolar stop, [da], and theEnglish voiceless unaspirated alveolar stop, (s)[taj. To determine if these differences aresimilar to a phonemic contrast used in one of the world’s languages, it would be necessaryto test adults raised in a language environment which uses these phones phonemically39.Even though there was no order effect, I decided to conduct an analysis similar tothe post hoc analysis done in Experiment 2 to investigate the possibility that adults respondThe information provided by the acoustic analysis suggests that no language uses these precise phoneticforms phonemically.[da] referent55SAME more often in one order than in another. Recall, a perceptual magnet effect wouldpredict that when [da] is the referent, there should be more SAIvIE responses. If there is anorder effect showing greater proportion of SAME responses when (s) [ta] is the referentthan when [da] is the referent, it would replicate the results from the second study andprovide further support for the notion that adults are not subject to a magnet effect whentested with these consonants. As in Experiment 2, it is of interest to examine both correctSAME responses (correct rejections) and incorrect SAME responses (misses). In thismixed-model ANOVA, the group variable was order ([da] referent versus (s)[ta] referent)and the repeated measure was trial type (same versus different). The dependent variablewas proportion of SAME responses.There was a main effect for order, F(1,lO)=9.17, p<.Oi, revealing that adults weremore likely to respond SAME when (s)[ta] was the referent (M=.604, SD=.215) thanwhen [da] was the referent (M=.403, SD=.213). This pattern is identical to that revealed inExperiment 2. There was also a main effect for trial type, F(1,1O)=23.20, p<.0007,indicating that adults were more likely to respond SAME when the trial type was the same(correct rejections M=.65 1, SD=.202) than when the trial type was different (missesM=.357, SD=.160).Although there was no interaction between order and trial type, Figure 8 is brokendown by these two factors to allow comparison to the results from Experiment 2. As canbe seen in this figure, when (s)[ta] is the referent, correct rejections are well above chance(indicating that subjects were correctly responding SAME during these trial types) butmisses do not differ from chance. In contrast, when [da] is the referent, correct rejectionsdo not differ from chance but misses are well below chance (indicating that subjects werecorrectly responding DIFFERENT during these trial types). Thus, although the trend torespond SAME is in the identical direction in both referent conditions (thus no interaction),the errors occur for different reasons (see Figure 8).56PercentofSAI\’IEresponses30 -20-10-ChanceThis experiment revealed that adults show moderate discrimination of the nativelanguage phonetic distinction (s) [taj versus [da] when the phone following 1sf is removedfrom its standard context and presented in initial position. Recall, Repp and Lin (1989)presented the phonemes fbi and ipi following Is! and adults had difficulty discriminating thephonemes. The present research, however, shows that adults can discriminate a languagespecific phonetic distinction when both phones are presented in initial position. As will beelaborated in the General Discussion, these results provide support for the notion of alanguage-specific phonetic factor in adult speech perception (Werker & Pegg, 1992).It should be noted that adults do not perform as well as would be expected fornative phonemic contrasts. When tested with native phonemes, adults reach close to 100%Figure 8: Percent of SAME responses for (s)[ta] versus [da] referents.80 [ Same Trial Types70- • Different Trial Types60 -50-40 -0-(s) [ta] referent [da] referentDiscussion57accuracy. In the present research, adults reached between 63% and 64% accuracyreinforcing the a-priori hypothesis that English speakers would find this native phoneticdistinction difficult to discriminate.Unlike the results from Experiment 2, the results from the present experimentrevealed no difference in ease of discrimination as a function of order of presentation.Despite the fact that there was no effect of order on accuracy, the proportion of SAMEresponses differs as a function of which category is the referent. Adults are more likely torespond SAME when (s)[ta] is the referent than when [da] is the referent. This effect is inthe direction of that revealed in Experiment 2 and is opposite to predictions derived fromthe perceptual magnet hypothesis. This bias to respond SAIVIE more often in one conditionprovides support for the notion that (s)[ta] exemplars, the least likely to be good examplesof initial position alveolar stops, form a more cohesive category than [da] exemplars.The implications of these results for the perceptual assimilation model of Best (Bestet a!., 1988; Best, 1994) are that discrimination of language-specific phones is notnecessarily affected by the degree to which they are assimilated to the same nativephonemic category. Experiment 1 revealed that adults judge both categories of phones tobe good examples of “D” and therefore both phones are assimilated to one phonemiccategory. According to the perceptual assimilation model, when two non-native phonemesare assimilated to one native category with no perceived differences in category goodness,discrimination should be very difficult. Yet in this experiment, discrimination of the nativephones is moderately good (63 to 64% correct). Thus, although the perceptual assimilationmodel may still explain differences in ease of discrimination for non-native phonemes, themodel does not generalize to perception of native language phonetic differences.Interestingly, this also suggests that native phones differ in important ways fromnon-native phonemes. Recall, Polka (1991) tested adults with non-native phonemes thatdiffered in native allophonic status but found that discrimination did not follow thehypothesized pattern. One possible explanation for her results is that even if the non-native58phonemes appear to be similar to native allophones and are transcribed identically to anative allophone, the non-native phonemes may differ in significant ways from nativephones. Thus, there is not necessarily a relationship between ease of discrimination of anon-native phonemic contrast (even if it has exemplars similar to native allophones) and anative phonetic distinction. Indeed these results suggest that adults do discriminate a nativephonetic distinction.In summary, exposure without native phonemic status appears to influence adultspeech perception by maintaining (or reinstating) their ability to discriminate language-specific phonetic variation, albeit with less ease than is shown for native phonemiccontrasts. The question of whether this adult pattern is best described as maintenance orrecovery can only be assessed after testing infants with the same distinction. This wasdone in the next experiment.59EXPERIMENT 4:Conditioned Headturn Procedure with InfantsThe final experiment in this thesis was designed to assess infant perception of theEnglish phones and compare the infant data to the adult data. To do this, infants weretested in the Conditioned Head Turn Procedure, the procedure used by Werker and Tees(1984a) to reveal the developmental reorganization. It was not known a-priori if infants of6- to 8-months of age could discriminate this phonetic distinction. On the one hand,previous evidence suggests that young infants can discriminate almost every contrast withwhich they have been presented. On the other hand, if the broad-based sensitivities shownby very young infants reflect sensitivity to the set of contrasts used as phonemes in theworld’s languages and if this particular combination of phonetic cues does not correspondto a phonemic contrast in any of the world’s languages, then young infants may fail thistask.The questions are slightly different for older infants. It is known that by 10 to 12months infant performance diminishes for non-native phonemic contrasts. Whether thisshift in speech perception extends to (non-phonemic) language-specific phonetic perceptionis not yet known. As mentioned previously, absence of exposure to a speech contrast doesnot always lead to decreased ability in 10- to 12-month-old infants, as infants this age andolder discriminate the Zulu click contrast (Best et al., 1988). What we do not yet know iswhether 10- to 12-month-old infants will discriminate speech segments to which they areexposed but which are not phonemic in the native language. The variants [da] and (s)[ta]allow us to test this possibility since these phones are part of the native language yet are notphonemically contrastive.Experiments 2 and 3 show that adults can discriminate (s)[taj versus [dal, but thisdoes not necessarily mean that infants wifi be able to discriminate these phones. Adultshave been shown to easily discriminate non-native German vowels contrasts but 10- to 12-60month-old English learning infants find discrimination of these same German vowelsdifficult (Polka & Werker, 1994). Similarly, Best et al. (in press) revealed that althoughadults can discriminate a voiced versus voiceless lateral fricative, <ca> <xa>, and a velarvoiceless aspirated /id versus an ejective 1k’! both from the Zulu language, 10- to 12-month-old infants only discriminate the first contrast.In a recent study, Christophe, Dupoux, Bertoncini, and Mehler (1994) examinedneonates’ ability to discriminate bisyllabic stimuli that either came from two words andcrossed a word boundary, or were from within the same word. For example, mati in“panorama typique” crosses the word boundary and mati in “mathematicien” does not.The results revealed that neonates could discriminate these two utterances. Such research isconsistent with other evidence (e.g., Hohne & Jusczyk, 1994) suggesting that younginfants can attend to (non-phonemic) phonetic information. Whether 10- to 12-month-oldinfants attend to such phonetic infonnation when it presented out of context is stillunknown.The results from this study will help to inform theory concerning the factorsunderlying the developmental reorganization in speech perception. If both younger andolder infants discriminate phonetic variants, it would suggest that initial sensitivity tophones includes sensitivity to this English phonetic distinction and that linguistic exposuredevoid of meaning is adequate to maintain discrimination. If on the other hand, youngerinfants discriminate but older infants fail to discriminate this distinction, support would begained for the hypothesis that rather than linguistic exposure, it is the phonemic status thatis important in maintaining infant discrimination. This result would lend credence to theview that the developmental re-organization is based on phonemic principles.Polka and Werker (1993a) have hypothesized that the change in speech perceptionmay first be evident as a perceptual magnet effect, at least with respect to vowels but not befully realized until 10 to 12 months (but see Polka & Bohn, 1994). If the perceptualmagnet effect extends to infant language-specific phonetic perception of consonants,61younger but not older infants should show an order effect. However, the order effectapparent with adults in Experiment 2 is potentially due to a phenomenon other than aperceptual magnet effect because they did not respond SAME more often to [da] first than(s)[ta] first pairings. In addition, adults did not evidence a order effect in Experiment 3.For these reasons, I do not predict that infants of either age will show evidence of aperceptual magnet effect in the present experiment.MethodSubjects: English-learning infants were recruited in several ways. Most parents whosebabies participated were initially contacted by a research assistant from our lab who visitedpost-partum mothers in the hospital. The research assistant described the type of studiesdone in our lab in general terms and asked mothers if they were interested in participating inour studies. If the mother indicated interest, she provided her address and phone numberand gave permission to be contacted at a later date. When the baby was the appropriateage, parents were contacted by phone and given detailed information about this study. Ifthey were still interested, an appointment was made. Other infants became subjects whentheir parents contacted the lab after seeing our posters or ads or after hearing advertisementswe had broadcast on local radio stations. Finally, some mothers and infants were recruitedby parents who had participated in our research. These mothers called us to makeappointments. All infants were healthy, born within 2-weeks of due date, from Englishspeaking homes, and not exposed to more than an estimated 10% of any other language.They were given a T-shirt and a certificate after participating.Data from 32 infants were used: 20 English-learning infants aged 6- to 8- months(6 females, 14 males, mean age = 6 months and 28 days) and 12 English-learning infantsaged 10- to 12-months (6 females, 6 males, mean age=10 months 15 days). An additional6230 infants aged 6- to 8- months were excluded from the study due to failure to condition tothe phonemic contrast (see next paragraph) on the first visit (15), failure to recondition tothe phonemic contrast on the second visit (4), crying or iii (6), equipment error (2) andfailure to return for the second visit (3). Seventeen additional older infants were excludeddue to failure to condition to the phonemic contrast on the first visit (9), failure torecondition to the phonemic contrast on the second visit (2), cried or iii (5) and equipmenterror (1)40.Stimuli: The stimuli used in the study consisted of the five exemplars of (s)[ta] and the fiveexemplars of [dai previously described. Since it was necessary to ensure that infants canand will perform in this task, infants were initially tested using an English phonemiccontrast. We selected the phonemic distinction [thai versus [da] as a control because thiscontrast is acoustically and phonetically similar to the phonetic distinction (s)[ta] versus[da] (all are alveolar stops). This would ensure at least some measure of comparabilitybetween the control and experimental segments41. Thus, in addition to the five (s) [ta] andthe 5 [da] exemplars used in the adult studies, the five exemplars of the phoneme [thaidescribed in Experiment 1 were included in the set of stimuli.Apparatus and Procedure: The testing was conducted in a sound attenuated chamber inwhich there were two chairs facing each other, a small table between the chairs, and areinforcer box (speakers below) situated to the left of the infant. An infant was seated onthe parent’s lap and one experimenter (El) sat in front of the infant. Without speaking, Elshowed the infant noiseless toys during the procedure to entertain the infant and to prevent40 After testing half of the infants, we were concerned about the high number of infants failing tocondition on the first visit. In an attempt to reduce this attrition, we changed the experimental team andmoved the reinforcer to the left of the infant. Several pilot infants were tested during this time. Thesechanges reduced attrition slightly.41 The high number of infants failing to condition on this contrast on the first visit suggests that [thaiversus [dai may be somewhat more difficult for infants than other phonemic contrasts on which they havebeen tested.63the infant from constantly checking the reinforcers. The parent was instructed to sit quietlyand if needed, reassure the infant with a smile or a nod. Both the parent and El wore headphones with loud vocal music playing to mask the stimuli being presented to the infant.Thus neither the parent nor El were aware of the signals being presented to the infant. Asecond experimenter (E2) was outside the chamber and watched the infant through a one-way mirror. E2 communicated with the computer via button presses.As in the adult procedure, infants in the Conditioned Head Turn Procedure arepresented with a repeating background signal with a 1500 ms ISI. At random intervals, thespeech sound changes from the background signal to the target signal. Whereas adultsindicate discrimination by raising their hand, infants are conditioned to turn their headtoward the visual reinforcer when they hear a change in the speech signal. When the infantperforms a head-turn, E2 pushes a button and if there was a change trial, one or more ofthe boxes in the visual reinforcer is activated. In addition, El gives verbal praise to theinfant.Infants are tested in three stages: a familiarization stage, a conditioning stage andone testing phase. During the first two stages, infants are presented with only oneexemplar from each the two categories of speech sounds. During the testing stage, infantsare presented with several exemplars and thus must attend to categorical differences amongthe speech signals.The first stage is to familiarize the infant with the sounds and the reinforcers andinvolves three to five change trials. To begin, E2 pushes a button to tell the computer topresent the repeating background phone. When E2 judges that the infant is ready (lookingtowards El and appears to be listening to the speech sounds) she pushes another buttonwhich tells the computer to present a change trial. The background sound changes to threepresentations of the target signal. The visual reinforcer is automatically activated by thecomputer after the first change signal is presented and remains on for the duration of the64three stimuli. El directs the infant attention towards the activated animal(s) if needed andgives verbal praise.During the second stage, the infant is taught the contingency between a change inthe speech signal and the activation of the visual reinforcers. To condition the infant, E2delays the activation of the reinforcers to provide an opportunity for the infant to make ahead-turn prior to their activation. In the event that the infant does not make a head-turn,E2 activates the reinforcers on the third change trial and two additional presentations of thechange stimulus accompany the duration of the reinforcers. When an infant has correctlyperformed three consecutive head-turns or the number of trials reaches a maximum of 15,the second stage of testing ends.The third stage is the testing stage during which E2 turns off the intercom so thatshe cannot hear if there is a change trial or a control trial. E2 pushes a button for anobservation interval and the computer randomly selects control versus experimental trials.Control trials occur on approximately 40% of the trials with the restriction that no morethan three consecutive control or 3 consecutive change trials occur. Again, E2 pushes abutton if the infant makes a head-turn and the computer activates the reinforcers only ifthere has been a change trial. If an infant fails to respond on three successive change trials,retraining occurs in which the infant is presented with only single exemplars. During theseretraining trials, if no head-turn occurs the reinforcer is automatically activated on the thirdchange stimulus and 2 more presentation of the change stimulus are presented. After thethree retraining trials, the program automatically returns to testing. Each infant is limited toa maximum of two retraining sets and these trials are not included in the analyses. Thetesting stage includes no less than 25 and no more than 30 trials. This number of trials hasbeen shown to be optimal for determining whether infants can discriminate and avoids thepitfall of determining what they can be trained to discriminate if given more trials (e.g.,Werker & Lalonde, 1988).65Infants were tested on two days: the first day they were presented with the Englishphonemes [thai versus [da] to ensure that they would perform in the task. Infants whofailed to show evidence of discriminating [thai from [dai were not tested on a second day.Those infants who successfully discriminated the phonemes [thai versus [da] on the firstday were presented with the phones (s)[ta] and [da] on the second day. If they failed todiscriminate between these phones they were re-tested with the [tha] versus [da] phonemiccontrast to ensure that their failure was not due to forgetting the task. If infants failed torecondition on the phonemic contrast, their data were not included in the analyses42.To allow for the possibility that infants would demonstrate a perceptual magneteffect (Kuhl et al., 1992), order of presentation of the speech sounds was counterbalancedboth for the control condition on the first visit and the experimental condition on the secondvisit. Thus, there were four possible orders for each age group (see Table 4).Table 4: Experimental Design for infants in Head Turn ProcedureAge Control (Day 1) n Experimental (Day 2) N[daj to [thaI 4 [da] to (s)[ta] 106-8 [thaI to [dai 6Months[dai to [thai 4 (s)[tai to [da] 10[thai to [dai 6[thai to [dai 3 [da] to (s)[ta] 610-12 [dai to [thaIMonths[thai to [da] 3 (s)[taJ to [daJ 6[dai to [thai42 Data from two infants in the 10- to 12-month age range who failed to recondition were included in theanalyses. These infants cried before criterion could be reached during the control session on day two but hadresponded correctly to five out of six trials before becoming fussy.66In a head-turn procedure, several indicators of performance are possible. First, Iexamined the data to determine the number of infants reaching a criterion of 7 correctresponses out of 8 consecutive trials in a total of 25 trials. The probability of an infantreaching this criterion in 25 trials is between p=.O5 and p=.OO1 (Werker & Lalonde, 1988).In addition, A-Prime scores were calculated for each infant (see Experiment 2 for formula).Finally, overall percent correct scores to both change and control trials were calculated foreach infant.ResultsTo allow comparison of 6- to 8- versus 10- to 12-month-old infants, an Analysis ofProportions43 (ANPRO, Marascuilo, 1966) was conducted comparing the number ofinfants reaching criterion in each group. This analysis revealed that there were significantlymore 6- to 8-month-old infants reaching criterion than 10- to 12-month-old infants,2=1 1.62, p<.003. There was no effect of order of presentation. Whereas 11 of the 20younger infants reached criterion (6 when [da] was the referent and 5 when (s) [ta] was thereferent), only 1 of the 12 older infants did so (and this was when [da] was the referent).To compare the infant to the adult data, a second ANPRO was conducted includingadults, as well as 6- to 8- and 10- to 12-month-old infants. This analysis revealed asignificant difference between the proportion of adults and the proportion of 10- to 12-month-old infants reaching criterion,X2=13.67, p<.OOi, indicating that discrimination ofthe phones was attenuated in 10- to 12-month-old infants (see Figure 9). The differencebetween the proportion of adults and the proportion of 6- to 8-month-old infants reachingcriterion was not significant.3 Analysis of Proportions is a non-parametric analog of Chi-square that allows comparison of unequal n’sand does not require a minimum number in each cell.67Figure 9: Proportion of adults and infants reaching criterion.0.80.6ProportionReachingCntenonAdults 10-12 monthsBefore conducting an analysis using the A-Primes, I inspected the infant’s A-Primevalues to ensure that no A-Prime values were extremely low. As mentioned previously, A-Prime values should fall between 1 and .5. Although it is mathematically possible to scorebelow .5, if the score is too low it suggests that the infant is consistently using a strategydifferent from that required for the task. three infants had low A-Primes; one older infant(A’=.09) and two younger infants, one when (s)[ta] was the referent (A’=.26) and onewhen [da] was the referent (A’=. 15). All other A-Primes were over .30. Data from thethree infants with very low A-Primes were dropped from subsequent analyses.After excluding the three infants, I conducted the 3 (age) X 2 (order) ANOVA usingA-Prime as the dependent variable. In this analysis, there was a significant main effect forage, F(2,35)=6.89, pc003. Order was not significant and there was no significantinteraction. The follow-up comparison of the main effect revealed that older infant APrimes were significantly lower than younger infant A-Primes, F(1)=7.29, p<.0i, andthan adult A-Primes, F(l)= 19.34, pc0003, but there was no difference between youngerinfants and adults (see Figure 10). This result parallels the results from the ANPRO and6-8 months68provides further support for the assertion that younger infants and adults perform equallywell at discriminating (s)[ta] from [da].Figure 10: A-Primes for adults, young infants and old infants80•70Ix6050_______Chance10-12 MonthsI also conducted an analysis to determine if infants revealed a greater proportion ofSAME responses in the [da] referent than in the (s) [ta] referent condition. This analysis issimilar to the final analysis in Experiment 3 conducted with adults. Recall, if infantsresponded SAME more often when (s)[taj is the referent than when [da] is the referent, itwould reveal a pattern consistent with the adult data from Experiment 2 and 3 and a patterninconsistent with predictions from the perceptual magnet hypothesis.In this mixed-model ANOVA, the group variables were age (old versus younginfants), order ([da] referent versus (s)[ta] referent) and trial type (same=correct rejectionsversus different=misses) and the dependent variable was number of SAME responses.There were no main effects and there was no interaction between the two factors indicatingthat neither age group of infants shows any trend to respond SAME more often in oneorder over another.Adults 6-8 Months69DiscussionThe results from this study reveal that significantly more 6- to 8- than 10- to 12-month-old infants discriminate the phones (s)[ta] and [da] and the A-Primes aresignificantly better than chance for younger but not older infants. This evidence of adevelopmental change in phonetic perception during the the first year of life supportsprevious cross-language research. First, previous research has shown that young infantsdiscriminate most speech distinctions with which they are presented. This early ability hasbeen termed broad-based sensitivity because it appears to involve discrimination of theworld’s phonemic inventory. Because (s) [ta] versus [da] is an English phonetic distinctionand may not be similar to any phonemic contrast, these results suggest that young infantsalso discriminate within-language phonetic variation.More importantly, older infants fail to discriminate the English phonetic distinction(s)[ta] versus [da]. Recall, the phones used in this research are not native phonemes andyet do occur in the linguistic input of these English infants. Furthermore, the exposureinvolves a regular and predictable phonetic alternation, the context-dependent allophone(s)[ta], and a commonly occurring initial position stop [da]. Thus in this research,phonemic status and exposure were controlled allowing assessment of the question ofwhether the developmental reorganization is due to phonemic status or due to exposure.On the basis of the evidence from this experiment, I can conclude that the developmentalreorganization in speech perception occurring at 10 to 12 months is not a function oflinguistic exposure per Se. Furthermore, the results are consistent with the notion that theshift is based on the phonemic status of the contrast.There is no evidence of an order effect indicating that neither younger nor olderinfants are subject to a perceptual magnet effect when tested with these phones. Furthersupport for the lack of a perceptual magnet effect arises from the analysis of SAMEresponses: neither age group of infants showed a tendency to respond SAME more often70in one condition than the other. This result differs from that with adults who respondedSAME more often during the (s)[taj referent condition even though there was no evidenceof a perceptual magnet effect and suggests that exposure to the native phones after infancyleads to a bias in perception opposite to that predicted by the perceptual magnet effect.The fact that adults and 6- to 8-month-old infants can discriminate this contrast but10- to 12-month-old infants cannot means that adults regain the ability to discriminate anative phonetic distinction. To determine at what point in development this shift occurs, itwould be necessary to assess discrimination in children of several ages.In summary, the difference in discrimination among older and younger infantssupports previous research indicating that the linguistic environment has an effect onphonetic perception by the end of the first year of life. The present research goes one stepfurther, however, by illustrating that older infants do not discriminate even a nativephonetic distinction to which they are exposed but which is not part of the native phonemicinventory. Thus, perception is modified not simply by exposure to the native language butrather relates in important ways to the phonemic inventory of the native language.71GENERAL DISCUSSIONThis series of experiments addressed three main questions. First, do adults attendto (non-phonemic) native language phonetic information presented out of a meaningfulcontext and if so, under what circumstances? Second, does non-phonemic linguisticexposure to language specific phonetic information maintain discrimination in adults?Third, is it linguistic exposure per se or phonemic status that accounts for thereorganization in phonetic perception between 6 and 12 months of age? In addition, theresearch allowed elaboration of three approaches to understanding adult and infant phoneticperception, the perceptual assimilation model (Best et al., 1988; Best, 1994), theperceptual magnet effect (Kuhl, 1991; Kuhl et al., 1992; Polka & Werker, 1994), and thefour-factor model (Werker & Pegg, 1992). In short, the purpose of these studies was toinvestigate adult and infant perception of a native-language phonetic distinction and tofurther elucidate our understanding of the developmental reorganization in speechperception.These studies revealed that adults judge all exemplars of both (s)[ta] and [da]categories to be better examples of a ‘D” than a “T” and there was no indication that adultsconsidered exemplars of (s)[ta] as less good members of the “D” category than exemplarsof [da] (Experiment 1). Nevertheless, when English adults were tested in an AX task with500 ms 151 (Experiment 2), they discriminated (s)[ta] versus [da] better than chance inthree of four conditions. Even when tested in a category change procedure (Experiment 3),adults discriminated this English phonetic distinction. Even though adults clearlyassimilated both phones to the same phonemic category, they were able to detect theacoustic/phonetic difference between them. This shows that it is not necessary for acontrast to be phonemic in order for adults to be able to discriminate it.The last study in this series revealed that a significant number of adults and 6- to 8-month-old infants could discriminate the phones when tested in the Conditioned Head Turn72Procedure whereas infants of 10 to 12 months of age could not discriminate the phones(Experiment 4). Furthermore, mean A-Primes for younger infants and adults did not differbut both were significantly greater than mean A-Primes for older infants. Only the mean A-Primes for the 6- to 8-month-old infants and the adults were greater than chance. Thisevidence suggests that the developmental reorganization in infant speech perceptionoccurring at the end of the first year of life (Werker & Tees, I 984a) is not governed solelyby linguistic exposure but rather appears to be influenced by the native language phonemicstatus of the contrast. Infants raised in English environments are exposed to the nativelanguage phone (s)[ta] as well as to the phone [da], but this exposure does not involve acontrast in meaning. The finding that 10- to 12-month-old infants cannot discriminate(s)[taj from [da] shows that simple exposure to a speech segment cannot account formaintenance of the full ability to discriminate a speech contrast.This finding is particularly noteworthy since the (s)[taj versus [da] distinction is anintegral part of English language phonology: It involves a context dependent allophone.Recall that context dependent allophones are phonologically relevant in that they aremandatory alternations in production. When an alveolar stop follows an Is!, it must alwaysbe produced as [ta] and never as [thai but when an alveolar stop occurs in initial position itcan never be [ta] but rather must be [daj or [thai. Thus, the developmental change inphonetic perception appears to be reorganized in accordance with the phonemic status of thecontrast and not simply the phonological relevance of a phonetic variation.When Best et al. (1988; Best, 1994) proposed the perceptual assimilation model,created to explain differences in discrimination of non-native phonemic contrasts, theysuggested that ease of discrimination may differ as a function of the degree to which a nonnative phonemic contrast is assimilated to a native language phonemic category. Further,Best (1994) proposed that when both non-native phonemes are judged as equally goodmembers of the same native category, discrimination will be poor. The evidence presentedhere reveals that although adults judge both native language phones as equally good73members of the same native phonemic category, they discriminate the phones moderatelywell. This raises the possibility that the perceptual assimilation model cannot begeneralized to native language phonetic perception without modification.With regard to the perceptual magnet effect, these studies suggest that the perceptualmagnet effect does not extend to native language consonant perception. Although adultstested in the AX procedure discriminated better when (s) [ta] was in first position than when[da] was in first position, and although this order effect was in a direction consistent withthe perceptual magnet effect in vowel perception (Kuhi et al., 1992; Polka & Werker,1994), three pieces of evidence suggest that the order effect may be due to a differentphenomenon. First, Kuhl demonstrated that vowels judged as more prototypical of thecategory acted like perceptual magnets. In the current experiments, there was no indicationof differences in judgments of category goodness for (s)[ta] versus [da].The second reason for arguing against a perceptual magnet effect was alsomentioned previously: If the perceptual magnet effect explained these results, it wouldmean that adults perceive pairings beginning with [da] as the SAME more often thanpairings beginning with (s)[ta]. Contrary to this, adults responded SAME less often when[dal was in first position. This effect was independent of whether the pairs were from thesame or different categories. In addition, when [da] was first and the pairs were the same,adults made significantly more errors potentially because they were detecting withincategory differences.Third, when tested in the category change procedure in Experiment 3 (the procedureused by Kuhi to reveal a perceptual magnet effect), adults discriminated the (s)[ta] versus[da] distinction moderately well regardless of order and yet still demonstrated a tendency torespond SAME more often to (s)[ta] first than [daj first pairings. In summary, the resultsfrom these three studies indicate that the order effect apparent in the AX procedure is notdue to a perceptual magnet effect but to some other phenomenon. This leads to the74conclusion that the perceptual magnet effect may not generalize to native languageconsonant perception.Possible explanations for the order effect may be found in the old literature on ordereffects reviewed by Warren (1985). Warren (1985) reviewed previous researchconcerning a perceptual phenomenon called the criterion shift rule. This rule states thatrepeated presentations of stimuli from one end of a perceptual continuum results in a shiftin the criterion used to determine if a stimulus from the other end is the same or different.The direction of the shift depends on several factors such as modality being tested, ISI,duration of exposure during the experiment, amount of training, memory for thecontinuum, the order of presentation, and the range of exemplars presented (Warren citesreviews by Hellstrom, 1985; McKenna, 1984).Asymmetrical biases (order effects) are also apparent and could be due to thedifferential influence of assimilation versus contrast effects. Assimilation is described asthe tendency of subjects to label boundary exemplars as the same as the referent categorywhile contrast is the tendency of subjects to perceive boundary exemplars as different.Extending these terms to the present research indicates that when (s) [ta] is the referent,assimilation occurs and when [da] is the referent, contrast effects occur. As mentioned inthe general introduction, Sawush and Jusczyk (1981) found results of contrast effects forboth /spal and /ba/ adapters, an effect opposite to the perceptual magnet effect and similar tothe [da] referent effect. In addition, Trehub et al. (1990) showed evidence of contrasteffects for good Western melodies and assimilation for poor Western melodies. PerhapsWarren’s criterion shift rule thus provides a better account for the range of phenomena thatoccur than the perceptual magnet hypothesis.The four-factor model of Werker and Pegg (1992) was designed to describe thepattern of developmental changes in phonetic perception. Young infants’ abilities aredescribed as revealing a broad-based phonetic factor because young infants discriminatemany contrasts that do not hold phonemic status in their native language. To date, virtually75all studies tested infants on contrasts that are phonemic in a language other than the nativelanguage. This thesis research provides the first evidence that young infants can alsodiscriminate native phonetic distinctions that are possibly never used phonemically in anylanguage. This broadens our understanding of the language-general phonetic factor.The original description of the language-general phonetic factor in young infantswould predict that virtually all younger infants should discriminate any contrast on whichthey were tested. Thus, even though infants aged 6 to 8 months performed as well asadults on the (s)[ta]-[da] distinction, it is important to note that several of the youngerinfants failed to discriminate (s) [ta] from [da]. This raises the possibilities that either the(s)[ta] versus [da] distinction does not correspond to a phonemic contrast in any languageor that the linguistic environment has already had an effect on perception by 6 to 8 monthsof age (as shown by Polka & Werker for vowel perception, 1994). To assess the firstpossibility, it would be necessary to assess perception of (s)[ta] versus [da] in adults andinfants from a language environment that uses a voiced/voiceless distinction phonemically.To assess the second possibility, it would be necessary to test infants under 6 months todetermine if they easily discriminate this English phonetic distinction.Werker and Pegg (1992) described the ability of 10- to 12-month-old infants asrevealing a language-specific phonetic factor because it is not yet fully phonemic. Recall,research reviewed in the introduction has shown that 10- to 12-month-old infants candiscriminate native but not non-native phonemes. In addition, there is growing evidencethat infants do not use the new categories in a contrastive way, that is in a fully phonemicfashion, until at least several months later (e.g., Stager & Werker, 1995; Werker &Baldwin, 1991; Werker & Pegg, 1992). It was for these reasons that Werker and Peggintroduced the term language-specific phonetic.The evidence from the present research provides further understanding of 10- to 12-month-old infant’s abilities by revealing that older infants do not discriminate even nativelanguage phones. Strictly speaking, therefore, 10- to 12-month-old infants’ speech76perception does not include language-specific phonetic perception. Perhaps a different termwould help to reflect this new evidence. A term like pre-.phonemic includes the notion of aphonemic bias in perception but also captures the idea that older infants’ perceptualcategories for speech are not fully phonemic because they are not yet functionallyphonemic. This term also avoids the implication that older infants might discriminate nativephonetic categories. This research, therefore, has provided more definition of the speechperception capabilities of 10- to 12-month-old infants.The term language-specific phonetic is still useful to describe adult speechperception. Although Werker and Pegg used the term language-specific phonetic todescribe 10- to 12-month-old infants’ declining discrimination and suggested that adultsmay also show evidence of a similar factor, they did not suggest that language-specificphonetic perception would be identical in infants and adults. Indeed, the evidence providedin these experiments suggests that there are significant differences between adult and 10- to12-month-old infant speech perception capabilities with respect to native phoneticinformation. Whereas adults show moderate discrimination of native language phoneticvariation in this set of experiments, older infants do not. These differences between adultand 10- to 12-month-old infant perception still need further investigation. One possibilitywould be to test children of various ages to determine when in development native-language phonetic distinctions become discriminable. Nevertheless, the language-specificphonetic factor still accurately describes adult speech perception.None of the above approaches provide an adequate explanation of why or howinfants undergo a reorganization in phonetic perception during their first year. Onesuggestion that has been proposed by some researchers is that infants develop the cognitiveabilities to form functionally based categories toward the end of the first year of life(Lalonde & Werker, in press; Morgan & Saffran, 1995). By this age, but not before, theinfant can discern relevant versus irrelevant variability in stimulus input and pay selective77attention to those differences that are criterial in defining a category. But how do infants“know” which variability is criterial to phonemic categories?This problem is in some ways similar to the problem of invariance which hasplagued speech perception theories for many years. Simply stated, it is difficult tounderstand how people come to identify a phonetic variant as belonging to specificphonemic category because there are no invariant acoustic or phonetic cues to phonemes.Some speech perception theories include the notion of a lexicon impacting on perception(e.g., TRACE Theory, McClelland & Elman, 1986). Such “top down” processingconstrains perception thus making perception of a phoneme possible. However, in infantperception, this view is not an option because infants’ lexicons are poorly formed, ifformed at all.There is a theory in the early stages of development that may help to provide aframework for understanding the evidence herein. This theory has been proposed toexplain the developmental changes in the more general aspects of speech perception (ratherthan being specific to the development of phonetic perception as the previous approacheswere). Nevertheless, this model can provide a heuristic with which to develop hypothesesconcerning the developmental reorganization in phonetic perception. This model isJusczyk’s (1993) Word Recognition and Phonetic Structure Acquisition (WRAPSA)model.Jusczyk’s model rests on the assumption that processing of language must be rapidand efficient because the speech stream is produced very rapidly. The model outlines oneway in which such efficient processing might develop44. The central component ofWRAPSA is of interest to this thesis. This component involves the developmentalformation of a weighted processing scheme (for details see Jusczyk, 1993).44 Davis (1992) has proposed a similar model which has been developed on the basis of productionevidence. The main difference between the models is that Jusczyk uses the syllable as the processedsegment and Davis uses the phoneme.78Jusczyk argues that early in life, speech processing is language general in that allpotential sources of information have equal valence because weighting has not yet takenplace. With exposure, the speech analyzers become weighted to reflect the distributionalproperties of various features in the input. Attention is focused on the most commonlyoccurring properties (and/or the most salient properties) of the native language whilefeatures that are not used in the native language are ignored. Weighting occurs as aconsequence of attention. Furthermore, Jusczyk suggests that the various components ofspeech are weighted sequentially. For example, the native language prosodic structures(stress, rhythms, and intonation) may be the first components of the native language tobecome perceptually weighted45.As this native prosodic information becomes weighted, itis processed more quickly and more efficiently. This, in turn, allows the infant to attend toanother level of detail in the native language such as clausal boundaries. Attention, then,moves from a focus on the more global aspects of the native language to a focus on fmerand/or more specific features.Jusczyk is careful to point out that just because attention is tuned to more globalfeatures early in life, does not mean that infants are unable to detect differences among fmerfeatures of a language. Indeed, as reviewed previously, young infants can discriminatemost contrasts whether they are native or non-native. My understanding of WRAPSA isthat as the weighting scheme develops, certain features of the native language become more“automatically” processed. That is, as each feature of the native language becomesperceptually weighted, the processing of these features becomes more efficient andattention can be focused on the next level of detail in turn, re-tuning perception at that level.‘ This weighting of native language prosodic information may occur prenatally since much auditoryinformation is available to the fetus (e.g., Fifer & Moon, 1988) or with minimal exposure postnatally(e.g., Mehler, Jusczyk, Lambertz, Haistead, Bertoncini, & Amiel-Tison, 1988) since infants show apreference for native language prosody just a few days after birth (Moon, Cooper, & Fifer, 1993). Also seeWalton and Bower (1993) which shows that neonates can form prototypes from as little as 1 minute ofexposure to a visual signal. It is possible, therefore, that infants can learn to process speech differentlyfrom other sounds with minimal exposure and thus may not be born with this ability.79The fact that adults and younger infants are moderately good at discriminating(s) [ta] from [da] is consistent with Jusczyk’s model. In Jusczyk’ s model, however, theunderlying mechanism allowing adults to discriminate native phonetic information could bedifferent than the mechanism used by young infants. WRAPSA seems to imply that younginfants discriminate (s) [ta] from [da] because the weighting schemes for detectingphonemic and native language phonetic information are not yet influenced by the nativelanguage. Adults, on the other hand, can and do attend to phonetic detail both in and out ofcontext because the weighting scheme has developed to such an extent that attention can beturned to the finest level of detail of the native language, that of language-specific phoneticvariation. Thus, Jusczyk implies that young infants and adults may show a similar patternof phonetic discrimination but for different reasons.When considering the performance of 10- to 12-month-old infants, it is necessaryto expand Jusczyk’s model but first I will briefly review the evidence of developmentalchanges occurring prior to this age. Jusczyk et al. (in press) discussed evidence of changesin infant preference for various aspects of the native language.. Early in life, infants show apreference for native over non-native speech even when the speech is low-passed filtered(Mehler, Jusczyk, Lambertz, Halstead, Bertoncini, & Amiel-Tison 1988; Moon, Cooper,& Fifer, 1993). By 4.5 months, infants show a preference for sentences with pausesinserted at natural boundaries over those with pauses inserted randomly in both native andnon-native languages (Jusczyk, Hirsh-Pasek, Kemler-Nelson, Kennedy, & Schomberg,1992). By 6 months, this preference holds only for native language utterances. It is notuntil about 9 months that infants reveal sensitivity to phrasal units showing that attentionturns to less global features with development (Jusczyk, Hirsh-Pasek, Kemler-Nelson,Kennedy, Woodward, & Piwoz, 1992). Finer grained levels of detail become salientaround this age as well. For example, infants of 9 months show preference for lists ofwords that conform to the phonotactic (ordering) constraints of their native language(Jusczyk, Friederici, Wessels, Svenkerud, & Jusczyk, 1993). Infants do not, however,80show the same preference for native over non-native lists when the words are low passedfiltered indicating that it is the segmental information that infants are attending to and notsimply the prosodic structure. Furthermore, when presented with two lists of native wordsthat include either highly probable ordering of segments or less probable but stillpermissible ordering, 9-month-old infants preferred the lists with more frequentlyoccurring sequences (Jusczyk, Luce, & Charles-Luce, 1994). There are several otherpieces of evidence supporting the notion that sensitivity to components of the nativelanguage changes over the first year of life but this should suffice to make the point (for areview, see Jusczyk et al., in press).Presumably, the developmental changes outlined above allow the 10-month-oldinfant to begin to attend to native phonemic information. Because they have developed aweighting scheme for many global features and for some detailed features, they can attendto more specific and finer detail in the native language. If this is the case, we would wantto know what features of a phoneme might become weighted and thus lead the 10 to 12-month-old infant to attend to phonemic contrasts and ignore phonetic distinctions.Deciding on which features are critical to distinguishing meaning within a particularlanguage is a very complex process. We know that each language uses a different set ofphonemes but it is also likely that a phoneme will differ in the set of language-specificfeatures that an infant needs to attend to if s/he is going to be able to detect a phoneme. Inaddition, these features may not necessarily be specified in the formal description of aphonemic inventory.I would like to suggest, nevertheless, that infant phonetic processing may becomeweighted as a result of language-specific information, information that has consequenceson perception of phonemes but is not identical to a formal description of a particularphoneme. This weighting may result in processing that mimics phonemic processing butdoes not involve an awareness of what a phoneme is nor would it involve the ability to usea phoneme contrastively. An example may help to make this suggestion clearer.81One natural grouping of phonemes in English is two categories of stop consonants,voiced and unvoiced. Previous evidence has shown that English speakers rely on VOT todifferentiate between stop consonants while other languages may rely on aspiration (Lisker& Abramson, 1970; Lotz et al., 1960). In English, aspiration and VOT co-occur so anEnglish infant may attend to either of these features in order to discriminate. Thus, onepotential candidate for a feature that may be critical to English is voicing. If 10- to 12-month-old English learning infants have successfully integrated native prosody and nativephonotactic rules as Jusczyk suggests, and if they are beginning to integrate informationconcerning voicing, they will attend to voicing differences in stop consonants but will failto attend to other phonetic cues. A weighting scheme focusing on VOT, the cirtierial cue tovoicing in English, would lead English infants to discriminate phonemic stop consonantsand ignore other phonetic cues differentiating (s)[ta] from [da]. Because the onlydifference between (s) [ta] and [da] is F2 onset47,an infant listening for voicing would failthis discrimination.The present research suggests that 10- to 12-month-old English learning infantsattend to VOT and/or aspiration differences since they fail to discriminate between (s) [ta]and [da]. It is impossible to determine if English infants of 10 months attend to one cuemore than to another or attend to both equally without testing infants with a native languagephonetic (non-phonemic) distinction that differs in aspiration but not voicing (or thereverse). Unfortunately, there is no such distinction in English. What we do know is thatEnglish infants of 10 months do not discriminate the subtle F2 differences in these stopconsonants.A further extension of Jusczyk’s model offers an explanation for why younginfants and adults can discriminate (s)[ta] from [da]. It is possible that the weighting46 In English, voicing includes both VOT and aspiration. In addition, VOT differences co-occur withpresence or absence of aspiration and thus English speakers do not have to rely on aspiration to distinguishbetween voiced and unvoiced stop consonants.47 It may be important to note that although the linguistic transcriptions suggest that (s)[ta] is voicelessand [da] is voiced (thus constitutes a voicing distinction), the acoustic analysis reveals no difference invoicing in the English phonetic variants.82scheme of young English-learning infants is not sufficiently developed to allow them tofocus on voicing and/or aspiration. Indeed, they may still attend equally to both. Thus,they discriminate both voicing and aspiration distinctions as well as F2 differences. On theother hand, the weighting scheme of adults may be fully developed. As mentioned above,Jusczyk suggests that when linguistic processing is fully weighted, it allows the listener toattend to fmer detail. Thus, it is possible that adults discriminate (s)[ta] from [da] becausedetection of voicing (and/or aspiration) has become so efficient or automated that they canattend to finer phonetic detail occurring in the linguistic input. Finer phonetic detail in thepresent research includes F2 differences, the only phonetic cue differentiating (s)[ta] from[da].One extension of this possibIlity involves an empirical question. If the proposal iscorrect, infants raised in environments where only aspiration is phonemic, would attend toaspiration differences while infants raised in environments where only VOT is phonemicwould attend to VOT. On the other hand, infants born in language environments usingboth VOT and aspiration as a phonemic distinction (e.g., Thai), could develop in one oftwo ways. They could attend to both VOT and aspiration differences (performing the sameway as very young infants). Alternatively, these infants may have a bias to attendpreferentially to one before the other. Factors which might influence such developmentcould include the frequency of occurrence of each of the phonemes or differential salienceof either VOT or aspiration within their native language (see Davis, 1992, concerningdifferent rates of acquisition in production for voicing versus aspiration).What remains to be done: (1) To extend this research to infants under 6 months todetermine if the linguistic environment has an effect on perception prior to 6 months. (2)To follow the developmental trajectory with respect to perception of native languagephonetic perception (including vowels) to determine at what point children become moreadult-like in their phonetic perception. (3) To assess perception of this English allophoniccontrast in adults and infants from a linguistic community in which the voiced, voiceless83distinction is phonemic and does not co-occur with aspiration (e.g., Thai speakers). (4) Toassess perception in adults and infants who are from a language environment in which thiscontrast does not occur either in the phonemic or allophonic inventory.One other logical next step in this research is to determine if infants discriminate aphonemic contrast ([da] versus [thai) when it is presented in the context of Is!. Recall,Repp and Lin (1989) revealed that adults fail to discriminate the phonemic contrast /palversus /ba/ in the Is! context. By determining if older infants but not younger infants fail todiscriminate a phonemic contrast when that contrast is presented following Is!, we canfurther elaborate on several models. Jusczyk’s model suggests that because 6-month-oldinfants are not yet sensitive to native phonotactic rules, they should discriminate thephonemic contrast. If pre-phonemic adequately describes 10- to 12-month-old infantabilities, they should not discriminate this contrast because although this is a phonemiccontrast when it occurs in initial position, it is not phonemic when following Is!.In conclusion, the evidence provided in this research has helped to inform theoryconcerning the developmental changes in phonetic perception. It does not appear that theperceptual magnet effect can be generalized to consonant perception, and the perceptualassimilation model needs adjusting if it is going to be applied to native language phoneticperception. In addition, the four-factor model needs to rename the factor concerning 10- to12-month-old infant perception. I have suggested using the term pre-phonemic. By usingJusczyk’s model as a heuristic framework, we can begin to understand how pre-phonemicperception emerges without having to resort to complicated mechanisms. That is, the prephonemic factor develops as a function of probabilistic exposure. The evidence that bothadults and 6- to 8-months old infants can discriminate an English phonetic distinction, but10- to 12-month-old infants cannot, has led us one step closer to understanding how it isthat children develop an tacit understanding of their native language phonology within sucha short period of time.84REFERENCESAbramson, A. S., & Lisker, L. (1970). Discriminability along the voicing continuum:Cross-language tests. Proceedings of the Sixth International Congress ofPhoneticScience. Prague: Academia, 569-573.Aslin, R. N., Pisoni, D. B., Hennessy, B. L., & Perey, A. J. (1981). Discrimination ofvoice onset time by human infants: New findings and implications for the effect ofearly experience. Child Development, 52, 1135-1145.Bertoncini, J., Bijeljac-Babic, J., Jusczyk, P. W., Kennedy, L., & Mehler, 1. (1988). Aninvestigation of young infants’ perceptual representations of speech sounds.Journal ofExperimental Psychology: General, 117, 2 1-33.Best, C. T. (1990). Adult perception of nonnative contrasts differing in assimilation tonative phonological categories, Journal of the Acoustical Society ofAmerica, 88,S 177.Best, C. T. (1991, June). Phonetic influences on the perception ofnon-native speechcontrasts by 6-8 and 10-12 month olds. In Biennial Meeting of the Society forResearch in Child Development, Seattle.Best, C. T. (1994). Learning to perceive the sound pattern of English. In C. RoveeCollier & L. Lipsitt (Eds.), Advances in Infancy Research. Ablex: Norwood, NJ.Best, C. T., LaFleur, R, McRoberts, G., & Silver-Isenstadt, J. (in press). Divergentdevelopmental patterns for infants’ perception of two non-native consonantcontrasts. Journal ofExperimental Psychology: Human Perception andPeiformance.Best, C. T., & McRoberts, G. W (1989, April). Phonological influences on infants’perception of two non-native speech contrasts. Paper presented at the Society forResearch in Child Development, Kansas City.Best, C. T., McRoberts, G. W., & Sithole, N. N. (1988). The phonological basis ofperceptual loss for non-native contrasts: Maintenance of discrimination among Zuluclicks by English-speaking adults and infants. Journal ofExperimentalPsychology: Human Perception and Performance, 14, 345-360.Burnham, D. K., & Earnshaw, L. J., & Clark, J. E. (1991). Development of categoricalidentification of native and non-native bilabial stops: Infants, children and adults.Journal of Child Language, 18, 231-260.Carney, A. E., Widin, 0. P., & Viemeister, N. F. (1977). Non-categorical perception ofstop consonants differing in VOT. Journal ofAcoustical Society ofAmerica, 62,961-970.Christophe, A., Dupoux, E., Bertoncini, J., and Mehler, J. (1994). Do infants perceiveword boundaries? An empirical study of the bootstrapping of lexical acquisition.Journal of the Acoustical Society ofAmerica, 95(3), 1570-1580.85Davis, K (1992). Phonetic and phonological contrasts in the acquisition ofvoicing: Alinguistic and developmental study of Voice Onset Time production in Hindi andEnglish. Unpublished doctoral dissertation, Cornell University. Ithaca.Dutton, D. (1992). The role ofallophonic variation in the perception ofword boundaries.Unpublished doctoral dissertation, State University of New York, Buffalo.Eilers, R. E., Gavin, W.J., & Oiler, D. K. (1982). Cross-linguistic perception in infancy:Early effects of linguistic experience. Journal of Child Language, 9, 289-302.Eilers, R. E., & Minifie, F. D. (1975). Fricative discrimination in early infancy. Journalof Speech and Hearing Research, 18, 158-167.Eimas, P. D. (1975). Developmental studies in speech perception. In L. B. Cohen and P.Salapatek (Eds.), Infant Perception: From Sensation to Cognition. NY: AcademicPress.Eimas, P. D., Siqueland, E. R., Jusczyk, P., & Vigorito, J. (1971). Speech perception ininfants. Science, 171, 303-306.Fernald, A., & Mazzie, C. (1991). Prosody and focus in speech to infants and adults.Developmental Psychology, 27, 209-221.Fifer, W. P., & Moon, C. (1988). Auditory experience in the fetus. In W. Smotherman& S. Robertson (Eds.), Behavior of the Fetus. West Caldwell, NJ: Telford Press.Fry, F. B., Abramson, A. S., Eimas, P. D. & Liberman, A.M. (1962). The identificationand discrimination of synthetic vowels. Language and Speech, 5, 17 1-189.Grieser, D., & Kuhi, P. K. (1989). Categorization of speech by infants: Support forspeech-sound prototypes. Developmental Psychology, 25 (4), 577-588.Hohne, E., & Jusczyk P. J. (1994). Allophonic variation and word segmentation in infantspeech perception. Perception and Psychophysics, 56 (6), 6 13-623.Hellström, A. (1985) The time-order and its relatives: Mirrors of cognitive processes incomparing. Psychological Bulletin, 97, 35-61.Jamieson, D. G., & Morosan, D. E. (1986). Training non-native speech contrasts inadults: Acquisition of the English /aI-18/ contrast by francophones. Perception &Psychophysics, 40(4), 205 215.Jusczyk, P. W. (1993). From general to language-specific capacities: The WRAPSAmodel of how speech perception develops. Journal ofPhonetics: Special Issue:Phonetic Development, 21(1-2), 3-28.Jusczyk, P. W., & Derrah, C. (1987). Representation of speech sounds by young infants.Developmental Psychology, 23, 648-654.Jusczyk, P. W., Friederici, A. D., Wessels, J., Svenkerud, V. Y., & Jusczyk, A. M.(1993). Infants’ sensitivity to the sound patterns of native language words.Journal ofMemory and Language , 32(3), 402-420.86Jusczyk, P. W., Hirsch-Pasek, K., Kemler-Nelson, D.G., Kennedy, L. J., &Schomberg, T. (1992) Perception of acoustic correlates of major phrasal units byyoung infants. Cognitive Psychology , 24(2), 252-293.Jusczyk, P. W., Hohne, B. A., Mandel, D. R. (in press). Picking up the regularities inthe sound structure of the native language. To appear in W. Strange (Ed.), SpeechPerception and Linguistic Experience: Theoretical and Methodological Issues inCross-Language speech research: Timonium, FA; York PressJusczyk, P. W., Luce, P. A., & Charles-Luce, J. (1994). Infants’ sensitivity to highfrequency versus low frequency phonetic sequences in the native language. JournalofMemory and Language, 33, 630-645.Kent, R. D., & Read, C. (1992). The Acoustical Analysis of Speech. San Diego:Singular Publishing Group Inc..Kuhi, P. K. (1979). Speech perception in early infants: Perceptual constancy forspectrally dissimilar vowel categories. Journal ofAcoustic Society ofAmerica, 66,1168-1179.Kuhi, P. K. (1983). Perception of auditory equivalence classes for speech in earlyinfancy. Infant Behavior and Development, 6, 263-285.Kuhl, P. K. (1985). Methods in the study of infant speech perception. In G. Gottlieb &N. A. Krasnegor (eds.), Measurement of Audition and Vision in the First Year ofPostnatal Life: A methodological Overview. Norwood, NJ: Ablex.Kuhl, P. K. (1991). Human adults and human infants exhibit a prototype effect forphoneme categories; monkeys do not. Perception and Psychophysics, 50(2), 93-107.Kuhl, P. K. Williams, K. A., Lacerda, F., Stevens, K. N., & Lindblom, B. (1992).Linguistic experience alters phonetic perception in infants by 6 months of age.Science, 255(5044), 606-608.Ladefoged, P. (1982). A Course in Phonetics. : San Diego: Harcourt Brace JavanovichInc..Layer, J. (1994). Principles ofPhonetics. Cambridge: University Press.Lalonde, C. E., & Werker, J. F. (in press). Cognitive influences on changes in cross-language speech perception. Infant Behavior and Development.Laskey, R. E., Syrdal-Laskey, A., & Klein, R. B. (1975). VOT discrimination by four- tosix--and-a-half-month-old infants from Spanish environments. Journal ofExperimental Child Psychology, 20, 2 15-225.Lisker, L., & Abramson, A. S. (1964). A cross-language study of voicing in initial stops:Acoustical measurements. Word, 20(3).Lisker, L., & Abrainson, A. S. (1967). Some effect of context on voice onset time inEnglish stops. Language and Speech, 10, 1-28.87Lisker, L., & Abramson, A. S. (1970). The voicing dimension: Some experiments incomparative phonetics. In B. Hala, M. Romporti, & P. Janota (Eds.) Proceedingsof the 6th International Congress ofPhonetic Sciences. Prague: Academia. 563-567.Lotz, J., Abramson, A. S., Gerstman, L. J., Ingemann, R., & Nemeser, W.J. (1960).The perception of English stops by speakers of English, Spanish, Hungarian, andThai: A tape cutting experiment. Language and Speech, 3,71-77.Marascuilo, (1966). The Scheffe Theorem. Psychological Bulletin.MacKain, K. S., Best, C. T., & Strange, W. (1981). Categorical perception of English Inand Ill by Japanese bilinguals. Applied Psycholinguistics, 2, 369-390.MacKain, K. S., & Stern, D. N. (1982). The concept of experience in speechdevelopment. In K. E. Nelson (Ed.), Children’s Language, Vol Vi. NY: GardnerPress.McKenna, F. P. (1984). Assimilation and contrast in perceptual judgments. QuarterlyJournal of Experimental Psychology, 36A, 531-548.McClelland, J. L., & Elman, J. L. (1986). The TRACE model of of speech perception.Cognitive Psychology, 18, 1-86.Mehler, J., Bertoncini, J., & Barriere, M. (1978). Infant recognition of mother’s voice.Perception, 7(5) 491-497.Mehler, J., Jusczyk, P. W., Lambertz, G., Halstead, N., Bertoncini, J., & Amiel-Tison,C. (1988). A precursor of language acquisition in young infants. Cognition, 29,144- 178.Miyawaki, K., Strange, W., Verbrugge, R., Liberman, A. M., Jenkins, J. J., &Fujimura, 0. (1975). An effect of linguistic experience: The discrimination of Inand Ill by native speakers of Japanese and English. Perception and Psychophysics,18, 331-340.Moon, C., Cooper, R. P., & Fifer, W. P. (1993). Two day olds prefer their nativelanguage. Infant Behavior and Development, 16 (4), 495-520.Morgan, J. L., & Saffron, J. R. (1995). Emerging integration of segmental and suprasegmental information in prelingual speech segmentation. Child Development, 66(4), 911-936.Papousek, M., Bomstein, M. H., Nuzzo, C., Papousek, H., & Symes, D. (1990).Infant responses to prototypical melodic contours in parental speech. InfantBehavior and Development, 13(4), 539-545.Papousek, M., Papousek, H., & Symes (1991) The meanings of melodies in motherese intone and stress languages. Infant Behavior and Development, 14,415-440.Pisoni, D. B., Aslin, R. N., Perey, A. J., & Hennessy, B. L. (1982). Some effects oflaboratory training on identification and discrimination of voicing contrasts in stopconsonants. Journal ofExperimental Psychology: Human perception andPeiforinance, 8: 297-314.88Pisoni, D. B, Logan, J. S., & Lively, S. E. (1994). Perceptual learning of nonnativespeech contrasts: Implications for theories of speech perception. In H. C.Nusbaum & J. Goodman (Eds.), Development ofSpeech Perception: TheTransition from Recognizing Speech Sounds to Spoken Words. Cambridge: MiTPress.Pisoni, D. B., & Tash, J. (1974). Reaction times to comparisons within and acrossphonetic categories. Perception and Psychophysics, 15, 285-290.Polka, L. (1991). Cross-language speech perception in adults: Phonemic phonetic, andacoustic contributions. Journal ofAcoustical Society ofAmerica, 89(6), 296 1-2977.Polka, L. (1995). Linguistic influences in adult perception of non-native vowel contrasts.Journal ofAcoustical Society ofAmerica, 97(2),Polka, L., & Bohn. 0-5. (1994, June). A cross language comparison of vowelperception in English-learning and German-Learning infants. Paper presentation atThe International Conference on Infant Studies, Paris, France.Polka, L., & Werker, J. F. (1994). Developmental changes in perception of nonnativevowel contrasts. Journal ofExperimental Psychology: Human Perception &Performance, 20(2), 421-435.Repp, B. H. (1984). Categorical perception: Issues, methods, findings. In N. J. Lass(Ed.), Speech and Language: Advances in Basic Research and Practice. NewYork: Academic Press.Repp, B. H., & Lin, H. (1989). Effects of preceding context on discrimination of voiceonset times. Perception and Psychophysics, 45(4), 323-332.Sawush, J. R., & Jusczyk. P. W. (1981). Adaptation and contrast in the perception ofvoicing. Journal ofExperimental Psychology: Human Perception & Performance,7(2), 408-421.Schouten, M. E. H., & van Hessen, A. J. (1992). Modelling phoneme perception I:Categorical perception. Journal ofAcoustical Society ofAmerica, 92(4), 184 1-1855.Stager, C., & Werker, J. F. (1995, June). Phonetic similarity influences learning word-object associations in 14-month-old infants. Presentation at the Annual Conferenceof the American Psychological Society, New York.Streeter, L. L. (1976). Language experience of two-month-old infants shows effect ofboth innate mechanisms and experience. Nature, 259, 39-4 1.Studdert-Kennedy, M. (1976). Speech perception. In N. J. Lass (Ed.), ContemporaryIssues in Experimental Phonetics. NY: Academic Press.Studdert-Kennedy, M. (1986). Sources of variability in early speech development. In J.S. Perkell & D. H. Klatt (Eds.), Invariance and Variability ofSpeech Processes.Hillsdale, NJ: Eribaum.89Sussman, J. E., & Lauckner-Morano, V. J. (1995). Further tests of the “perceptualmagnet effect” in the perception of [i]: Identification and change/no changediscrimination. Journal ofAcoustical Society ofAmerica, 97(1), 539-552.Swoboda, P. J., Morse, P.A., & Leavitt, L. A. (1976). Continuous vowel discriminationin normal and at risk infants. Child Development, 47,459-465.Tees, R. C., & Werker, J. F. (1984). Perceptual Flexibility: Maintenance or recovery ofability to discriminate non-native speech sounds. Canadian Journal ofPsychology,38, 579-590.Trehub, S. E. (1976). The discrimination of foreign speech contrasts by infants andadults. Child Development, 47, 466-472.Trehub, S. E., Thorpe, L. A., & Trainor, L. J. (1990). Infants’ perception of good andbad melodies. Psychomisicology, 9(1).Treisman, A. (1986). Features and objects in visual processing. Scientific American,255(5), 114B-125.Walton, G., & Bower, T. G. R. (1993). Amodal representation of speech in infants.Infant Behavior and Development, 16, 233-243.Warren, R. M. (1985). Criterion shift rule and perceptual homeostasis. PsychologicalReview, 94(4), 574-584.Werker, J. F. (1989). Becoming a native listener. American Scientist, 77, 54-59.Werker, J. F. (1991). Ontogeny of speech perception. In I. G. Mattingly & M. StuddertKennedy (Eds.), Modularity and the Motor Theory ofSpeech Perception.Hillsdale, NJ: Lawrence Eribaum.Werker, J. F. (1992). Cross-language speech perception: Developmental change does notinvolve loss. In J. Goodman and H. Nusbaum (Eds.), Speech Perception andWord Recognition. Cambridge: MIT Press.Werker, J. F. (1993). The contribution of the relation between vocal production andperception to a developing phonological system. Journal ofPhonetics: SpecialIssue: Phonetic development. 21(1-2), 177-180.Werker, J. F., & Baldwin, D. A. (1991, April). Speech perception and lexical acquisition.Paper presented at the biennial meeting of the Society for Research in ChildDevelopment. Seattle, WA.Werker, 3. F., Cohen, L., & Lloyd, V. (submitted). A new procedure for assessing wordlearning.Werker, 3. F., Gilbert, J. H. V., Humphrey, K., & Tees, R. (1981). Developmentalaspects of cross-language speech perception. Child Development, 52, 349-353.Werker, 3. F., & Lalonde, C. (1988). Cross-language speech perception: Initialcapabilities and developmental change. Developmental Psychology, 24, 672-683.90Werker, J. F., & Logan, J. S. (1985). Cross-language evidence for three factors in speechperception. Perception and Psychophysics, 37, 35-44.Werker J. F., & Pegg, J. E. (1992). Infant speech perception and phonologicalacquisition. In C. Ferguson, L. Menn, and C. Stoel-Gammon (Eds.),Phonological Development: Models, Research, and Implication. Parkton,Maryland: York Press.Werker, J. F., & Polka, L. (l993a). Developmental changes in speech perception: Newchallenges and new directions. Journal ofPhonetics., 21, 83-101.Werker, 3. F., & Polka, L. (1993b). The ontogeny and developmental significance oflanguage-specific phonetic perception. In de Boysson-Bardies, S. de Schonen, P.Jusczyk, P. McNeilage, & J. Morton (Eds.), Developmental Neurocognition:Speech and Face Processing in the first year of life (pp. 275-288); NATO ASISeries. Norwell, MA: Kiuver Academic Press.Werker, J. F., & Tees, R. C. (1984a). Cross language speech perception: Evidence forperceptual reorganization during the first year of life. Infant Behavior andDevelopment, 7, 49-63.Werker, J. F., & Tees, R. C. (1984b) Phonemic and phonetic factors in adult cross-language speech perception. Journal ofAcoustical Society ofAmerica, 75, 1865-1878.Werker, J. F., & Tees, R. C. (1983). Developmental change across childhood in theperception of non-native speech sounds. Canadian Journal ofPsychology, 37,278-286.91APPENDIX 1Definition of a Few Linguistic TermsTwo terms used to describe speech sounds are phones and phonemes. Phones is aterm used to describe speech sounds used in languages of the world and are transcribed insquare brackets, as in [d]. Each specific language uses only a subset of these phones.Among the set of phones that are particular to a language is a smaller set called phonemes.Phonemes are phonologically defined linguistic segments transcribed in slanted brackets, asin Id!. Although there are several definitions of phonemes (e.g., Layer, 1994), the notionthat is important here is that phonemes signal a difference in meaning between a minimalpair. An example of a minimal pair in English is “tip” versus “dip” which are identicalexcept for the difference between It! and Id!. Thus we know that It! and Id! are Englishphonemes because there is a difference in meaning between this minimal pair.Speakers may produce phonetic variations of phonemes and these variations arecalled allophones. One type of allophone is a phonetic variation of a phoneme that occursrandomly and is therefore said to occur in free variation. Examples of this type ofallophone include dialectic differences among speakers, phonetic differences between male,female and child voices, and even phonetic variation within an individual speaker (for amore complete list see Best, 1994; Studdert-Kennedy, 1976, 1986). Because this type ofallophone is free to vary and is not specified by context in a regular and predictable fashion,it is not specified by the phonology of the language. Thus, we might say that theseallophones are phonologically irrelevant. Although not all aspects of this kind ofallophonic variation have been studied, the research that has been done with infantssuggests that they can ignore this kind of phonetic variation when the testing situationrequires them to perceive categories (e.g., Kuhi, 1979, 1983; Bertoncini, Bijeljac-Babic,Jusczyk, Kennedy, & Mehier, 1988; Jusczyk & Derrah, 1987).92A second kind of allophone is specified by context but is optional. That is, aspeaker may or may not use the allophonic variant. An example of this kind of allophone isthe way in which the It! in “little” is sometimes produced as a flap [TI but carefullyarticulating speakers will produce a It! albeit sometimes still phonetically different from aninitial position It’. Thus, although a It! occurring in medial position may be produced as anallophonic variant, it is not a mandatory rule in English.The type of allophone relevant to this thesis is the context-dependent allophonewhich is specified by context and is mandatory—speakers must use the allophone in theappropriate context. These allophones are said to occur in complementary distributionbecause they occur within a specified context: one context is reserved for one allophone theother context is reserved for the other allophone. An example of an English context-dependent allophone occurring in complementary distribution is the It! in “stack” versus theIt! in “tack”. Although there are several phonetic differences between these segments, thereis one phonetic difference which is phonologically relevant (but is not phonemic) inEnglish. Specifically, the It! in “tack” is aspirated while the It! in “stack” is not. Aspirationis the puff of air which follows the release of the stop48. Aspiration occurs during theVOT, that is, before the onset of voicing, and is transcribed as 1h, Thus, the It! in “tack” isphonetically transcribed as [thi and the ItJ in “stack” is phonetically transcribed as [t].The phonology of a language includes structured principles about these context-dependent allophones and every speaker has an tacit (but not necessarily explicit)understanding of these language-specific context-dependent rules. The Englishphonological rule concerning the allophone described above states that there are twoallowable alveolar stops in initial position, the voiceless aspirated [thi and the voiced48 In English, aspiration is called a redundant feature because it co-occurs with other features such as voice-onset-time (VOT) and, thus, is not contrastive. In other languages, however, aspiration may be a phonemicfeature. For example, Thai uses the phonemes, It! and /th/, which differ in presence or absence of aspirationand in Thai, aspiration is called a distinctive feature because it serves to differentiate meaning. Whether theEnglish allophones [tJ and [thi are phonetically identical to the Thai phonemes It) and it1’! is not known butdetailed acoustic analyses might be predicted to reveal phonetic differences between these segments.93unaspirated [d], phonemically transcribed as It! and Id!. However, when an alveolar stop ispreceded by [s], it becomes the voiceless unaspirated [t] (e.g., Ladefoged, 1975). Thisphonological rule applies to all stops including bilabial stops [ba] and [pha] and velar stops[gal and [kha]. In formal terms, the general phonological rule states:i.e.: The phonemes become the allophones in these contexts:/p/,ItJ, /k! —* #[ph], [thi, [kh] — (i.e., syllable initial position)—s[p], [t], [k] — (i.e., following Is?)Thus, in the context of [SI English speakers consistently and predictably produce anallophonic variant of initial position stop consonants. In fact, in linguistic terms, both [t1’]and [t] are allophonic variants of the phoneme It).The reason English phonology groups (s)[t] with [th] relates to the fact that in theunderlying representations, voicing is constant in a consonant cluster. Since [sI isunvoiced, the stops following [sJ will also be unvoiced (Ladefoged, 1982). In otherwords, a constraining factor about the underlying phonological structure of English is that[sd] is an unacceptable cluster. In fact, [ds] is also an unacceptable cluster. Any Englishspeaker would realize that [ds] is not allowed in word initial position but they may notrealize that it is also not allowed in fmal position. This is not inherently obvious to non-linguistically trained English speakers because we have words like “bids” spelled with an‘s’. However, if you listen carefully to the way you produce the “s” following the voicedstop, you will hear that it sounds more like a [z] than an [s]. Because [zI is the voicedversion of [s], the phonological rule is maintained even when the cluster occurs in the wordfinal position. This variation in Is! following a voiced versus an voiceless segment isanother example of an English allophonic rule.94APPENDIX 2Example of Coding Sheet for Experiment 1SESSION #___________ SUBJECT CODE______staPOOR EXCELLENT1 2 3 4 5 6 71.________________21.______________41.2._______22. 42.3._____23.______43.4. 24. 44.5. 25. 45.6._26._____46.7. 27._47.8.28. 48.9. 29. 49.10.30. 50.11. 31. 51.12. 32. 52.13._33._53.14. 34. 54.15. 35. 55.16. 36. 56.17. 37. 57.18. 38. 58.19. 39. 59.20. 40. 60.


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items