referential labeling can facilitate phonetic learning in...

Referential Labeling Can Facilitate Phonetic Learning in Infancy

H. Henny Yeung, Lawrence M. Chen, and Janet F. WerkerUniversity of British Columbia

All languages employ certain phonetic contrasts when distinguishing words. Infant speech perception israpidly attuned to these contrasts before many words are learned, thus phonetic attunement is thought toproceed independently of lexical and referential knowledge. Here, evidence to the contrary is provided.Ninety-eight 9-month-old English-learning infants were trained to perceive a non-native Cantonese tone con-trast. Two object–tone audiovisual pairings were consistently presented, which highlighted the target contrast(Object A with Tone X; Object B with Tone Y). Tone discrimination was then assessed. Results showedimproved tone discrimination if object–tone pairings were perceived as being referential word labels, althoughthis effect was modulated by vocabulary size. Results suggest how lexical and referential knowledge couldplay a role in phonetic attunement.

For many types of perceptual stimuli (i.e., speech,faces, musical rhythms, etc.), human infants quicklyattune to relevant perceptual differences, and learnto ignore irrelevant ones (Scott, Pascalis, & Nelson,2007). Phonetic perception offers a classic exampleof attunement, as experience with the native lan-guage begins affecting the perception of phoneticcontrasts from the 1st year of life (e.g., Best,McRoberts, & Goodell, 2001; Narayan, Werker, &Beddor, 2010; Rivera-Gaxiola, Silva-Pereyra, &Kuhl, 2005; Sato, Kato, & Mazuka, 2012). A classicdevelopmental pattern is maintenance for the dis-crimination of native contrasts, and a decline forthe discrimination of non-native contrasts (Werker& Tees, 1984).

One reason that phonetic attunement is interest-ing is because its acquisition describes a learningproblem. Early thinkers (e.g., Jakobson, 1942/1968)suggested that a language’s unique set of phoneticcontrasts could be learned by comparing minimalpairs of words (e.g., the pair of words bat andpat minimally differ in the first consonant). How-ever, such a mechanism is seen as implausible in

infancy, as learners would require sufficientnumbers of phonological neighbors—words with sim-ilar phonological forms—to compare with oneanother to deduce native phonetic categories, andthis has not yet occurred by the end of the 1st year(e.g., Coady & Aslin, 2003). How then can phoneticattunement occur if infants do not have a suffi-ciently sizable lexicon?

Mechanisms of Phonetic Attunement

One approach considers how native languagesound patterns can be deduced from the statisticalfrequency of phonetic features in acoustic space(Gauthier, Shi, & Xu, 2007; Jusczyk, 1993; Kuhl,1993; McMurray, Aslin, & Toscano, 2009; Salminen,Tiitinen, & May, 2009; Vallabha, McClelland, Pons,Werker, & Amano, 2007; Werker et al., 2007). Mayeand colleagues tested this hypothesis in the labora-tory by presenting 6- to 8-month-olds with phonetictokens that varied along a single acoustic dimen-sion for just a few minutes. Tokens were distributedaround either a single mode (the unimodal group),or two separate modes (the bimodal group), simu-lating exposure to languages that had one or twophonetic categories defined by this dimension. Onlythe bimodal group discriminated the relevantcontrast, demonstrating distributional learning fromsmall amounts of speech exposure (Maye, Werker,& Gerken, 2002), an effect now replicated severaltimes (Cristia, 2011; Cristia, McGuire, Seidl, &

H. Henny Yeung is now at the Universit�e Paris Descartes &CNRS (Laboratoire Psychologie de la Perception, UMR 8158).

This work was supported by the McDonnell Foundation(412783-001G) to Richard N. Aslin and Janet F. Werker, and bythe National Sciences and Engineering Research Council of Can-ada (81103) to Janet F. Werker. We thank Jacqueline Chong andMoko Chen for help preparing stimuli, as well as Laurel Fais,Alison Greuel, and Priya Kandhadai for valuable suggestions onan earlier draft.

Correspondence concerning this article should be addressedto H. Henny Yeung, Laboratoire Psychologie de la Perception,Universit�e Paris Descartes, 45 rue des Saints-P�eres, 75006Paris, France. Electronic mail may be sent to [email protected].

© 2013 The AuthorsChild Development © 2013 Society for Research in Child Development, Inc.All rights reserved. 0009-3920/2014/8503-0015DOI: 10.1111/cdev.12185

Child Development, May/June 2014, Volume 85, Number 3, Pages 1036–1049

Francis, 2011; Maye, Weiss, & Aslin, 2008; Yoshida,Pons, Maye, & Werker, 2010).

Recent work has suggested, however, that statis-tical information from frequency distributions aloneis not sufficient to explain phonetic attunement(e.g., McMurray et al., 2009; Werker, Yeung, &Yoshida, 2012). Supplementary mechanisms mayconspire with statistical information. For example,phonetic learning is enhanced when speech inputcomes from “social” scenarios involving live experi-menters (Kuhl, Tsao, & Liu, 2003). Distributionallearning may also be supplemented by redundantvisual cues about lip movements (Teinonen, Aslin,Alku, & Csibra, 2008). Moreover, other kinds of co-occurring cues, like visual objects (Hayes-Harb,2007), may conspire with statistical information insupplementary ways, perhaps helping infants rec-ognize two categories when statistical cues aremuddled. For example, infants may frequently hearspeech of one form (e.g., doll) co-occurring with onekind of object, and speech of another form (e.g.,ball) co-occurring with another. Phonetic learningfrom this kind of supplementary visual cue is thefocus of this study.

A recent report demonstrated infant learningfrom these sorts of visual co-occurrences: Nine-month-olds’discrimination of a non-native speechcontrast was improved when one type of speechsound was consistently paired with one object,and the other type was consistently paired with adifferent object (Yeung & Werker, 2009). This phe-nomenon was attributed to a learning principle,termed acquired distinctiveness (Kluender, Lotto,Holt, & Bloedel, 1998; Lawrence, 1949; Thiessen,2011), such that when Token A is consistentlyheard in Context B (e.g., co-occurring with oneobject), and Token A’ is consistently heard in Con-text C (e.g., co-occurring with another object), thenthe perceptual distance between Tokens A and A’

is increased.Phonetic learning from acquired distinctiveness

has some similarities to the earlier described mini-mal pair theory of phonetic learning (i.e., maintain-ing a /b/–/p/ distinction because bat and patare contrastive words). However, two importantfeatures of acquired distinctiveness circumvent thelearning problem inherent to the minimal pairtheory. First, acquired distinctiveness is a basic,domain-general learning principle that is likelyavailable to infants from early in life. While younginfants may not have words for concepts like bat(the animal) or pat (the gesture), they may still hearthese speech sounds in different environmentalcontexts. Second, acquired distinctiveness draws

from a potentially richer data set than just thenumber of known words. Indeed, distinctive con-texts come not only from co-occurring visual cues(Hayes-Harb, 2007; Yeung & Werker, 2009), butalso from the auditory contexts (i.e., lexical formsthat infants may have already segmented from thespeech stream) in which phonetic tokens occur(Feldman, Myers, White, Griffiths, & Morgan, 2013;Swingley, 2009).

The Role of the Lexicon

The mechanisms of infant phonetic learning thatare described above suggest that the lexicon mayindeed be important for phonetic category acquisi-tion, but only insofar as it structures the statisticalaspects of phonetic input, and the contexts inwhich input is given. Nevertheless, some research-ers still hypothesize ways in which the actual lexi-cal knowledge of an infant can act upon phoneticcategory learning. Specifically, certain kinds of lexi-cal or referential cues may increase the efficacy ofphonetic learning (Heitner, 2004; Werker et al.,2012).

Consider a closely related perceptual categoriza-tion study, where adults learned a visual distinctionbetween two kinds of “Greebles,” an artificiallycreated set of alien-like figures (Lupyan, Rakison, &McClelland, 2007). The two kinds were learnedmore easily when cued by a lexical contrast(i.e., knowing a different label for each kind ofGreeble), than when cued by a nonlexical contrast(i.e., knowing the two locations where each kind ofGreeble lived). Top-down information about thestatus of a “label” may have deepened perceptuallearning of the Greeble categories through somefeedback process not available in the nonlexicalcondition (the “label feedback hypothesis,” Lupyan,2012).

Visual categorization in infancy is similarlyaffected, as hearing linguistic stimuli promotesobject categorization in ways that hearing nonlin-guistic stimuli does not (e.g., Ferry, Hespos, &Waxman, 2010). Here, we ask whether a similarkind of “label feedback” can influence phoneticlearning as well. That is, does the lexical status of alabel affect categorization, not just of visual objectcategories but also of phonetic information in thelabels themselves (see also Heitner, 2004)? Acquireddistinctiveness contexts offer an elegant way to testthis question. In such contexts, are infants better atdifferentiating two kinds of speech sounds (pairedwith two objects) when these sounds are knownto be lexically contrastive (i.e., when the sounds

Labeling Guides Infant Phonetic Learning 1037

represent two labels) compared to when they arenonlexically contrastive?

Two experiments are reported, both of whichadapt the acquired distinctiveness training proce-dure from Yeung and Werker (2009). Nine-month-old infants in that previous report were trained todiscriminate a non-native phonetic contrast usingtwo consistent object–sound pairings. Phoneticlearning (i.e., improved discrimination of the targetcontrast) was seen when consistent pairings weretrained, compared to when no training or inconsis-tent training was given. The present Experiment 1used a similar procedure to again train infants, butthis time on an even harder non-native contrast.Experiment 2 then asked whether phonetic learningcould be facilitated when object–sound pairingswere presented in a more referential manner (i.e.,when pairings were interpreted as word labels,rather than just co-occurrences).

Experiment 1

Experiment 1 probed the role of acquired distinc-tiveness contexts (i.e., the co-occurrence of twodifferent speech sounds with two visually distinctobjects) in the establishment of the native languagephoneme repertoire. Two differences from theoriginal Yeung and Werker (2009) study were intro-duced.

The first difference was that a new phonetic con-trast was tested. Yeung and Werker (2009) testedEnglish-learning infants’ discrimination of two /d/-like phones: a dental [d̪] and a retroflex [ɖ], whichdiffer along a familiar phonetic dimension (i.e., placeof articulation, or where in the vocal tract a consonantis produced). English normally has a single interme-diate category (alveolar /d/) in this part of thedimension, and thus the dental–retroflex contrastwas one that partitioned the native /d/ into aslightly more fronted /d/ category (a dental placeof articulation), or a slightly more back /d/ category(a retroflex place of articulation). Thus, acquired dis-tinctiveness in Yeung and Werker (2009) was shownto create new categories along a familiar phoneticdimension (rather than bring attention to a contrastdifferentiated along a novel phonetic dimension).Moreover, the perception of this dental–retroflexcontrast in English-learning infants is in the midst ofdecline at 9 months of age (Werker & Tees, 1984),and it was not clear if acquired distinctiveness coulddo more than simply enhance an already weaklydiscriminable contrast. For these two reasons, thetarget phonetic contrast was changed.

Here, we chose to use Cantonese lexical tones,which are cued primarily by pitch variations withina single syllable (Khouw & Ciocca, 2007). Tonerepresents a stronger test of perceptual learning forthe two reasons above. Tone distinctions are basedon a phonetic dimension (pitch) that is not, byitself, lexically relevant in English. Moreover, toneperception is more entrenched than dental–retroflexcontrasts at this age: English-learning 9-month-oldsconsistently fail to discriminate tones (Mattock &Burnham, 2006; Mattock, Molnar, Polka, & Burnham,2008; Yeung, Chen, & Werker, 2013), and furtherignore pitch variation in the recognition of seg-mented words (Singh, White, & Morgan, 2008).Thus, evidence of a newly learned tone contrastwould constitute a clearer, more unambiguous caseof learning (not just enhancement) in phoneticperception.

The second difference from the acquired distinc-tiveness procedure used by Yeung and Werker(2009) was a slight, but theoretically importantmodification. In the earlier work, the co-occurrenceof speech sounds and objects was purposely con-trolled: The presentation of the syllables wassynchronized with the onset of object movement.This tight coupling of movement and speechsounds can occur in actual speech input fromparents (Gogate, Bahrick, & Watson, 2000), andfacilitates associative learning (Gogate & Bahrick,2001). However, real word labels do not alwaysoccur in perfect temporal synchrony with theirreferents, and such synchrony can often promote adifferent interpretation of the speech sound (as an“amodal” feature of the object). To eliminate thesecomplications in the present work, the co-occur-rence of the tones and the objects was maintained,but synchrony in the timing of presentation waseliminated.

Method

Participants

Infants were healthy, full-term infants around9 months of age recruited from a database of par-ents who expressed interest in research studies(primarily recruited at local maternity wards).Infants were exposed to English at least 80% of thetime in their environment, and heard no other tonelanguages. Twenty infants were tested (eight boys,M = 270 days, range = 260–282 days). Nine addi-tional infants were not included for the followingreasons: excessive fussiness (n = 6) and experi-menter or equipment error (n = 3).

1038 Yeung, Chen, and Werker

Stimuli

Target auditory stimuli were identical to thoseused in another report, in which 9-month-oldEnglish-learning infants failed to discriminate a risingversus level tone contrast, even though English-learning 4-month-olds, and Chinese-learning 4- and9-month-olds succeeded (Yeung et al., 2013). Thetones used are native to Cantonese, and it is com-mon to notate these tones with numeric sequencesdescribing the starting and ending pitch on a sub-jective scale from 1 to 5 (higher numbers = higherpitch). Adult-directed consonant–vowel syllables,“chee” (/ʨ ͡h i/), were spoken by a female nativeCantonese speaker with either a high-rising “Tone25” (此 this; 始 start), or a midlevel “Tone 33” (次next; 刺 thorn). Four tokens of each type wererecorded and well matched in f0 range, duration,and amplitude (Figure 1a; see Yeung et al., 2013,for further details).

Visual stimuli during the training phase con-sisted of a set of animated cartoon objects(Figure 1b), which were displayed while infantsheard one of the tone stimuli. The visual stimulusduring the test phase was a static checkerboard(Figure 1b).

Procedure

Infants sat on their parent’s lap approximately 36 in.from a 27-in. screen placed in a sound-attenuatedand dimly lit room. Parents listened to music overheadphones, and were instructed not to speak orpoint at the screen. A video camera allowed theexperimenter to see the infant from a separate room.Sound was presented at about 65 dB, and each trialconsisted of stimuli presentation until the infant

looked away for 2 s. An attractive silent animationwas then presented, and the next trial began when theinfants refixated the screen. Stimuli presentation wascontrolled using Habit X software (http://habit.cmb.ucdavis.edu/) running on OS X on a Mac Pro (AppleInc., Cupertino, CA).

Training trials consisted of an animation display-ing a rotating object moving horizontally on thescreen for 7 s. Five phonetic tokens from one tonecategory (four unique) were also presented with avariable interstimulus interval (ISI) of around 850 ms,which prevented audiovisual synchronization. Thenext trial started immediately, featured a differentobject moving in the same manner, and presentedwith five tokens (four unique) from the other tonecategory. Trials continued until 2 min of lookingwas accumulated.

The test phase followed training, and consistedof a modified alternating/nonalternating discrimi-nation procedure. The first trial contained a singlecheckerboard as a visual stimulus, presented insilence to give infants a chance to process the newvisual stimulus after the training phase. Eight trialsfollowed, all containing the same visual checker-board. Nonalternating (Non-Alt) trials contained allthe unique tokens of one tone type, whereas alter-nating (Alt) trials also contained four uniquetokens, ordered such that one tone type always fol-lowed a contrasting type for the first four stimuluspresentations, and then both tone types occurredrandomly. Maximum looking to each trial was setto 30 s, and all auditory stimuli were presentedwith a 1-s interstimulus interval.

The logic of this procedure is that looking timeswill be different for Alt and Non-Alt trials if thephonetic contrast is discriminable, and previouswork suggested that longer looking to the Alt trial

Pre-Training Training Test

RisingTone 25

LevelTone 33

Tone Discrimination

b)a)

Pi tch (Hz)

Time (s)0 0.1 0.2 0.3 0.4 0.5 0.6

180

200

220

240

260

280High Rising (25)

Mid Level (33)

Figure 1. (a) Pitch tracks for each stimulus token. (b) Visual stimuli used Experiments 1 and 2, and the picture of the experimentalsetup (far right). Infants in Experiment 1 and in the nonreferential condition of Experiment 2 were presented with two novel objects,each uniquely paired with one tone type during training. Infants in the referential condition of Experiment 2 were presented with thesame training, as well as three familiar object trials before training.


type is seen when discriminating this (or similar)tone contrasts (Mattock et al., 2008; Yeung et al.,2013). Alt trials were always followed by Non-Alttrials, and vice versa. Altogether, four Alt trials andfour Non-Alt trials were presented, counterbalanc-ing which trial type occurred first, as well as whatorder the two kinds of Non-Alt trials occurred (seeYeung et al., 2013, for details). Trained observersblind to the auditory stimuli coded looking offlinefor statistical analysis.

Results

Results showed no evidence of tone discrimina-tion, as looking in the test phase to Alt trials wasnot different from looking to Non-Alt trials,t(19) = .42, p = .82 (Figure 2).

Discussion

Infants in Experiment 1 failed to show tone dis-crimination, even after being trained on consistentobject–tone pairings that highlighted the contrast.This differs from the results reported by Yeung andWerker (2009), in which a dental–retroflex conso-nant contrast was tested. It is not clear if this failureto discriminate stems from the procedural changesmade in this study—particularly the desynchroniza-tion of object and sound—or from the use of a tonal

contrast rather than a consonant distinction. Never-theless, this result allowed us to directly testwhether phonetic learning could be improved ifobject–tone pairings were presented in a morereferential context. Experiment 2 used the sameprocedure as Experiment 1, but manipulated thereferential interpretation of the training.

Previous work from the word mapping literatureprovided inspiration for the experimental manipula-tion (Fennell & Waxman, 2010). Although infants at14 months of age can discriminate many phoneticcontrasts, they have difficulty using this perceptualknowledge in word mapping. For example, eventhough infants successfully map two phoneticallydistinct forms (e.g., lif with Object A; neem withObject B), they have difficulty mapping phoneti-cally similar forms (e.g., bin with Object A; din withObject B; see Stager & Werker, 1997). Fennell andWaxman (2010) found that they could make thistask more “referential,” and improve mapping abili-ties. They did this by first presenting infants withthree trials containing familiar object–label pairings(e.g., car with an image of a car). They argued thatthis provided a referential context because infantsknew that they would be seeing images of objectsand hearing their correct, referential labels. The rec-ognition of this referential context was consideredhelpful to infants, as they better accessed phoneticdetail when learning novel word mappings.

0

2

4

6

8

(n = 20) (n = 19) (n = 21) (n = 18) (n = 20)

Vocab Unknown Low Vocab High Vocab Low Vocab High Vocab

Nonreferential Nonreferential Referential

Experiment 1 Experiment 2

Lo

okin

g T

ime (

s)

Non-Alt Alt

*

Figure 2. Looking times to alternating and nonalternating trial types during the test phase from the Experiments 1 and 2. In Experiment2, each experimental condition (nonreferential or referential) is separated into low- and high-vocabulary groups (based on a mediansplit). Error bars indicate standard errors.*p < .05.


In Experiment 2, a similar manipulation wasused to test 9-month-old infants’ sensitivity toreferential contexts, but unlike Fennell and Wax-man (2010), the focus was not on word mapping.Instead, this study assessed whether the referentialmanipulation introduced by Fennell and Waxmancould enhance phonetic learning of a non-nativecontrast. If infants were indeed sensitive to thereferential value of the manipulation, it shouldfacilitate phonetic learning.

Experiment 2

Two conditions were tested. In the nonreferentialcondition, a group of infants participated in the sameprocedure as Experiment 1. In the referential condi-tion, another group of infants did the same task,except three trials containing familiar object labelspreceded the training phase (see Fennell & Waxman,2010). If this referential distinction was meaningfulto infants, it was predicted that improved phoneticdiscrimination—as measured in the test phase—would be seen in the referential condition (whereinfants are trained on two-word labels) compared tothe nonreferential condition (where infants aretrained on two co-occurring sounds).

Because our hypotheses were related to thelexical knowledge of infants, and because 9-month-olds in our sample were younger (and likely hadsmaller vocabularies) than the 14-month-olds testedin Fennell and Waxman (2010), a measure of vocab-ulary size was also recorded. The short form of theMacArthur Communicative Development Invento-ries (CDI) was used, as it offers an easy-to-completechecklist of common words that infants may com-prehend or produce (Fenson et al., 2000). Here,we modified the short form to include the familiarobject labels used in the experimental procedure.

Method

Participants

Infants were recruited as in Experiment 1. Thenonreferential condition included 40 infants (19boys, M = 273 days, range = 255–302 days). Twelveadditional infants were not included for the follow-ing reasons: hearing a tone language (n = 2),excessive fussiness (n = 7), and experimenter/equipment error (n = 3). The referential conditionincluded 38 infants (20 boys, M = 270 days,range = 256–294 days). Seventeen additional infantswere not included for the following reasons:

excessive fussiness (n = 10), experimenter or equip-ment error (n = 5), and being out of camera range(n = 2).

Stimuli

Auditory and visual stimuli were identical tothose in Experiment 1, except that four auditorytokens each of banana, car, and keys were recordedfrom the same speaker who produced the tonestimuli. These words were chosen because they arehighly frequent for infants, and because animatedcartoon stimuli similar to the novel objects inExperiment 1 were also available for these items(Figure 1b).

Procedure

The procedure was identical to Experiment 1 inboth conditions, except that in the referential condi-tion, three pretraining trials were shown. Here,animations of the familiar objects (banana, car, andkeys) were paired with five tokens (four unique) oftheir corresponding labels. Pretraining trials werepresented for 14 s each, but not in an infant-con-trolled manner, ensuring that these trials were notinadvertently cut short by looks away. All parentsalso completed our modified version of the Mac-Arthur CDI (Fenson et al., 2000), which consisted ofthe standard short form, altered such that banana,car, and keys all appeared.

Results

A preliminary analysis tested for differences inattention during the training phase between condi-tions. We measured how many infant-controlledtrials were needed to accumulate this preset look-ing criterion. This is similar to examining totallooking time during habituation to control forattentional differences across conditions (Cohen,Deloache, & Rissman, 1975), but because thetotal amount of looking was fixed to 2 min,the only freely varying parameter was the num-ber of trials needed to reach the criterion. Ananalysis of this variable showed no differencebetween experimental conditions, Mnonreferential = 21.10,SDnonreferential = 2.60, Mreferential = 20.26, SDreferential =2.31, t(76) = 1.50, p = .14.

Overall Patterns

Was phonetic discrimination of the tone contrast(as measured in the test phase) improved in the ref-


erential condition relative to the nonreferential one?A mixed analysis of variance (ANOVA) with abetween-subjects factor of condition (nonreferentialor referential) and a within-subjects factor of trialtype (Alt or Non-Alt trials) revealed no significantmain effects or interactions (as with all ANOVAcomparisons here, alpha = .05). This suggested nei-ther an overall effect of phonetic discrimination norany differences across conditions.

To test our hypothesis that the lexical knowl-edge of young infants might interact with pho-netic learning, two additional analyses wereconducted. First, parental reports of their infants’knowledge of the pretraining items (i.e., banana,car, or keys from the referential condition) wereexamined. It was hypothesized that infants mightonly benefit from the referential manipulation ifthey knew at least one of the words presented inpretraining phase (see also Fennell & Waxman,2010). Second, the overall vocabulary scores ofinfants were examined. It was hypothesized thatbetter word learners should also be better atunderstanding referential contexts in general, irre-spective of knowing the specific pretrainingwords. One possibility is that simply seeing afamiliar object on the screen and hearing a labelmay have cued certain individuals (i.e., lexicallysavvy infants) to the fact that they were in a“word learning” context. Both analyses aredescribed separately below.

Comprehension of Pretraining Words

A preliminary analysis examined the number ofpretraining words that infants were reported tohave known from the CDI. Slightly less than half ofthe parents in our total sample reported knowledgeof banana, car, or keys (n = 31 reported that theirinfant knew one word, n = 1 knew two words, andn = 2 knew all three words). The proportion ofinfants knowing at least one pretraining worddid not differ between groups, Mnonreferential = .38,SDnonreferential = .49, Mreferential = .50, SDreferential = .51,t(76) = 1.12, p = .27.

An ANOVA with between-subjects factors ofpretraining (understood no pretraining words, orunderstood at least one pretraining word) andcondition, as well as a within-subjects factor of trialtype showed no significant effects. Results thus sug-gested that parental reports of whether their infantsunderstood any of the pretraining words did notbear on tone discrimination, and this did not signif-icantly change according to the condition in whichinfants were tested.

Overall Vocabulary

A preliminary analysis examined overall CDIscores. These scores did not significantly differbetween groups, Mnonreferential = 8.10, SDnonreferential =6.61, Mreferential = 12.31, SDreferential = 15.07, t(77) =1.61, p = .11, but differences in means and variancesbetween conditions were sizable. A further analysisshowed that these differences stemmed from ahighly non-normal, and positively skewed, distribu-tion of raw scores (common with checklists; see alsoFenson et al., 2000), and in neither condition wasthere a normal distribution, S-Wnonreferential(40) = .91,p = .004, S-Wreferential(38) = .73, p < .001. Threedifferent analyses were conducted to correct for thisproblem.

A categorical analysis. The first method was a cat-egorical analysis, which assigned infants to groupsbased on the overall median score (seven words),which was the same median within both experi-mental conditions. Infants with CDI scores belowthe median were counted as having low vocabular-ies, whereas the others were counted as havinghigh vocabularies. A preliminary ANOVA on thenumber of trials needed to complete the trainingphase with between-subjects factors of vocabulary(low or high vocabulary) and condition revealed nosignificant effects.

The critical ANOVA examined looking timesduring the test phase with between-subjects factorsof vocabulary and condition, and a within-subjects factor of trial type. This showed only a sig-nificant three-way interaction, F(1, 74) = 12.22,p = .029, g2 = .063, such that tone discriminationdiffered as a function of vocabulary group and con-dition. Follow-up two-way ANOVAs were conductedwithin each condition.

In the nonreferential condition, no significanteffects or interactions were found. In the referentialcondition, only the interaction between vocabularyand trial type was significant, F(1, 36) = 5.56,p = .024, g2 = .13, indicating that looking times toAlt and Non-Alt trials differed in low- versus high-vocabulary groups. Planned comparisons withineach level of vocabulary (Bonferroni-corrected,pooled variance) showed that looking times to thetwo types of test trials were not different in thelow-vocabulary group, t(37) = .64, p = .53, but thatinfants looked significantly longer at Alt comparedto Non-Alt trials in the high-vocabulary group,t(37) = 2.55, p = .015, d = .41. Results thus showedthat only high-vocabulary infants succeeded at tonediscrimination after referential training, but low-vocabulary infants did not (Figure 2).


Removing outliers. The second method was toremove outliers. A cutoff of 20 on the CDI was cho-sen because it was the lowest threshold at which nooutliers were present (i.e., all scores were within the1.5 interquartile ranges). As shown in Figures 3aand 3b, outlier-removed CDI scores did not differbetween groups, Mnonreferential = 7.54, SDnonreferential =5.65, Mreferential = 6.73, SDreferential = 4.60, t(70) = .66,p = .51.

A hierarchical multiple regression was con-ducted on “discrimination scores” from the testphase (i.e., looking to Alt trials–looking to Non-Alttrials), using outlier vocabulary (outlier-removedCDI scores), condition, and their interaction aspredictors. A reduced model including justoutlier vocabulary and condition was marginallyimproved by addition of their interaction,F(1, 68) = 3.16, p = .080, change in R2 = .043,although the complete model remained nonsignifi-cant, F(3, 68) = 1.76, p = .16. These results are

weak, but suggest that the interaction betweenvocabulary scores and referential context couldexplain some of the variance in discriminationscores beyond the portion explained by vocabularyor referential context alone. As seen in Figures 3aand 3b, discrimination scores were not correlatedwith outlier-removed CDI scores in the nonreferen-tial condition, r(39) = .001, p = .99, but these vari-ables were marginally correlated in the referentialcondition, r(33) = .30, p = .088.

Normalizing vocabulary scores. The third methodwas to transform CDI scores. In this analysis, aln(x + c) transform was applied: ln was the naturallog, x was the raw CDI score, and c was a constant.As several infants had CDI scores of zero, a constanthad to be added (the constant 2 was the lowest inte-ger that resulted in a normal distribution of scores,S-W(78) = .98, p = .15, but identical statisticalpatterns are found with either 1 or 2). As shownin Figures 3c and 3d, log-transformed scores did

-8

-6

-4

-2

0

2

4

6

8

10

0 12 24 36 48

Alt

min

us

No

n-A

lt L

oo

kin

g (

s)

Vocabulary: Raw CDI Scores

A) Nonreferential: Removing Outliers

Included Excluded

r(39) = .001

-8

-6

-4

-2

0

2

4

6

8

10

0 12 24 36 48

Alt

min

us

No

n-A

lt L

oo

kin

g (

s)

Vocabulary: Raw CDI Scores

B) Referential: Removing Outliers

Included Excluded

r(33) = .30^

-8

-6

-4

-2

0

2

4

6

8

10

0.5 1 1.5 2 2.5 3 3.5 4 A

lt m

inu

s N

on

-Alt

Lo

okin

g (

s)

Vocabulary: Log-Transformed CDI Scores

D) Referential: Normalizing Scores

LN (X + Constant)

r(38) = .37*

-8

-6

-4

-2

0

2

4

6

8

10

0.5 1 1.5 2 2.5 3 3.5 4

Alt

min

us

No

n-A

lt L

oo

kin

g (

s)

Vocabulary: Log-Transformed CDI Scores

C) Nonreferential: Normalizing Scores

LN (X + Constant)

r(40) = -.051

Figure 3. Results from Experiment 2 showing the relation between phonetic learning and vocabulary, controlling for skewedMacArthur Communicative Development Inventories (CDI) scores both by removing outliers, and by normalizing CDI scores. Plotsshow discrimination scores on the y-axis (looking to alternating trials minus looking to nonalternating trials) against: (a) raw CDI scores(outliers removed) in the nonreferential condition, (b) raw CDI scores (outliers removed) in the referential condition, (c) normalizedscores (log-transformed) in the nonreferential condition, and (d) normalized scores (log-transformed) in the referential condition. They-axis always crosses just below the median CDI score (seven words), and the solid lines indicate trend lines for each plot. Pearson coef-ficients are indicated.^p < .10. *p < .05.


not differ between groups, Mnonreferential = 2.08,SDnonreferential = .73, Mreferential = 2.23, SDreferential = .77,t(76) = .90, p = .37.

Another hierarchical multiple regression wasconducted on discrimination scores from the testphase, using log vocabulary (log-transformed CDIscores), condition, and their interaction as predic-tors. A reduced model including just log vocabu-lary and condition was significantly improved byaddition of their interaction, F(1, 74) = 4.95,p = .029, change in R2 = .060, and the completemodel was significant, F(3, 74) = 2.22, p = .038,R2 = .11. This analysis more clearly shows that theinteraction between vocabulary size and referentialcontext explains a unique portion of the variance indiscrimination scores beyond the portion explainedby vocabulary or referential context alone. As seenin Figures 3c and 3d, discrimination scores werenot correlated with log-transformed CDI scores inthe nonreferential condition, r(40) = �.051, p = .75,but were significantly correlated in the referentialcondition, r(38) = .37, p = .022.

Discussion

Results suggest that phonetic learning (asassessed by tone discrimination) was affected byboth the referential context of training and infants’overall vocabulary scores. This was shown in threeanalyses, correcting for a non-normal and positivelyskewed distribution of vocabulary scores. First, acategorical analysis showed that infants receivingthe referential training had improved tone discrimi-nation—but only if they also had high vocabularies(i.e., scores at the median or higher)—whereasinfants in the nonreferential condition showed nolearning, regardless of vocabulary size (Figure 2).The second and third analyses used regression toshow that an individual infants’ success at tone dis-crimination was predicted by vocabulary size(using either raw or normalized scores), but only inthe referential condition (Figure 3). Together, theseresults provide convincing evidence that tone dis-crimination is preferentially improved (a) when areferential context helps to indicate that two objectlabels are distinguished by tones, and (b) wheninfants have greater overall vocabulary sizes. Theeffects of referential context and vocabulary are dis-cussed separately below.

Nonreferential Versus Referential Contexts

Previous findings show that the perceptual dis-tinctiveness of speech sounds is enhanced in

acquired distinctiveness contexts (Yeung & Werker,2009). The current results indicate that the nature ofthese contexts matters. Specifically, 9-month-oldinfants—at least those with high vocabularies—paidspecial attention to the acoustic properties that dis-tinguished two (linguistically important) soundswhen those sounds were potential word labels.Infants did not differentiate these same acousticproperties when they simply co-occurred with thetwo objects. Theoretically, this finding is closelyrelated to the notion that words play a special rolein both adult (Lupyan, 2012) and infant (Ferryet al., 2010) categorization. In that literature, visualfeatures of object categories are learned more easilywhen referential cues (i.e., labels) indicate categorymembership, compared to when nonreferential cues(i.e., tones, spatial positions) indicate the same.Here, we similarly show that the acoustic featuresof the labels themselves are learned more easily whenreferential cues (i.e., contrasting word labels) indi-cate category membership, compared to when non-referential cues (i.e., co-occurring visual contexts)indicate the same.

It remains important to note that the currentwork is different from—although perhaps relatedto—the literature on word learning. In that litera-ture, infants’ ability to map phonetic forms ontonovel object referents is difficult before 13–14 months of age (e.g., Hollich et al., 2000;Werker, Cohen, Lloyd, Casasola, & Stager, 1998),particularly when the phonetic forms to be mappedare similar (Stager & Werker, 1997; although seeFennell & Waxman, 2010). In this study, it istherefore unlikely that 9-month-old infants arelearning new lexical entries (successfully mappingnovel word forms onto their referents). Resultssuggest instead that some infants—those withhigh vocabularies—are better able to identify andcompare the phonetic content of speech soundswhen those sounds are possible word labels.These infants then used this newly acquired pho-netic knowledge to discriminate a perceptual pat-tern normally considered to be non-native in theirlanguage environment.

Effects of Overall Vocabulary Size

The above discussion raises another question:Why did vocabulary size interact with the referen-tial nature of the training? One hypothesis is thathaving a larger vocabulary size increased the chancethat infants had banana, keys, and/or car in their lex-icon, which might have been a prerequisite for rec-ognizing the referential context. This hypothesis


was not supported because the reported compre-hension of these individual items did not predictdiscrimination performance. However, even thoughCDIs are quite accurate at the item level by at least20 months of age (Ring & Fenson, 2000), it remainspossible that parental reports do not accuratelyreflect infants’ specific knowledge of individualwords at 9 months of age. Without any indepen-dent means of verifying the comprehension of thesethree particular words, it is thus difficult to defini-tively exclude the possibility that overall vocabularywas actually a better estimate of whether an infantreally understood banana, keys, or car than CDIreports of those specific items.

We consider another hypothesis to be morelikely, however. It may be the case that specificknowledge of the three pretraining words was notthe true source of infants’ referential understandingin the critical condition. Rather, simply seeingfamiliar objects and hearing an unknown labelbears striking similarity to other ostensive scenariosthat infants may have previously experienced. Con-sider book-reading scenarios, where infants arecommonly shown familiar objects whose labels theymay not yet have learned. Here, they are repeatedlypresented with the corresponding word label (e.g.,books showing familiar objects beginning withletters A–Z). Familiarity with activities like thesemay just be enough to help linguistically savvy 9-month-olds recognize the referential condition as aword learning “game,” where (phonetically contras-tive) word forms are heard. This may be where thecrucial difference between the referential and non-referential conditions lies because in the latter con-dition the referential nature of the object–tonepairings was not as transparent.

This latter hypothesis entails two distinct possi-bilities that further explain the interaction betweenthe experimental condition and vocabulary size.The first possibility is that two independent factorsacted to improve phonetic learning, pushing therecognition of distinct tone categories just abovethreshold only in the referential condition. One fac-tor is the top-down benefit from being trained withcontrasting word forms in a referential context, asdiscussed earlier (e.g., Lupyan, 2012). The other fac-tor is that infants who are better phonetic learnersmay also have had higher vocabularies, a proposalrelated to studies prospectively correlating lan-guage-specific phonetic sensitivity at younger ageswith larger vocabularies at older ages (Kuhl, Con-boy, Padden, Nelson, & Pruitt, 2005; Tsao, Liu, &Kuhl, 2004). As learning tone distinctions is quitedifficult for English-learning infants at the age

tested here, discrimination may only have been pos-sible when both of these factors conspired to aidlearning. That is, only good phonetic learners (i.e.,infants with high vocabularies), who were alsotrained in a referential context, succeeded at thetask.

The second possibility is that vocabulary size isrelated to infants’ abilities at recognizing the refer-ential context. For example, high-vocabulary infantsmight be exposed more frequently to ostensivelabeling situations like book reading, a reasonablehypothesis given correlations between vocabularysize and these sorts of infant–parent interactions(e.g., Raikes et al., 2006). Provocatively, this pre-dicts that referential understanding (as measuredby one’s overall vocabulary size) is one cause ofvariance in phonetic sensitivity, or at least of thevariance in infants’ ability to learn new phoneticcontrasts. That is, those infants with precocious lexi-cal knowledge may be better at identifying referen-tial contexts, and thus be better at recognizingrelevant perceptual contrasts between distinct wordlabels. Future work will need to further explore therelation between referential abilities, vocabulary,and phonetic sensitivity, but one prediction is thatsome measure of lexical or referential knowledge(one more appropriate for infants under 12 monthsof age than parent checklists) can prospectivelypredict infants who are better at native phoneticcontrasts (and worse at non-native phonetic con-trasts). Preliminary evidence for this hypothesiscomes from recent work, showing that the pace ofphonetic development can be predicted by the styleof mother–child interactions (Elsabbagh et al., 2013),which is then likely correlated with vocabularysize.

General Discussion

Young infants acquire native language phoneticpatterns defined by the lexicon, but this happenslong before infants have learned more than a hand-ful of words. Statistical (i.e., distributional) mecha-nisms of phonetic learning have been proposed(e.g., Cristia et al., 2011; Maye et al., 2002; Mc-Murray et al., 2009), which do not rely on an infant’slexical knowledge. Infants may also use informationfrom the unique (acquired distinctiveness) contextsin which phonetic tokens are heard, which couldsupplement statistical learning (Feldman et al.,2013; Swingley, 2009; Yeung & Werker, 2009).

A weakness of the acquired distinctivenessapproach is that the notion of informative visual or


auditory contexts is unconstrained. For example,many phonetically irrelevant acoustic properties con-sistently appear in unique contexts that are notlexically distinctive (see also the discussion of allo-phones in Feldman et al., 2013). In the visualdomain, for example, recent work reports adultphonetic learning when talker identity distinguishesa phonetic contrast (Mani & Schneider, 2013). Otherwork has also shown that infants are consistentlyexposed to unique pitch deviations in certain care-giving contexts, which have pragmatic meanings,rather than lexical ones (e.g., Stern, Spieker, &MacKain, 1982). An unsophisticated computationalmodel might mistakenly consider all such contextsto be distinctive, yet human infants successfullyattune to phonetic contrasts that generalize overlexically irrelevant acoustic information such astalker identity (e.g., Kuhl, 1983) and pitch (e.g.,Singh et al., 2008). From a theoretical point of view,acquired distinctiveness can thus only be effective ifinfants also recognize the uniquely referential con-texts where relevant phonetic contrasts are made.

This study suggests one kind of constraint onlearning from acquired distinctiveness: infants’understanding of the way that words are used(Heitner, 2004), as observed in their referentialunderstanding of labeling contexts. A proceduresimilar to that reported by Yeung and Werker(2009) was run here, where consistent pairingsbetween distinct objects and speech sounds high-lighted a difficult tone contrast. In Experiment 1,infants failed to learn the contrast from this train-ing, showing no evidence of tone discrimination. InExperiment 2, infants succeeded at tone discrimina-tion, but only when trained in a more referentialcontext (i.e., it was made clearer that the object–tone pairings were labels, and not simply co-occur-rences), although this effect was seen only in infantswith higher vocabularies.

Two possibilities describe how referential contextand vocabulary size might interact. The first is thata referential context boosts tone discrimination, asdoes having a high vocabulary (i.e., those infantsare better phonetic learners). Together, these inde-pendent factors pushed only those high-vocabularyinfants in the referential condition above threshold.The second possibility is that having a high vocabu-lary allows infants to more easily understand thereferential manipulation. That is, having a highvocabulary is likely associated with a better under-standing of referential contexts (i.e., knowing thatword labels are heard when seeing objects andhearing speech). Future research will need to distin-guish these possibilities, but both implicate lexical

knowledge (vocabulary size) and referential knowl-edge (sensitivity to labeling contexts) in phoneticlearning.

Word learning and phonetic attunement arerarely considered mutually influential, as it iswidely assumed that the latter influences the for-mer, and not the other way around. This assump-tion stems from the fact that the most active periodof attunement occurs in the 1st year of life, beforeinfants have learned many words. In conjunctionwith other work, however, there is increasingevidence that—in addition to statistical learning—lexical and referential factors can affect the learningof perceptual categories from early in development.For example, word labels appear to preferentiallyaid infants’ detection of category features (forobjects) from at least 3 to 4 months of age (Ferryet al., 2010), and this study reports a similarphenomenon in the phonetic domain at 9 monthsof age. Moreover, infants from 6 to 12 months ofage have already begun to accumulate several lexi-cal items (Bergelson & Swingley, 2012; Tincoff &Jusczyk, 1999, 2012), an indication that infants inthis age range already have some knowledge ofhow referential labels work.

All proposed mechanisms of phonetic develop-ment, including statistical, bottom-up approaches,depend crucially on the structure of the lexicon.However, only a few previous accounts of phoneticlearning (Heitner, 2004; Werker et al., 2012) hadalso proposed that phonetic learning might also beinfluenced by lexical and referential knowledge onthe part of the infants themselves. The currentresults suggest that these proposals have some trac-tion. However, the present findings also highlightthe need for broader research that tackles twofurther issues.

One issue is a computational one. Is acquireddistinctiveness a mechanism that depends strictlyon minimal pairs, which are thought to be rela-tively rare in parental input? Previous work usingan acquired distinctiveness approach (Feldmanet al., 2013; Thiessen, 2011) has shown that infantscan use a more probabilistic learning mechanism,learning phonetic contrasts (e.g., between ah [ɑ] andaw [ɔ] vowels) after be presented with nonminimalpairs of word labels (e.g., gutah and litaw). Itremains to be seen whether this kind of phoneticlearning from nonminimal pairs is seen in objectlabeling situations, and if so, if it is similarly subjectto the kinds of referential constraints observed here.A second future research topic concerns the preciserelation between lexical knowledge (i.e., the numberof lexical entries) and referential abilities (i.e., the


capacity to understand social and ostensive cueswhen using words), particularly when phoneticattunement is most active in the 1st year of life. Abetter understanding of this process may help inter-pret the mediating role that lexical knowledge hadin the currently reported studies where phoneticlearning was modulated by labeling objects.

References

Bergelson, E., & Swingley, D. (2012). At 6-9 months,human infants know the meanings of many commonnouns. Proceedings of the National Academy of Sciences,109, 3253–3258. doi:10.1073/pnas.1113380109

Best, C. T., McRoberts, G. W., & Goodell, E. (2001). Dis-crimination of non-native consonant contrasts varyingin perceptual assimilation to the listener’s nativephonological system. Journal of the Acoustical Society ofAmerica, 109, 775–794. doi:10.1121/1.1332378

Coady, J. A., & Aslin, R. N. (2003). Phonological neigh-bourhoods in the developing lexicon. Journal of ChildLanguage, 30, 441–469. doi:10.1017/S0305000903005579

Cohen, L. B., Deloache, J. S., & Rissman, M. W. (1975).The effect of stimulus complexity on infant visualattention and habituation. Child Development, 46,611–617.

Cristia, A. (2011). Fine-grained variation in caregivers’/s/predicts their infants’/s/category. Journal of the Acousti-cal Society of America, 129, 3271–3280. doi:10.1121/1.3562562

Cristia, A., McGuire, G. L., Seidl, A., & Francis, A. L.(2011). Effects of the distribution of acoustic cues oninfants’ perception of sibilants. Journal of Phonetics, 39,388–402. doi:10.1016/j.wocn.2011.02.004

Elsabbagh, M., Hohenberger, A., Campos, R., VanHerwegen, J., Serres, J., De Schonen, S., . . . & Karmiloff-Smith, A. (2013). Narrowing perceptual sensitivity to thenative language in infancy: Exogenous influences ondevelopmental timing. Behavioral Sciences, 3, 120–132.doi:10.3390/bs3010120

Feldman, N. H., Myers, E. B., White, K. S., Griffiths, T. L.,& Morgan, J. L. (2013). Word-level information influ-ences phonetic learning in adults and infants. Cognition,127, 427–438. doi:10.1016/j.cognition.2013.02.007

Fennell, C. T., & Waxman, S. R. (2010). What paradox?Referential cues allow for infant use of phonetic detailin word learning. Child Development, 81, 1376–1383.doi:10.1111/j.1467-8624.2010.01479.x

Fenson, L., Pethick, S., Renda, C., Cox, J. L., Dale, P. S., &Reznick, J. S. (2000). Short-form versions of the Mac-Arthur Communicative Development Inventories.Applied Psycholinguistics, 21, 95–116. doi:10.1017/S0142716400001053

Ferry, A. L., Hespos, S. J., & Waxman, S. R. (2010). Cate-gorization in 3- and 4-month-old infants: An advantageof words over tones. Child Development, 81, 472–479.doi:10.1111/j.1467-8624.2009.01408.x

Gauthier, B., Shi, R., & Xu, Y. (2007). Simulating theacquisition of lexical tones from continuous dynamicinput. Journal of the Acoustical Society of America, 121,EL190–EL195. doi:10.1121/1.2716160

Gogate, L. J., & Bahrick, L. E. (2001). Intersensory redun-dancy and 7-month-old infants’ memory for arbitrary syl-lable-object relations. Infancy, 2, 219–231. doi:10.1207/S15327078IN0202_7

Gogate, L. J., Bahrick, L. E., & Watson, J. D. (2000). Astudy of multimodal motherese: The role of temporalsynchrony between verbal labels and gestures. ChildDevelopment, 71, 878–894. doi:10.1111/1467-8624.00197

Hayes-Harb, R. (2007). Lexical and statistical evidence inthe acquisition of second language phonemes. SecondLanguage Research, 23, 1–31.

Heitner, R. M. (2004). The cyclical ontogeny of ontology:An integrated developmental account of object andspeech categorization. Philosophical Psychology, 17, 45–57.doi:10.1080/0951508042000202372

Hollich, G. J., Hirsh-Pasek, K., Golinkoff, R. M., Rebecca,J., Brown, E., Chung, H. L., . . . & Bloom, L., (2000).Breaking the language barrier: An emergentist coalitionmodel for the origins of word learning. Monographs ofthe Society for Research in Child Development, 65(SerialNo. 262).

Jakobson, R. (1968). Child language, aphasia, and phonologi-cal universals (A. R. Keiler, Trans.). Oxford, UK: Mou-ton. (Original work published 1942)

Jusczyk, P. W. (1993). From general to language-specificcapacities: The WRAPSA model of how speech percep-tion develops. Journal of Phonetics, 21, 3–28.

Khouw, E., & Ciocca, V. (2007). Perceptual correlates ofCantonese tones. Journal of Phonetics, 35, 104–117.doi:10.1016/j.wocn.2005.10.003

Kluender, K. R., Lotto, A. J., Holt, L. L., & Bloedel, S. L.(1998). Role of experience for language-specific func-tional mappings of vowel sounds. Journal of the Acousti-cal Society of America, 104, 3568–3582.

Kuhl, P. K. (1983). Perception of auditory equivalence clas-ses for speech in early infancy. Infant Behavior & Devel-opment, 6, 263–285. doi:10.1016/S0163-6383(83)80036-8

Kuhl, P. K. (1993). Innate predispositions and the effectsof experience in speech perception: The native languagemagnet theory. In B. de Boysson-Bardies, S. de Scho-nen, P. W. Jusczyk, P. McNeilage, & J. Morton (Eds.),Developmental neurocognition: Speech and face processingin the first year of life (pp. 259–274). Dordrecht, Nether-lands: Kluwer.

Kuhl, P. K., Conboy, B. T., Padden, D., Nelson, T., &Pruitt, J. (2005). Early speech perception and laterlanguage development: Implications for the “criticalperiod.” Language Learning and Development, 1, 237–264.doi:10.1080/15475441.2005.9671948

Kuhl, P. K., Tsao, F., & Liu, H. (2003). Foreign-languageexperience in infancy: Effects of short-term exposureand social interaction on phonetic learning. Proceedingsof the National Academy of Sciences, 100, 9096–9101.doi:10.1073/pnas.1532872100


Lawrence, D. H. (1949). Acquired distinctiveness of cues:I. Transfer between discriminations on the basis offamiliarity with the stimulus. Journal of ExperimentalPsychology, 39, 770–784.

Lupyan, G. (2012). What do words do? Toward a theoryof language-augmented thought. In B. H. Ross (Ed.),The psychology of learning and motivation (Vol. 57, pp.255–297). New York, NY: Academic Press. doi:10.1016/B978-0-12-394293-7.00007-8

Lupyan, G., Rakison, D. H., & McClelland, J. L. (2007).Language is not just for talking: redundant labels facili-tate learning of novel categories. Psychological Science,18, 1077–1083. doi:10.1111/j.1467-9280.2007.02028.x

Mani, N., & Schneider, S. (2013). Speaker identity sup-ports phonetic category learning. Journal of ExperimentalPsychology: Human Perception and Performance, 39, 623–629. doi:10.1037/a0030402

Mattock, K., & Burnham, D. (2006). Chinese and Englishinfants’ tone perception: Evidence for perceptual reor-ganization. Infancy, 10, 241–265. doi:10.1207/s15327078in1003_3

Mattock, K., Molnar, M., Polka, L., & Burnham, D. (2008).The developmental course of lexical tone perception inthe first year of life. Cognition, 106, 1367–1381. doi:10.1016/j.cognition.2007.07.002

Maye, J., Weiss, D. J., & Aslin, R. N. (2008). Statisticalphonetic learning in infants: Facilitation and featuregeneralization. Developmental Science, 11, 122–134. doi:10.1111/j.1467-7687.2007.00653.x

Maye, J., Werker, J. F., & Gerken, L. (2002). Infant sensi-tivity to distributional information can affect phoneticdiscrimination. Cognition, 82, B101–B111. doi:10.1016/S0010-0277(01)00157-3

McMurray, B., Aslin, R. N., & Toscano, J. C. (2009).Statistical learning of phonetic categories: Insights froma computational approach. Developmental Science, 12,369–378. doi:10.1111/j.1467-7687.2009.00822.x

Narayan, C. R., Werker, J. F., & Beddor, P. S. (2010). Theinteraction between acoustic salience and language expe-rience in developmental speech perception: Evidencefrom nasal place discrimination. Developmental Science,13, 407–420. doi:10.1111/j.1467-7687.2009.00898.x

Raikes, H., Pan, B. A., Luze, G., Tamis-LeMonda, C. S.,Brooks-Gunn, J., Constantine, J., Banks Tarullo, L., et al.(2006). Mother–child bookreading in low-incomefamilies: Correlates and outcomes during the firstthree years of life. Child Development, 77, 924–953.doi:10.1111/j.1467-8624.2006.00911.x

Ring, E. D., & Fenson, L. (2000). The correspondencebetween parent report and child performance for recep-tive and expressive vocabulary beyond infancy. FirstLanguage, 20, 141–159. doi:10.1177/014272370002005902

Rivera-Gaxiola, M., Silva-Pereyra, J., & Kuhl, P. K. (2005).Brain potentials to native and non-native speech con-trasts in 7- and 11-month-old American infants. Devel-opmental Science, 8, 162–172. doi:10.1111/j.1467-7687.2005.00403.x

Salminen, N. H., Tiitinen, H., & May, P. J. C. (2009). Mod-eling the categorical perception of speech sounds: Astep toward biological plausibility. Cognitive, Affective &Behavioral Neuroscience, 9, 304–313. doi:10.3758/CABN.9.3.304

Sato, Y., Kato, M., & Mazuka, R. (2012). Development ofsingle/geminate obstruent discrimination by Japaneseinfants: Early integration of durational and nondura-tional cues. Developmental Psychology, 48, 18–34. doi:10.1037/a0025528

Scott, L. S., Pascalis, O., & Nelson, C. A. (2007). Adomain-general theory of the development of percep-tual discrimination. Current Directions in PsychologicalScience, 16, 197–201. doi:10.1111/j.1467-8721.2007.00503.x

Singh, L., White, K. S., & Morgan, J. L. (2008). Building aword-form lexicon in the face of variable input: Influ-ences of pitch and amplitude on early spoken word rec-ognition. Language Learning and Development, 4, 157–178.doi:10.1080/15475440801922131

Stager, C. L., & Werker, J. F. (1997). Infants listen formore phonetic detail in speech perception than inword-learning tasks. Nature, 388, 381–382. doi:10.1038/41102

Stern, D. N., Spieker, S., & MacKain, K. (1982). Intonationcontours as signals in maternal speech to prelinguisticinfants. Developmental Psychology, 18, 727–735. doi:10.1037//0012-1649.18.5.727

Swingley, D. (2009). Contributions of infant word learn-ing to language development. Philosophical Transactionsof the Royal Society B: Biological Sciences, 364, 3617–3632.doi:10.1098/rstb.2009.0107

Teinonen, T., Aslin, R. N., Alku, P., & Csibra, G. (2008).Visual speech contributes to phonetic learning in6-month-old infants. Cognition, 108, 850–855. doi:10.1016/j.cognition.2008.05.009

Thiessen, E. D. (2011). When variability matters more thanmeaning: The effect of lexical forms on use of phonemiccontrasts. Developmental Psychology, 47, 1448–1458. doi:10.1037/a0024439

Tincoff, R., & Jusczyk, P. W. (1999). Some beginningsof word comprehension in 6-month-olds. PsychologicalScience, 10, 172–175. doi:10.1111/1467-9280.00127

Tincoff, R., & Jusczyk, P. W. (2012). Six-month-olds com-prehend words that refer to parts of the body. Infancy,17, 432–444. doi:10.1111/j.1532-7078.2011.00084.x

Tsao, F., Liu, H., & Kuhl, P. K. (2004). Speech perceptionin infancy predicts language development in the secondyear of life: A longitudinal study. Child Development, 75,1067–1084. doi:10.1111/j.1467-8624.2004.00726.x

Vallabha, G. K., McClelland, J. L., Pons, F., Werker, J. F.,& Amano, S. (2007). Unsupervised learning of vowelcategories from infant-directed speech. Proceedings of theNational Academy of Sciences of the United States of Amer-ica, 104, 13273–13278. doi:10.1073/pnas.0705369104

Werker, J. F., Cohen, L. B., Lloyd, V. L., Casasola, M., &Stager, C. L. (1998). Acquisition of word-object associations


by 14-month-old infants. Developmental Psychology, 34,1289–1309.

Werker, J. F., Pons, F., Dietrich, C., Kajikawa, S., Fais, L., &Amano, S. (2007). Infant-directed speech supportsphonetic category learning in English and Japanese. Cog-nition, 103, 147–162. doi:10.1016/j.cognition.2006.03.006

Werker, J. F., & Tees, R. C. (1984). Cross-language speechperception: Evidence for perceptual reorganization dur-ing the first year of life. Infant Behavior and Development,7, 49–63.

Werker, J. F., Yeung, H. H., & Yoshida, K. A. (2012).How do infants become experts at native-speech per-ception? Current Directions in Psychological Science, 21,221–226. doi:10.1177/0963721412449459

Yeung, H. H., Chen, K. H., & Werker, J. F. (2013).When does native language input affect phonetic per-ception? The precocious case of lexical tone. Journal ofMemory and Language, 68, 123–139. doi:10.1016/j.jml.2012.09.004

Yeung, H. H., & Werker, J. F. (2009). Learning words’sounds before learning how words sound: 9-month-olds use distinct objects as cues to categorize speechinformation. Cognition, 113, 234–243. doi:10.1016/j.cog-nition.2009.08.010

Yoshida, K. A., Pons, F., Maye, J., & Werker, J. F. (2010).Distributional phonetic learning at 10 months ofage. Infancy, 15, 420–433. doi:10.1111/j.1532-7078.2009.00024.x


referential labeling can facilitate phonetic learning in...

Documents