Phonological specificity of vowels and consonants in early ...

Journal of

www.elsevier.com/locate/jml

Journal of Memory and Language 57 (2007) 252–272

Memory andLanguage

Phonological specificity of vowels and consonantsin early lexical representations

Nivedita Mani *, Kim Plunkett

Department of Experimental Psychology, Oxford University, South Parks Road, Oxford OX1 3UD, UK

Received 8 August 2006; revision received 21 March 2007Available online 3 May 2007

Abstract

Infants become selectively sensitive to phonological distinctions relevant to their native language at an early age. Onemight expect that infants bring some of this phonological knowledge to bear in encoding the words they subsequentlyacquire. In line with this expectation, studies have found that 14-month-olds are sensitive to mispronunciations of initialconsonants of familiar words when asked to identify a referent. However, there is very little research investigatinginfants’ sensitivity to vowels in lexical representations. Experiment 1 examines whether infants at 15, 18 and 24 monthsare sensitive to mispronunciations of vowels in familiar words. The results provide evidence for vowels constraininglexical recognition of familiar words. Experiment 2 compares 15, 18 and 24-month-olds’ sensitivity to consonantand vowel mispronunciations of familiar words in order to assess the relative contribution of vowels and consonantsin constraining lexical recognition. Our results suggest a symmetry in infants’ sensitivity to vowel and consonantmispronunciations early in the second year of life.� 2007 Elsevier Inc. All rights reserved.

Keywords: Phonological specificity; Vowels; Consonants; Lexical representation

Introduction

New-born infants are precocious in their ability todiscriminate phonetic contrasts not present in theirnative language (Eimas, Siqueland, Jusczyk, & Vigoriti,1971; Kuhl, 1987). However, infants are no longer ableto discriminate non-native phonetic contrasts at theend of their first year of life (Werker & Tees, 1984; Wer-ker & Lalonde, 1988). Since infants also begin to pro-duce and understand words around this age, it hasbeen argued that this loss of language-universal phonetic

0749-596X/$ - see front matter � 2007 Elsevier Inc. All rights reserv

doi:10.1016/j.jml.2007.03.005

* Corresponding author.E-mail address: [email protected] (N. Mani).

discrimination is caused by the association of soundswith meaning (Werker & Tees, 1984). NeverthelessKuhl, Williams, Lacerda, Stevens, and Lindblom(1992) report that infants display greater sensitivity tonative than to non-native vowels by 6 months—wellbefore an understanding of word meaning has devel-oped. Indeed, recent research by Maye, Werker, andGerken (2002) finds that exposure to the distributionof the speech sounds in the infants’ native languagehas an extremely influential role in driving this prefer-ence for native language speech sounds—suggesting aless crucial role for the introduction of word-object asso-ciations. A part of the infant’s phonological repertoireappears to be in place before lexical acquisition is setin motion. Consequently, this phonological repertoire

ed.

mailto:[email protected]

N. Mani, K. Plunkett / Journal of Memory and Language 57 (2007) 252–272 253

might actually facilitate early lexical development, byproviding the means to discriminate words.

Previous studies have shown that infants’ lexical rep-resentations are at least partially phonologically speci-fied. These studies report that infants look longer at atarget object when its label is correctly pronounced thanwhen the label is mispronounced, even when the mispro-nunciation involves a single featural change. A mispro-nunciation effect has been reported for word-initialconsonants in familiar words as early as 14 months (Bai-ley & Plunkett, 2002; Ballem & Plunkett, 2005; Swingley& Aslin, 2000, 2002) and word-medial consonants at19 months (Swingley, 2003). This suggests that infantsdo not appear to encode words vaguely. A possibleinterpretation of this finding is that infants’ phonologi-cal representations are similar to adult representationsof familiar words—fully specified in phonetic detail.Although this adult-oriented perspective requires furtherconfirmation of infant sensitivity across a full range ofconsonantal feature changes, it is noteworthy that themajority of previous studies have only tested infants’sensitivity to mispronunciations of consonants in lexicalrecognition or word-learning tasks (Bailey & Plunkett,2002; Ballem & Plunkett, 2005; Fennell & Werker,2003; Stager & Werker, 1997). The role of vowels ininfant’s recognition of familiar words has not been sim-ilarly explored. The only studies to include vowel mis-pronunciations were Swingley and Aslin (2000, 2002).These studies analysed 14-month-old (Swingley & Aslin,2002) and 18- to 23-month-old infants’ (Swingley &Aslin, 2000) sensitivity to mispronunciations of sixwords, two of which were vowel mispronunciations,while the rest were consonant mispronunciations.Infants were presented with two pictures for 3 s, afterwhich one of the images was named in the carrier phrase‘Where’s the [target]?’ Swingley and Aslin report thatinfants at both age groups took longer to switch fromthe distracter to the target when the target label wasmispronounced than when it was correctly pronounced.However, a significant bias towards the target imageremained even following mispronunciations. This effectwas also significant by items, suggesting that there wereno differences between infants’ sensitivity to vowel andconsonant mispronunciations.

Swingley and Aslin do not provide a systematic anal-ysis of infants’ sensitivity to vowel mispronunciationsalone. Of the two words that were mispronounced bythe vowel—apple (mispronounced as opple) and car

(mispronounced as cur)—one involved a word-initialvowel mispronunciation in a bisyllabic word, and theother a word-final vowel mispronunciation in a mono-syllabic word. Conversely, all consonant mispronuncia-tions were word-initial in monosyllabic words alone.While this does not invalidate the overall mispronuncia-tion effect reported in both studies (Swingley & Aslin,2002), their results do not provide a general test of infant

sensitivity to vowel mispronunciations given that onlytwo words were mispronounced in an identical fashionfor all infants and the locus of the mispronunciation dif-fered in each word. A more stringent test of infant vowelsensitivity in lexical processing needs to examine agreater range of vowels and words, where the locationof the mispronunciation is carefully controlled.

There are good reasons to suppose that vocalic iden-tity might play a central role in lexical recognition: vow-els carry the major acoustic energy associated withspeech, being longer and louder than consonants andprovide important cues to prosodic structure. Infants’sensitivity to vowels in their native language is acquiredmuch earlier than similar sensitivity to language-specificconsonants (Kuhl et al., 1992). Bond (1954) argues thatvowel changes to words are more likely to lead to low-ered comprehensibility than consonant changes in adults.Finally, Gerken, Murphy, and Aslin (1995) found thatpre-schoolers were more sensitive to vowel changes inbisyllabic words than consonant changes, though no dif-ferences were found for monosyllabic words. These find-ings indicate a prominent role for vowels and consonantsin lexical recognition and suggest that vowel mispronun-ciations in familiar words would have a similar effect oninfant responding as consonant mispronunciations.

In addition, some studies suggest that production ofmost vowels is mastered early, with full mastery overconsonants only appearing later (Templin, 1957). Thelonger duration of vowels compared to consonantsmay make them easier for infants to articulate, basedon infants’ general preference for longer sounds in pro-duction (Smith, 1978). Until recently, not many produc-tion studies had focussed on infants’ age of acquisitionof vowels, because vowels were considered to be mas-tered much earlier than consonants and tended to beaccurately articulated. More recently, however, somestudies have found that infants accurately produce anumber of consonants and vowels at around the sameage—by about 15 months of age, infants appear to havemastered /b, d g, n, h, w/ in their consonant inventory(Robb & Bleile, 1994) and /i, u, o, A, �/ in their vowelinventory (Selby, Robb, & Gilbert, 2000; Stoel-Gam-mon & Herrington, 1990). Further analysis suggests thatinfants have acquired both rounded and unrounded,front and back, and high and low vowels in their vowelinventory. However, infants do not master any voicelessstops or most fricatives until 18–36 months of age. Welack evidence suggesting that infants’ mastery of produc-tion of a particular segment affects their perception ofmispronunciations of this segment. However, if percep-tion sensitivities were to reflect production abilities, wemight expect to observe a heightened sensitivity to vowelmispronunciations compared to consonant mispronun-ciations in early lexical recognition.

An alternative perspective based on typological anal-ysis is ‘that the task of distinguishing lexical items rests

254 N. Mani, K. Plunkett / Journal of Memory and Language 57 (2007) 252–272

more on consonants than on vowels’ (Nespor, Pena, &Mehler, 2003, p. 209). These authors argue that conso-nants are specialised for conveying information aboutthe lexicon whereas vowels provide information aboutprosody and grammar. Previous studies report thatadults are more likely to replace vowels than consonantsin tasks involving changes from non-words to words(Cutler, Sebastian-Galles, Soler-Vilageliu, & van Ooijen,2000; Van Ooijen, 1996). In addition, vowels are per-ceived less categorically than consonants, suggestingsome variability in the perception of vowels (Liberman,Delattre, Cooper, & Gerstman, 1954; Pisoni, 1973) andthe acoustic characteristics of vowels are also very sus-ceptible to inter-speaker differences in production (Pet-erson & Barney, 1952). From this perspective, onemight expect infants to show less sensitivity to mispro-nunciations of vowels than consonants when recognisingthe referent of a familiar word.

A study by Nazzi (2005) offers some support for theview that consonants play a more central role in lexicalacquisition than vowels. Using a name-based categorisa-tion task (Nazzi & Gopnik, 2001), Nazzi reports that 20-month-old French infants can learn simultaneously twonovel words that differ by either the word-initial or non-initial consonants while being unable to learn two wordsthat differ by their vowels. This result supports Nesporet al.’s (2003) claim that the task of distinguishing lexicalitems rests more on consonants than on vowels. It is alsonoteworthy in this context, that Caramazza, Chialant,Capasso, and Micell (2000) found a double dissociationbetween the processing of vowels and consonants inaphasic patients, and argued that this demonstratedcategorically distinct representations of vowels andconsonants that could not be reduced to a featural level.

Overall, the evidence for the phonological specificityof vowels in early lexical representations is equivocal.The studies by Swingley and Aslin (2000, 2002) suggesta degree of phonological specificity of vowels in familiarwords as early as 14-months-old. However, this findingis based on infant responses to just two words (and con-sequently mispronunciations of just two vowels). Hence,the generality of the finding must be questioned. In con-trast, Nazzi’s (2005) study suggests that infants fail todiscriminate vowels in a lexical acquisition task whereasthey do discriminate between consonants. Both the nat-ure of the tasks (named-based categorisation vs. mispro-nunciation) and the status of the lexical items (novel vs.familiar) differ in these studies, so direct comparison isnot straightforward. Nevertheless, the implications ofthe studies are divergent and therefore require empiricalclarification.

We describe two sets of experiments designed to eval-uate infant sensitivity to vowel identity in early lexicalrepresentations. In the first set of experiments, we usea mispronunciation task to evaluate the generality ofSwingley and Aslin’s finding that infants are sensitive

to vowel identity in familiar words. In line with Swingleyand Aslin, we predict that infants will be sensitive tovowel mispronunciations. In a second set of experi-ments, we compare infant sensitivity to vowel and con-sonant mispronunciations in familiar words. On thebasis of Nespor et al.’s claims and Nazzi’s findings,one might predict that infants will be more sensitive toconsonant mispronunciations than vowel mispronuncia-tions of familiar words.

Experiment 1

The first series of experiments in this study examineswhether infants at 15, 18 and 24 months are sensitive tomispronunciations involving changes in the word-medialvowels of familiar monosyllabic words. Using the inter-modal preferential looking task (for details see Bailey &Plunkett, 2002; Ballem & Plunkett, 2005), infants arepresented with correct and incorrect labels for familiarobjects. If infants have phonologically detailed lexicalrepresentations including information about vowel iden-tity, then infants should look longer at a target objectwhen the label is correctly pronounced than when it isincorrectly pronounced. Such a finding would providesupport for the hypothesis that vowels play a prominentrole during word recognition in the early developing lex-icon. If infants fail to show sensitivity to vowel mispro-nunciations of familiar words, the hypothesis thatvowels are not central to constraining lexical identityin the early lexicon would be supported.

Method

Participants

The participants in this experiment were 28 infants at15 months (M = 15.04 months; range = 14.1–15.4 months;13 M and 15 F), 30 infants at 18 months (M = 18.1 months;range = 17.3–18.6 months; 17 M and 13 F) and 31 infantsat 24 months (M = 24.28 months; range = 23.7–25.1months; 16 M and 15 F). Thirteen additional infants weretested but were excluded due to fussiness, parental interfer-ence, or experimenter error (8 at 15 months; 2 at 18 months;3 at 24 months). All infants had no known hearing orvisual problems and were recruited via the maternityward at the John Radcliffe Hospital in Oxford. Infantscame from homes where British English was the only lan-guage in use. All parents were asked to complete theOxford Communicative Developmental Inventory(OCDI; Hamilton, Plunkett, & Schafer, 2000), a Britishadaptation of the MacArthur CDI (Fenson et al., 1993).

Stimuli

The stimuli presented to infants at 18 and 24 monthswere 16 monosyllabic (CVC) nouns taken from theOCDI. Each infant heard eight labels, half of which


were correctly pronounced while the other half wereincorrectly pronounced. Following Ballem and Plunkett(2005), mispronunciations were created by changing oneor more of the dimensions of the vowel (i.e., height,backness, roundedness), usually resulting in a non-word.There were two exceptions (dog fi dig and doll fi dill)involving mispronunciations that resulted in real wordsthat were judged unknown to the infants according toOCDI reports. In four words, mispronunciationsinvolved changes to one of the dimensions of the vowelalone—backness (1) or height (3). In all other cases, mis-pronunciations involved large changes to at least twodimensions of the vowel—backness, roundedness, orheight (see Table 1 for details). Table 1 gives a completelisting of the words and their correspondingmispronunciations.

The speech stimuli were produced by a femalespeaker of British English in an enthusiastic, child-direc-ted manner. The audio recordings were made with a dig-ital audio tape recorder (DAT) in a sound-attenuated,recording booth. The audio stimuli were then digitisedat a sampling rate of 44.1 kHz and a resolution of16 bits. Each of the auditory stimuli was then splicedinto the carrier phrase ‘‘Look! target word!’’ using Gold-wave 5.10. The duration, fundamental frequency, andintensity of the correctly pronounced and mispro-nounced labels are given in Table 1. There was no sys-tematic difference in the duration (t(15) = �1.489;p = .15), fundamental frequency (t(15) = �.008,p = .99) and intensity (t(15) = .831, p = .419) of thecorrect and mispronounced labels.

Table 1Durations of the correctly pronounced and mispronounced labels pre

Correctpronunciation

Dur(ms)

f0

(Hz)Amp(dB)

Incorrectpronunciation

Dur(ms)

Ball 577 307 83 Bal /bæl/ 543Bed 537 338 82 Bod /b cd/ 528Bib 503 356 80 Bab /bæb/ 565Book 528 351 80 Bik /bIk/ 557Boot 616 342 77 Bot /b ct/ 680Bread 593 341 80 Brod /br cd/ 600Brush 703 315 82 Brish /brIS/ 715Bus 652 334 81 Bis /bIs/ 613Dish 655 358 79 Dush /dfS/ 644Dog 524 338 83 Dig /dIg/ 517Doll 520 299 83 Dill /dIl/ 599Duck 560 393 80 Dack /dæk/ 559Milk 667 343 79 Marlk /m Alk/ 724Moon 721 309 81 Marn /m An/ 675Sock 704 376 81 Souk /sfk/ 728Sun 659 288 81 Sen /sen/ 704

Mean 607 336 80 Mean 621

Infants at 15 months were tested on a slightly differentset of 10 monosyllabic (CVC) words. The change in test-ing materials was necessitated by the smaller vocabulariesof the younger infants. The changes involved reducing thenumber of test stimuli and adding two new words (seeTable 2). Again, the audio stimuli were spoken by afemale speaker of British English in an enthusiastic voice.The durations, fundamental frequency, and intensity ofthe correctly pronounced and mispronounced labels pre-sented to infants at 15 months are given in Table 2. Two ofthe mispronunciations involved changes in only one of thevowel dimensions—backness (1) or height (1). All theother mispronunciations involved large changes of twovocalic dimensions—backness, roundedness, or height(see Table 2 for details). There was no systematic differ-ence in the duration (t(9) = .944, p = .37), fundamentalfrequency (t(9) =�.516, p = .618) or intensity (t(9) = .874,p = .405) of the correct and mispronounced labels.

Visual stimuli were computer images created fromphotographs, with one image for each word. Imageswere judged by three adults (the authors and an indepen-dent observer) as typical exemplars of the labelledcategory.

Procedure

During the experiment, all infants sat on their care-giver’s lap approximately 80 cm away from a projectionscreen (1.3 m · .35 m). Two cameras mounted directlyabove the visual stimuli recorded infants’ eye-move-ments. Synchronised signals from the two cameras werethen routed via a digital splitter to create a recording of

sented to infants at 18 and 24 months (Experiment 1)

f0

(Hz)Amp(dB)

Main dimension changes Distracter

Back Round Height

334 83 � � Bed339 81 + + Ball351 82 � Boot356 78 � � Bus332 81 � Bib348 83 + + Brush338 78 � + Bread323 78 � + Book316 81 + + Doll357 78 � � + Duck312 81 � � + Dish375 82 � Dog336 80 + � Moon297 81 � � Milk314 78 + Sun361 78 � + Sock

336 80

Table 2Durations of the correctly pronounced and mispronounced labels presented to infants at 15 months (Experiment 1)


Dur(ms)

f0

(Hz)Amp(dB)

Incorrectpronunciation

Dur(ms)

f0

(Hz)Amp(dB)

Main dimension changes Distracter

Back Round Height

Ball 627 254 79 Bal /bæl/ 597 201 77 � � BedBed 573 302 80 Bod /b cd/ 568 266 77 + + BallBib 569 260 80 Bab /bæb/ 563 264 80 � BookBook 617 306 80 Bik /bIk/ 510 367 77 � � BibBread 670 275 79 Brod /br cd/ 685 258 82 + + BrushBrush 729 273 77 Brish /brIS/ 696 269 79 � + BreadCup 667 400 78 Kip /kIp/ 778 381 77 � + KeysDog 578 308 77 Dig /dIg/ 542 328 78 � � + DuckDuck 647 244 79 Dack /dæk/ 602 316 76 � DogKeys 729 283 78 Koos /ku:z/ 700 322 78 + + Cup

Mean 640 290 78 Mean 624 297 78


two separate time-locked images of the infant. Caregiv-ers wore headphones throughout the experiment andwere instructed to keep their eyes shut.

Infants at 18 and 24 months were each presented witheight trials. In each trial, infants saw images of twofamiliar objects, side-by-side, for 5 s. The distancebetween images was 15 cm. One of the objects was thennamed in the carrier phrase ‘‘Look! target word!’’ witheither a correct label or a mispronunciation. The audi-tory signal was delivered through a centrally locatedloudspeaker situated immediately above the projectionscreen. Onset of the target word began halfway intothe trial at 2500 ms. This onset divided the trial into apre- and post-naming phase. Infants saw each objectonly once during the experiment paired with anotherdistracter object whose label began with the same onsetconsonant. The requirement that target and distracterhave labels with the same onset consonant ensures thatinfants cannot use the label onset to identify the targetreferent (Fernald, Swingley, & Pinto, 2001). Infantsmust therefore pay attention to the vowel nucleus forsuccessful target identification. The target–distracterpairings are listed in Table 1. Across infants, imagesappeared as target or distracter with equal frequency.Likewise, words were equally likely to be correctlypronounced and mispronounced. Half of the labelspresented to infants were correctly pronounced whilethe other half were incorrectly pronounced. Infantsnever heard the same object labelled with both an incor-rect and a correct pronunciation. Targets appearedequally often to the left and to the right. Likewise, cor-rect and incorrectly pronounced words identified leftand right targets equally often. Order of presentationof trials was randomised across infants.

The current study presents infants with just eighttrials, compared to the Swingley and Aslin (2000,2002) studies which presented infants with 24 test trials.The relatively small number of trials is motivated by

the requirement that both target and distracter havenames that begin with the same onset consonant (seeabove) and an attempt to avoid any priming effectsdue to repetition of stimuli, thus ensuring that infantswere not presented with the same picture twice in thesame experiment and that they did not hear both cor-rect and incorrect pronunciations of the same word.While the number of trials could have been increasedby adding bisyllabic words or vowel-initial words, thecurrent study attempted to localise the source of anymispronunciation effects: infants were only presentedwith word-medial vowel mispronunciations of closed,monosyllabic words. Infants’ sensitivity to mispronun-ciations in different word positions and mispronuncia-tions of bisyllabic words have yet to be systematicallyexplored.

Fifteen-month-old infants know fewer words andhence it is even more difficult to find pairs of objectswhose names begin with the same onset consonant. Fif-teen-months-olds were presented with 10 trials, eachtrial having the same overall form as that presented tothe older infants. The trials were divided into two testingblocks where infants saw each pair of objects once ineach block, i.e., infants were presented with five pairsof images twice. For each pair, one of the objects waslabelled in one of the blocks while the other object waslabelled in the other block. Hence, the second block oftesting permitted evaluation of vowel sensitivity to agreater range of words for every infant. If the label ofone of the objects of a pair was correctly pronouncedin the first block, the label for the other object of the pairwas mispronounced in the second block and vice versa.Therefore, an object in a pair was never the target objectin both blocks. This ensured that infants did not hearcorrect and incorrect pronunciations of the same wordin the experiment. Half of the labels were correctly pro-nounced and the other half were incorrectly pronouncedand each image appeared equally often as target and


distracter. Across infants, words were equally likely tobe correctly pronounced and mispronounced. Half ofthe infants heard 3 mispronunciations and 2 correct pro-nunciations in the first block and 2 mispronunciationsand 3 correct pronunciations in the second block. Forthe other half of the infants, this was reversed. Orderof presentation of the trials was randomised within ablock and image pairs were counterbalanced fortarget-side and mispronunciations.

Scoring

A digital-video scoring system was used to assessvisual events on a frame-by-frame basis (every 40 ms).This technique enabled tracking of every eye fixation.A second skilled coder evaluated the data from 10% ofthe participants. Coders achieved a high level of agree-ment (r = .95, p < .001).

The coded video frames were used to determine theamount of time infants’ looked at the target (T) anddistracter (D) images in the two phases of each trial;before and after the onset of the target word. Similarly,we also calculated the length of infants’ longest fixa-tions at the target (t) and distracter (d) for the twophases of each trial. As in previous research, it wasassumed that the amount of time required by infantsto initiate an eye-movement was 367 ms (Swingley &Aslin, 2000, 2002; Swingley, Pinto, & Fernald,1999).Consequently, analysis of the post-naming phase ofthe trial was initiated 367 ms after the onset of the tar-get word. This ensures that the analyses only considerchanges in infants’ looking behaviour that can reason-ably constitute a response to the spoken word. In addi-tion, only those trials in which infants fixated both thetarget and the distracter during the pre-naming phaseof the trial were included. For the 15-month-oldinfants, we relaxed this criterion slightly to require onlythat they fixated both pictures sometime during theentire course of the trial. This was necessary becausethe 15-month-olds made fewer saccades during thecourse of stimulus presentation than the older infants.On the basis of these criteria, we excluded 31 trials(14 trials at 15 months; 14 trials at 18 months; 3 trialsat 24 months) from the analysis.

We calculated the difference (t � d) between infants’longest look (LLK) at target (t) and distracter (d) imagesbefore and after target word onset (Bailey & Plunkett,2002; Ballem & Plunkett, 2005; Meints, Plunkett, &Harris, 1999; Schafer & Plunkett, 1998). A differencemeasure is used to calculate the target preference duringeach phase of the trial because the longest looks involveonly single fixations on target and distracter. Systematicincrements in infants’ longest look at the target acrossthe two phases of the trial can be interpreted as a mea-sure of the child’s understanding of the target word (Bai-ley & Plunkett, 2002; Meints et al., 1999; Reznick, 1990;Schafer & Plunkett, 1998; Swingley & Aslin, 2000).

We also calculated the amount of time infants spentlooking at the target (T) over the amount of timeinfants’ spent looking at the target and distracter(T + D) in order to determine the proportion of timeinfants spent looking at the target—proportion of targetlooking measure (PTL). As with the LLK measure, asignificant increase in infants’ preference for the targetacross the two phases of the trial indicates infants’ asso-ciation of the target label and target object.

Results of the main ANOVAs using the LLK andPTL measure will be reported in tables. However,planned comparisons will mainly be reported using theLLK measure. Planned comparisons using the PTLmeasure will only be reported if there are differences inthe results between the two measures. Previous studieshave found that the LLK measure provides a more sen-sitive index of infants’ comprehension (Meints et al.,1999; Schafer & Plunkett, 1998; Southgate & Meints,1999). Schafer and Plunkett argue that the PTL measureis more susceptible to decreasing infant participationduring the course of the trial. As the trial proceeds,infants may become distracted and start to display look-ing patterns not necessarily related to the auditory stim-ulus—one might reasonably expect gaze switches fartherfrom the time of presentation of auditory stimulus to beless related to the occurrence of the auditory stimulus.The PTL measure may, therefore, provide a more gen-eral measure of the level of infant interest in the targetobject aroused upon presentation of an auditory stimu-lus. The LLK measure, on the other hand, may providea more direct measure of the time it takes the infant tocheck that the label matches the object.

Some studies also report the amount of time taken byinfants to switch from the distracter image to the targetimage upon hearing the target label as an index ofinfants’ preference for the target image (Fernald et al.,2001; Swingley & Aslin, 2000, 2002; Swingley et al.,1999), arguing that longer latencies reflect the extra timerequired to identify a semantic mismatch (Swingley &Aslin, 2000, pp. 151–2). Since it is usually assumed inthese studies that the minimum amount of time requiredby infants to initiate an eye-movement is around theorder of 367 ms, only eye-movements 367 ms after theonset of the target word were considered. A rapidchange in gaze after this point is taken as a measure ofinfants’ detection of a mismatch between the picturecurrently fixated and the target label.

Results

15 months

Fig. 1 plots the difference between the 15-month-olds’ preference for the target in the pre- and thepost-naming phases of Block 1 and Block 2 using theLLK measure. Fig. 1 suggests that there was no system-atic difference in infants’ preference for the target

Fig. 1. Experiment 1: mean difference between the pre- andpost-naming phase of the correctly pronounced and mispro-nounced trials in Blocks 1 and 2 presented to 15-month-olds(Longest look data).


between the pre- and the post-naming phase when thetarget label was mispronounced. Conversely, there wasan increase in preference for the target from the pre-and the post-naming phase when the target label wascorrectly pronounced in Block 2.

The results of a 3 · 2 repeated-measures ANOVAwith the factors naming (pre- and post-naming), accu-racy of pronunciation (correct and incorrect) and blockare reported in Table 3 by subjects and by items for bothmeasures. The associated min F 0 values are also reportedin Table 3. A significant main effect of block (by sub-jects) and a significant interaction between naming andpronunciation (by items) indicated that there was a sys-tematic difference in infants’ looking behaviour in thetwo blocks and that there was a significant differencein infants’ looking behaviour following correct andincorrect pronunciations. The data from the two blockswere analysed separately.

Block 1. The results of a 2 · 2 repeated-measuresANOVA using only the factors naming and accuracyof pronunciation on the data from the first block arealso reported in Table 3. These results indicate thatinfants failed to respond systematically to correct orincorrect pronunciations of the target label during thefirst block of testing.

Block 2. The results of a 2 · 2 repeated-measuresANOVA using the factors naming and pronunciationon the data from Block 2 are also reported in Table 3.The significant interaction between naming and pronun-ciation found in the ANOVA indicates systematic differ-ences in infants’ looking behaviour between the pre- andpost-naming phases in the correctly pronounced and

mispronounced conditions. The means and confidenceintervals of the increment in target looking for the differ-ent pronunciation conditions are reported below.

Planned comparisons using the LLK measurerevealed that there was a significant difference in infants’preference for the target from the pre- to the post-nam-ing phase following correct pronunciations (M = 383 ms(CI: 18,748)). The difference was near-significant usingthe PTL measure. Conversely, there was no significantdifference in infants’ preference for the target from thepre- to the post-naming phase following incorrect pro-nunciations using either measure (LLK: M = �95 ms(CI: �229,421)).

Latency analysis of Block 2. Analysis of the responselatencies in Block 2 yielded similar results. Infants werefixating the distracter picture at the disambiguationpoint (367 ms after onset of target word) in 42.4% ofthe trials. Of these trials, infants switched from the dis-tracter to the target image 84.2% of the time, therebyproviding response latencies on 35.5% of all trials(.42 · .84). Swingley and Aslin (2002) argue that youn-ger infants are less likely to provide reliable latency mea-sures. Consequently, latency measures for the younginfants in the Swingley and Aslin study were analysedusing unaggregated data. Using the same method ofanalysis, the 15-month-olds in the current study tookan average of 617 ms to switch from the distracter tothe target image upon hearing correct pronunciationsof the target label. Conversely, infants took an averageof 966 ms to switch from the distracter to the targetimage upon hearing incorrect pronunciations of the tar-get label. Infants exhibited delayed response times tomispronunciations compared to correct pronunciationsof the target label, suggesting that infants were sensitiveto mispronunciations of the target label. A one-wayANOVA indicated that the difference between condi-tions was significant ((F(1,47) = 5.809; p = .02); meandifference = 349 ms (CI: 57,641)).

These results indicate that infants at 15 months aresensitive to mispronunciations of familiar words whenthe mispronunciations involve a change to the vocalicnucleus. However, this sensitivity was only observedduring the second block of testing.

18 and 24 months

Since the 18- and 24-month-olds were presented withthe same experiment, the data from both age groups wasanalysed together. Fig. 2 plots the difference betweeninfants’ preference for the target in the pre- and thepost-naming phase using the LLK measure. Fig. 2 sug-gests that there was no systematic difference in infants’preference for the target between the pre- and thepost-naming phase when the target label was mispro-nounced. Conversely, there was an increase in preferencefor the target between the pre- and the post-naming

Tab

le3

An

alys

iso

fva

rian

cere

sult

sfo

rE

xper

imen

t1

(15-

mo

nth

-old

s)(s

ign

ifica

nt

effec

tsar

eh

igh

ligh

ted

inb

old

)

Eff

ect

Blo

ckF

1(s

ub

ject

anal

ysis

)F

2(i

tem

anal

ysis

)M

inF0

PT

LL

LK

PT

LL

LK

PT

LL

LK

Fd

fp

Fd

fp

Fd

fp

Fd

fp

Nam

ing

1an

d2

.092

1,27

.763

.005

1,27

.944

.990

1,9

.346

.208

1,9

.660

F(1

,31)

=.0

84F

(1,2

8)=

.004

Pro

nu

nci

atio

n1

and

2.9

211,

27.3

46.4

861,

27.4

925

.17

71

,9.0

49

2.07

51,

9.1

84F

(1,3

4)=

.781

F(1

,35)

=.3

93B

lock

1an

d2

5.3

90

1,2

7.0

28

3.69

01,

27.0

654.

342

1,9

.067

7.5

18

1,9

.02

3F

(1,2

4)=

2.43

2F

(1,3

5)=

2.47

5N

*P

1an

d2

2.32

01,

27.1

392.

155

1,27

.154

9.8

37

1,9

.01

26

.27

11

,9.0

34

F(1

,35)

=1.

877

F(1

,36)

=1.

603

N*

B1

and

2.2

031,

27.6

561.

563

1,27

.222

.994

1,9

.345

6.0

58

1,9

.03

6F

(1,3

5)=

.168

F(1

,36)

=1.

242

P*

B1

and

21.

732

1,27

.199

5.0

24

1,2

7.0

33

2.85

01,

9.1

264.

503

1,9

.063

F(1

,35)

=1.

077

F(1

,26)

=2.

374

N*

P*

B1

and

22.

033

1,27

.165

3.34

71,

27.0

781.

900

1,9

.201

2.08

31,

9.1

83F

(1,2

6)=

.982

F(1

,21)

=1.

283

Nam

ing

1.0

271,

28.8

71.6

261,

28.4

35.0

171,

9.8

991.

004

1,9

.343

F(1

,21)

=.0

10F

(1,3

3)=

.385

Pro

nu

nci

atio

n1

.024

1,28

.877

.348

1,28

.560

.213

1,9

.656

.494

1,9

.500

F(1

,32)

=.0

21F

(1,3

2)=

.204

N*

P1

.002

1,28

.966

.022

1,28

.883

.013

1,9

.911

.272

1,9

.615

F(1

,35)

=.0

01F

(1,3

2)=

.020

Nam

ing

2.3

361,

28.5

67.9

111,

28.3

483.

784

1,9

.084

5.1

40

1,9

.05

0F

(1,3

2)=

.308

F(1

,35)

=.7

73P

ron

un

ciat

ion

22.

721

1,28

.111

3.31

81,

28.0

803.

083

1,9

.113

5.3

24

1,9

.04

6F

(1,2

9)=

1.44

5F

(1,3

3)=

2.04

4N

*P

24

.22

91

,28

.05

06

.90

51

,28

.01

43.

397

1,9

.098

4.46

81,

9.0

64F

(1,2

4)=

1.88

3F

(1,2

2)=

2.71

2Fig. 2. Experiment 1: mean difference between the pre- andpost-naming phase of the correctly pronounced and mispro-nounced trials presented to 18- and 24-month-olds (longestlook data).


phase when the target label was correctly pronounced.The data were analysed using a mixed model ANOVAwith the factors naming (pre- and post-naming) and pro-nunciation accuracy (correct and mispronounced) aswithin-subjects factors and age (18 and 24 months) asa between-subjects factor (see Table 4 for results of anal-ysis by subjects and by items, together with associatedmin F 0 values—significant F1, F2 and min F 0 values arein bold).

The significant interaction between naming and pro-nunciation indicates systematic differences in infants’looking behaviour between the pre- and post-namingphase in the correctly pronounced and mispronouncedconditions. Importantly, there was no main effect ofage or any significant interactions with age. The meansand confidence intervals of the increment in target look-ing for the different pronunciation conditions arereported below, separated by age using the LLKmeasure.

18 months. There was a significant difference in infants’preference for the target from the pre- to the post-nam-ing phase following correct pronunciations (M = 256 ms(CI: 26,485)). Conversely, there was no significant differ-ence in infants’ preference for the target from the pre- tothe post-naming phase following incorrect pronuncia-tions (M = �109 ms (CI: �336,117)).

24 months. The results with the 24-month-olds weresimilar to the 18-month-old infants. There was a signif-icant difference in infants’ preference for the target fromthe pre- to the post-naming phase following correct pro-nunciations (M = 318 ms (CI: 73,563)). Conversely,there was no significant difference in infants’ preference

Tab

le4

An

alys

iso

fva

rian

cere

sult

sfo

rE

xper

imen

t1

(18-

and

24-m

on

th-o

lds)

(sig

nifi

can

teff

ects

are

hig

hli

ghte

din

bo

ld)

Eff

ect

F1

(su

bje

ctan

alys

is)

F2

(ite

man

alys

is)

Min

F0

PT

LL

LK

PT

LL

LK

PT

LL

LK

Fd

fp

Fd

fp

Fd

fp

Fd

fp

Nam

ing

2.66

31,

59.1

082.

276

1,59

.137

2.22

61,

30.1

461.

661

1,30

.146

F(1

,75)

=1.

212

F(1

,71)

=.9

60P

ron

un

ciat

ion

.478

1,59

.492

2.15

41,

59.1

48.7

421,

30.3

962.

509

1,30

.396

F(1

,88)

=.2

90F

(1,8

3)=

1.15

8A

ge.0

071,

59.9

32.0

081,

59.9

29.0

011,

30.9

33.0

001,

30.9

33F

(1,3

9)=

.000

F(1

,59)

=.0

07N

*A

.071

1,59

.790

.047

1,59

.830

.206

1,30

.653

.106

1,30

.653

F(1

,36)

=.0

52F

(1,8

9)=

.032

P*

A.3

611,

59.5

50.2

061,

59.6

52.7

131,

30.4

05.2

481,

30.4

05F

(1,8

9)=

.239

F(1

,84)

=.1

12N

*P

11

.205

1,5

9.0

01

10

.578

1,5

9.0

02

8.2

85

1,3

0.0

07

9.7

66

1,3

0.0

07

F(1

,71

)=

4.7

63

F(1

,77

)=

5.0

77

N*

P*

A.0

341,

59.8

54.0

911,

59.7

641.

234

1,30

.276

.084

1,30

.276

F(1

,62)

=.0

33F

(1,7

7)=

.043


for the target from the pre- to the post-naming phasefollowing incorrect pronunciations (M = �122 ms (CI:�388,143)).

For both measures, and at both age-groups, correctpronunciations resulted in systematic increments inlooking from the pre- to the post-naming phase of thetrial, whereas mispronunciations did not. These resultsindicate that infants at 18 and 24 months of age weresensitive to mispronunciations of the vowels in familiarwords.

Latency analysis

Analysis of the response latencies yielded similarresults to the LLK and PTL measures. Infants fixatedthe distracter picture at the disambiguation point(367 ms after onset of the target word) in 50.5% of tri-als. Of these trials, infants switched from the distracterto the target image 80.3% of the time. Consequently,response latencies were measured on 40.6% of all trials(.50 · .80). Twenty infants did not provide latency mea-sures in both pronunciations conditions and wereexcluded from the analysis. The data from the remain-ing infants showed that infants took an average of633 ms to switch from the distracter to the target imageupon hearing correct pronunciations of the target label.Conversely, infants took an average of 792 ms to switchfrom the distracter to the target image upon hearingincorrect pronunciations of the target label. The datawere analysed using a mixed model ANOVA with pro-nunciation condition (correct and mispronounced) as awithin-subjects factor and age as a between subjects fac-tor (18 and 24 months). There was a significant effect ofpronunciation (F(1,40) = 4.779, p = .035; meandifference = 158 (CI: 14,302)). The interaction betweenage and pronunciation was not significant (F(1,40) = .03,p = .86).

These results indicate that infants switch faster fromthe distracter to the target when the target label is cor-rectly pronounced compared to when it is mispro-nounced and, therefore, that infants are sensitive tomispronunciations involving changes to the vowels offamiliar words at 18 and 24 months. In addition, therewere no systematic differences in the performance ofthe 18- and 24-month-olds with both age groups show-ing similar sensitivity to mispronunciations.

Effects of vocabulary size

We calculated the mean receptive percentile vocabu-lary size of the different age-groups based on parentalOCDI reports (15 months: M = 22.2%; SD = 13.5%;range: 3–61%; 18 and 24 months: M = 58.7%;SD = 24.8%; range: 14–99%). Earlier studies havefound no evidence of a relationship between vocabularysize and infants’ sensitivity to mispronunciations (Bai-ley & Plunkett, 2002; Ballem & Plunkett, 2005; Swing-ley & Aslin, 2000, 2002). We measured the correlation

1 Note that we cannot directly compare the performance ofinfants in the Ballem and Plunkett design and the current studydue to the differences between the two experiments, such as thepresentation of training blocks prior to testing blocks, and thepresentation of novel objects during training and testing whichmay have sensitised infants to the mispronunciations.


between infants’ vocabulary size and their sensitivity tomispronunciations, as measured by the difference innaming effect between correct and mispronouncedlabels. There was no evidence for any correlation at15 months (r = �.059; p > .5) or with the older infants(r = �1.7, p > .2).

We also calculated the mean productive percentilevocabulary size of the different age-groups based onparental OCDI reports (15 months: M = 2.4% words;SD = 2.1%; range: 0–8%; 18 and 24 months:M = 34.7%; SD = 27.8%; range: 0–94%). We measuredthe correlation between infants’ productive vocabularysize and their sensitivity to mispronunciations, as above.There was no evidence for any correlation at 15 months(r = �.282; p > .1), or with the older infants (r = �.11;p > .3). Hence, we find no evidence for a relationshipbetween the phonological specificity of infants’ lexicalrepresentations and the size of infants’ productive orreceptive vocabulary (as measured by the OCDI),thereby confirming the findings of earlier studies.

Note that we did not check to see whether infants’ability to produce the words presented to them in theexperiment influenced their sensitivity to the mispronun-ciations. This was for two reasons. First, CDI data can-not tell us whether infants were accurately producing theword, if they were able to say them at all. Second, wecannot assume that there is a deterministic influencebetween infants’ ability to say a word and infants’ sensi-tivity to a mispronunciation of that word. Infants maybe able to produce a word accurately but fail to showa mispronunciation sensitivity (or vice versa). Our datado not speak to the otherwise interesting issue of therelationship between perception and productionmechanisms.

Discussion

The primary goal of this study was to determinewhether infants would recognise that a familiar wordwas mispronounced when the locus of the mispronunci-ation was the vowel nucleus. Although a considerablebody of research has accumulated regarding infant sen-sitivity to mispronunciations of consonants in familiarand novel words, there are no studies that have system-atically explored infant sensitivity to vowel changes forfamiliar word recognition in the second year of life.Our results provide clear evidence that infants are sensi-tive to mispronunciations of the vowels in familiarmonosyllabic words when the word is used to identifya referent. Infants looked significantly longer at a targetobject in the post-naming phase than in the pre-namingphase of a trial when the labels were correctly pro-nounced. Conversely, they did not look significantlylonger at a target object in the post-naming phase thanin the pre-naming phase when the labels were mispro-nounced. Likewise, infants were faster to switch from

the distracter to the target image on hearing correct pro-nunciations compared to mispronunciations. All the agegroups tested in the study showed this sensitivity tovowel mispronunciations, indicating that changes tothe vocalic nucleus of familiar labels systematicallyaffected identification of target objects, perhaps asyoung as 15-months-old, and certainly as young as18-months-old. This suggests that infant lexical repre-sentations contain sufficient phonological informationregarding the vowels in familiar words to recognise amispronunciation of the vowel when identifying a targetreferent: early lexical representations are not phonolog-ically underspecified, at least for the vowels of theclosed, monosyllabic, familiar words tested in this study.

The performance of the 15-month-olds might beinterpreted as indicating that the phonological represen-tations for vowels are less robustly specified than for theolder infants: while infants displayed sensitivity to mis-pronunciations of familiar words in the second blockof testing, there was no evidence of infants recognisingeven the correctly pronounced words in the first block.This finding contrasts with Ballem and Plunkett (2005)who found that 14-month-old infants displayed sensitiv-ity to consonant mispronunciations of familiar wordseven in the first block of testing. This provides aninteresting contrast to the results obtained with the15-month-olds in the current experiment where sensitiv-ity to mispronunciations was found in only the secondblock of testing, after infants had already been exposedto the object pairs in Block 1.

This might suggest that infants at 15 months only dis-play sensitivity to mispronunciations of vowels uponrepetition of object pairs. Does this differ from the pat-tern of infant sensitivity to consonant mispronunciationsfound in earlier studies? Note that although infants inthe Ballem and Plunkett study showed sensitivity to con-sonant mispronunciations in Block 1, they had been pre-sented with the same object pairs four times in the firstblock—twice with correct pronunciations and twice withmispronunciations. Consequently, the consonant mis-pronunciation effect displayed in Block 1 in Ballemand Plunkett (2005) might have been facilitated by therepeated presentation of object pairs.1

A similar pattern is found upon closer analysis of theSwingley and Aslin (2000, 2002) design. Infants in theseexperiments were also presented with each object pairfour times. In the absence of any other data on infants’sensitivity to mispronunciations of consonants followingthe first presentation of the picture pairs in these studies,


it is difficult to reach a firm conclusion regarding thedependence of a consonant mispronunciation effect onrepetition of picture pairs.

Overall, the results of the current study support theexpectations raised by the Swingley and Aslin studies(2000, 2002) that infants will also show sensitivity tovowel mispronunciations in familiar words. The currentstudy goes further in that it provides a systematic explo-ration of infants’ sensitivity to vowel mispronunciationsacross a much wider range of words, in the absence ofconfounding factors such as syllabicity of the wordbeing tested, and change in position of the vowel mis-pronunciation (Swingley & Aslin, 2000, 2002). However,due to the differences between earlier experimentsreporting infants’ sensitivity to consonant mispronunci-ations and the current study, any difference in infants’performance between these experiments cannot reliablybe attributed to a putative asymmetry between vowelsand consonants in infants’ lexical representations. Clo-ser consideration of the design of earlier experimentsreporting a consonant mispronunciation effect suggeststhat the latter may be influenced by the repetition ofthe picture pairs. In order to provide a more direct com-parison of the role of consonants and vowels during lex-ical processing by infants, Experiment 2 tests infants’sensitivity to consonant and vowel mispronunciationsof the same words in the same experiment. In addition,Experiment 1 tested infants’ sensitivity to mispronuncia-tions of different degrees (along one-dimension andalong many-dimensions). Analysis of infants’ sensitivityto the different degrees of mispronunciations suggestedthat none of the age-groups tested were sensitive toone-dimension changes to the vowels (see Appendix).However, the absence of a significant result may havebeen caused by the small number of infants who contrib-uted to the analysis. Consequently, in Experiment 2, wecompare infants’ sensitivity to single-dimension changesto the vowels and consonants in order to provide a moreappropriate test of infants’ sensitivity to small mispro-nunciations of familiar words.

Experiment 2

Although Experiment 1 provides confirmatory sup-port for the hypothesis that vowels play a prominentrole during word recognition, we cannot reject thehypothesis that vowels are less central in constraininglexical identity than consonants. Nazzi (2005) presentsthe only systematic comparison of the differences ineffects of consonants and vowels on early word-learning.He found that 20-month-old French infants were unableto learn two words which differed only in their vowels,while being able to learn two words differing in eitherword-initial or word-medial consonants. He argued thatthis provides evidence that consonants are more crucial

to lexical acquisition than vowels. The striking differencebetween the ability of the English infants in Experiment1 to recognise vowel changes in familiar words and thefailure of French infants to learn two words whichformed a minimal pair on the vowel in the Nazzi(2005) study raises questions about the relative impor-tance of vowels and consonants in lexical access inEnglish infants.

Experiment 2 tests infants’ sensitivity to mispronun-ciations of the vowels and consonants in familiar wordsin order to provide a systematic comparison of thephonological specificity of vowels and consonants inearly lexical representations. In this task, infants werepresented with correct and incorrect labels for familiarobjects. Unlike Experiment 1, half of the mispronunci-ations involved word-medial, vowel changes of thefamiliar CVC labels whereas the other half involvedword-initial, consonant changes. If Nazzi’s conclusionsregarding the relative importance of vowels and conso-nants in guiding lexical acquisition also relate to lexicalaccess, infants should show greater sensitivity to conso-nant mispronunciations than to vowel mispronuncia-tions. On the other hand, if vowels and consonantssimilarly constrain lexical access, there should be nodifference in infants’ sensitivity to vowel and consonantmispronunciations. Furthermore, Experiment 2 exam-ines infant sensitivity to a change to only one-dimen-sion of the vowels and consonants of the familiarwords, thereby providing a more stringent test of thephonological specificity of the underlying lexicalrepresentations.

Method

Participants

The participants in this experiment were 29 infants at15 months (M = 14.72 months; range = 13.6–15.6months, 11 M and 18 F), 27 infants at 18 months(M = 18.1 months; range = 17.7–18.8 months, 12 Mand 15 F) and 28 infants at 24 months (M = 23.84;range = 23.0–24.7, 15 M and 13 F). Twenty additionalinfants were tested but were excluded due to experi-menter error (2), were outliers from the normal popula-tion (5), only looked at one picture in each trial (3) orbecause they did not complete the experiment (10). Allinfants were recruited according to the same criteria asExperiment 1. Again, all parents were asked to completethe OCDI.

Stimuli

The stimuli presented to infants were eight monosyl-labic (CVC) nouns taken from the OCDI. Each infantheard eight labels, half of which were correctlypronounced while the other half were incorrectly pro-nounced. Each infant heard two vowel mispronuncia-tions, and two consonant mispronunciations. Although


it is difficult to ensure equality in the perceived similarityor phonological equivalence of consonant and vowelschanges, we attempted to ensure that the changes causedwere roughly equivalent by manipulating only one-dimension of either consonant or vowel characteristics.Indeed, previous research has found that the degree ofcorrelation between adult ratings of perceived similarityof phonemes and phonemic feature similarity measuressupports the use of the latter in psycholinguistic testing(Bailey & Hahn, 2005, p. 356).

Note also that the stimuli tested make it easier forinfants to display sensitivity to consonant mispronunci-ation than to vowel mispronunciations, since all of theconsonant mispronunciations were word-initial, whileall vowel mispronunciations were word-medial. Previousresearch has demonstrated that the word-initial positionis extremely salient to infants, with the word onset beingadequate to guide infants’ recognition of the targetobject (Fernald et al., 2001). If infants display sensitivityto vowel mispronunciations despite the position of thesemispronunciations being less salient compared to theposition of the consonant mispronunciations, then thiswould provide evidence for clear sensitivity to vowelmispronunciations at an early age. Mispronunciationswere created by changing one of the dimensions of thevowel (height or backness) or of the consonant (placeor voice) and usually resulted in a non-word. There weresix exceptions (cat fi cart; bed fi bud; bus fi bass;

bus fi pus; ball fi gall; dog fi bog) involving mispro-nunciations that resulted in real words that were judgedunknown to the infants according to OCDI reports.Across infants, words were equally likely to receive con-sonant mispronunciations and vowel mispronuncia-tions. Table 5 gives a complete listing of the wordsand their corresponding mispronunciations anddistracter pairs.

The speech stimuli were produced by a femalespeaker of British English in an enthusiastic, child-direc-ted manner. The duration, fundamental frequency, andintensity of the correctly pronounced and mispro-

Table 5Durations of the correctly pronounced and mispronounced labels pre


Dur(ms)

f0

(Hz)Amp(dB)

Vowelmispronunciation

Dur(ms)

f0

(Hz

Ball 621 255 80 Bule /bu:l/ 706 236Bib 440 307 77 Bab /bæb/ 517 278Bed 563 324 82 Bud /b�d/ 484 280Bus 596 364 84 Bas /bæs/ 635 307Cat 557 353 81 Cart /c A:t/ 618 290Cup 575 425 81 Cep /kep/ 453 408Dog 532 295 83 Doog /dfg/ 602 274Keys 789 268 76 Kas /kæz/ 644 336

Mean 584 323 80 Mean 582 301

nounced labels are given in Table 5. There was no sys-tematic difference in the overall duration(F(2,21) = .54, p = .58), intensity (F(2,21) = 1.569,p = .23) or fundamental frequency (F(2,21) = .37,p = .69) of the stimuli in the three pronunciationconditions.

Visual stimuli were computer images created fromphotographs, with one image for each word. As inExperiment 1, images were judged by three adults (theauthors and an independent observer) as typicalexemplars of the labelled category.

Procedure

The 18 and 24-month-olds were presented with eighttrials each. The timing of presentation of the auditoryand visual stimuli was identical to that of Experiment1. Unlike Experiment 1, distracter objects were neverlabelled due to experimental constraints on the stimuli:first, the labels for the target and distracter objectsalways began with the same consonant. Second, mispro-nunciations were always one-dimension mispronuncia-tions of the target label. Together, these constraintsmade it difficult to find words that permitted one-dimen-sion vowel mispronunciations that did not sound similarto words already known to infants.

In addition, the 15-month-old infants were providedwith two wake-up trials before the main experiment inorder to familiarise them with the task. Although Exper-iment 2 had no block design, we hoped to encourageinfant participation in the early trials of the experimentthrough the inclusion of these wake-up trials. The trialswere identical to trials in the main experiment, exceptthat there were no mispronunciation trials. The pairsof words presented to infants in the wake-up trials werefish-sock and shoe-bird. Targets appeared equally oftento the left and to the right in the wake-up trials andthe main experiment. Likewise, correct and incorrectlypronounced words identified left and right targetsequally often. Order of presentation of trials wasrandomised across infants.

sented to infants (Experiment 2)

)Amp(dB)

Consonantmispronunciation

Dur(ms)

f0

(Hz)Amp(dB)

Distracter

78 Gall 558 265 78 Bear84 Dib 573 294 78 Boot84 Ped 455 297 82 Book83 Pus 536 314 82 Bike84 Gat 564 346 78 Cow79 Gup 522 418 83 Car84 Bog 455 288 83 Duck82 Tees 687 263 76 Coat

82 Mean 543 310 80

Tab

le6

An

alys

iso

fva

rian

cere

sult

sfo

rE

xper

imen

t2

(all

age-

gro

up

s)(s

ign

ifica

nt

effec

tsar

eh

igh

ligh

ted

inb

old

)

Eff

ect

F1

(su

bje

ctan

alys

is)

F2

(ite

man

alys

is)

Min

F0

PT

LL

LK

PT

LL

LK

PT

LL

LK

Fd

fp

Fd

fp

Fd

fp

Fd

fp

15

-mon

th-o

lds

Nam

ing

.040

1,28

.843

.679

1,28

.417

.006

1,7

.942

.176

1,7

.688

F(1

,9)

=.0

05F

(1,1

1)=

.139

Pro

nu

nci

atio

n.2

642,

27.7

70.1

182,

27.8

89.1

772,

6.8

42.6

592,

6.5

51F

(2,1

5)=

.105

F(2

,33)

=.1

00N

*P

4.2

72

2,2

7.0

24

2.9

59

2,2

7.0

69

7.6

84

2,6

.02

22.

758

2,6

.141

F(

2,2

7)

=2

.74

5F

(2,1

9)=

1.42

7

18

-a

nd

24

-mon

th-o

lds

Nam

ing

.088

1,53

.768

.682

1,53

.413

.389

1,13

.54

3.63

81,

13.0

79F

(1,6

6)=

.07

F(1

,64)

=.5

74P

ron

un

ciat

ion

.863

2,52

.428

1.12

22,

52.3

333.

015

2,12

.087

5.5

99

2,1

2.0

19

F(2

,63)

=.6

70F

(92,

64)

=.9

34A

ge.0

121,

53.9

14.1

581,

53.6

93.0

041,

13.9

53.0

001,

13.9

94F

(1,2

2)=

.003

F(1

,65)

=.1

36N

*A

.016

1,53

.900

1.12

01,

53.2

95.0

161,

13.9

0.0

141,

13.9

08F

(1,4

2)=

.008

F(9

1,13

)=

.013

P*

A.2

252,

52.8

00.3

492,

52.7

07.3

412,

12.7

1.2

922,

12.7

52F

(2,5

0)=

.135

F(2

,43)

=.1

58N

*P

7.8

43

2,5

2.0

01

6.5

14

2,5

2.0

03

4.5

98

2,1

2.0

33

4.3

51

2,1

2.0

38

F(

2,2

8)

=2

.89

F(2

,30

)=

2.6

08

N*

P*

A.5

012,

52.6

09.6

972,

52.5

02.4

522,

12.6

47.3

362,

12.7

21F

(2,3

7)=

.237

F(2

,40)

=.3

54


Scoring

The same digital-video scoring system as Experiment1 was used to analyse infant eye fixations. The codedvideo frames were used to calculate the LLK and PTLmeasures of looking behaviour. Only trials where infantsfixated both pictures during the entire course of the trialwere included in the analysis.

As in Experiment 1, main ANOVAs are reported intables (see Table 6) using both the PTL and LLK mea-sure. Planned comparisons using the PTL measure willbe reported only if they differ from the pattern of themain dependent variable. Note that the use of latencymeasures to compare consonant and vowel mispronun-ciations is inappropriate in this experiment as the locusof the mispronunciations types occurs at different timepoints in the word (consonant onset vs. vowel nucleus).Hence, latency measures in this experiment will only beused to compare consonant mispronunciations withcorrect pronunciations, and vowel mispronunciationswith correct pronunciations. In contrast, the LLKand PTL measures provide a global measure of prefer-ence over the complete duration of each phase of thetrial.

Results

Fig. 3 plots the difference between infants’ preferencefor the target in the pre- and the post-naming phase forthe three pronunciation conditions (correct, vowel andconsonant mispronunciation) using the LLK measureat 15, 18 and 24 months of age. Fig. 3 suggests that therewas no difference in infants’ preference for the targetbetween the pre- and the post-naming phase when thetarget label was mispronounced, irrespective of whetherthe mispronunciation changed the vowel or the conso-nant of the target label. Conversely, there was a system-atic increase in preference for the target between the pre-and the post-naming phase when the target label wascorrectly pronounced. Since the youngest infants werepresented with a slightly different version of the experi-ment (i.e., including two wake-up trials prior to the mainexperiment), the data from these infants was initiallyanalysed separately.

15 months

The results of the ANOVA with the factors naming(pre- and post-naming) and pronunciation accuracy(correct pronunciations; vowel mispronunciations; andconsonant mispronunciations) are presented in Table 6(reported by items and by subjects, including associatedmin F 0 values—significant results are highlighted inbold, while near-significant results are in italics). Theinteraction between naming and pronunciation in themain ANOVA suggests systematic differences in infants’looking behaviour between the pre- and post-namingphases in the three pronunciation conditions. This

Fig. 3. Experiment 2: mean difference between the pre- andpost-naming phase of the correctly pronounced and mispro-nounced trials (vowel and consonant mispronunciations)presented to 15-, 18- and 24-month-olds (longest look data).


interaction was significant using the PTL measure andmarginally significant using the LLK measure.

Planned comparisons revealed that correctpronunciations resulted in systematic increments inlooking from the pre- to the post-naming phase of thetrial (M = 297 (CI: 34,560)). Conversely, neither conso-nant (M = �181 (CI: �542,178)) nor vowel mispronun-ciations (M = 123 (CI: �279,526)) resulted in significantincrements in looking from the pre- to the post-namingphase of the trial. We also compared the mean differencein the increment from the pre- to the post-naming phasebetween the correctly pronounced and the two incor-rectly pronounced conditions. There was a significantdifference in the increment from the pre- to the post-naming phase between correct pronunciations andconsonant mispronunciations (M = 479 (CI: 75,883)).However, the difference in the increment from the pre-to the post-naming phase between correct pronuncia-tions and vowel mispronunciations was not significant(M = 173, (�721,373)). Finally, the difference in theincrement from the pre- to the post-naming phasebetween vowel and consonant mispronunciations wasalso not significant (M = 305, (�211,822)).

Latency analysis of the data from the 15-month-oldswas not possible. There was an insufficient number of tri-als where infants shifted their gaze from the distracter tothe target image within 2000 ms. Swingley and Aslin(2000, p. 9) have argued that saccades initiated more than2000 ms after the onset of the target word cannot bereliably interpreted as driven by the auditory stimulus.

18 and 24-month-olds

The results of a mixed model ANOVA with the fac-tors naming (pre- and post-naming) and pronunciation

accuracy (correct pronunciations; vowel mispronuncia-tions; and consonant mispronunciations) as within-sub-jects factors and age as a between-subjects factor arepresented in Table 6 (reported by items and by subjects,including associated min F 0 values—significant min F 0

values are in bold, while near-significant min F 0 valuesare in italics). The significant interaction between nam-ing and pronunciation in the main ANOVA suggestssystematic differences in infants’ looking behaviourbetween the pre- and post-naming phase in the threepronunciation conditions. As in Experiment 1 and inearlier studies (Swingley & Aslin, 2000), the absence ofa main effect of age or a significant interaction betweenage and any other factors suggests that there were nosystematic differences between the 18- and 24-month-olds. The data from these two age-groups were, there-fore, analysed together.

As with the younger infants, correct pronunciationsresulted in systematic increments in looking from thepre- to the post-naming phases of the trial (M = 357(CI: 180,534)). Conversely, neither consonant (M = �90(CI: �364,184)) nor vowel mispronunciations(M = �55 (CI:�338,226)) resulted in similar incrementsin looking from the pre- to the post-naming phases ofthe trial. We also analysed the mean difference in theincrement from the pre- to the post-naming phasebetween the correctly pronounced and the two incor-rectly pronounced conditions. There was a significantdifference in the increment from the pre- to the post-naming phase between correct pronunciations andvowel mispronunciations (M = 412 (CI: 129,696)). Sim-ilarly, there was a significant difference in the incrementfrom the pre- to the post-naming phase between correctpronunciations and consonant mispronunciations(M = 447 (CI: 134,760)). However, the difference inthe increment from the pre- to the post-naming phasebetween vowel mispronunciations and consonant mis-pronunciations was not significant (M = 34 (CI:�323,392)).

Despite the absence of a main effect of age, or anyinteractions with age, in order to ensure that the effectsreported above were not driven purely by the olderinfants, we separately analysed only the LLK data fromthe 18-month-olds. There was a significant interactionbetween naming and pronunciation at 18 months(F(2,25) = 5.741, p = .009). Planned comparisonsrevealed that there was a significant difference in 18-month-olds’ looking behaviour between correct pronun-ciations and vowel mispronunciations (M = 526 ms (CI:118,935)). Similarly, there was a significant difference in18-month-olds’ looking behaviour between correctpronunciations and consonant mispronunciations(M = 617 ms (CI: 171,1064)). However, there was nosignificant difference in 18-month-olds’ looking behav-iour between consonant mispronunciations and vowelmispronunciations (M = �90 ms (�608, 427)).


Latency analysis

Analysis of the response latencies yielded similarresults to the LLK and PTL measures. Infants fixatedthe distracter picture at the disambiguation point(367 ms after onset of the target word) in 45.5% of over-all trials. Of these trials, infants switched from the dis-tracter to the target image 80.3% of the time, yieldingresponse latencies for 36.5% of all trials (.45 · .80).Seven infants did not provide latency measures for morethan one condition. Consequently, latency analyses wereconducted using the data from 48 infants altogether (29infants for comparison of consonant mispronunciationswith correct pronunciations, and 34 infants for compar-ison of vowel mispronunciations with correctpronunciations).

Consonant mispronunciations. The data from the 29infants used in the latency analyses for consonant mis-pronunciations showed that infants took an average of630 ms to switch from the distracter to the target imageupon hearing correct pronunciations of the target label.In contrast, infants took an average of 485 ms toswitch from the distracter to the target image uponhearing consonant mispronunciations of the targetlabel. A paired samples t-test comparing latencies ofresponses for correct and consonant mispronunciationsconfirmed that this effect was significant (M = 144 (CI:14,274)), i.e., infants shift their gaze away from the dis-tracter picture to the target picture faster if the onsetconsonant is mispronounced than when it is correctlypronounced. Note that this is precisely the result onewould expect if, as Swingley and Aslin (2000) note, aswitch latency reflects ‘the child’s detection of a mis-match between the retrieved semantic category andthe initially fixated picture’ (p. 151). In this experiment,the target and distracter begin with the same conso-nant. Hence, an onset consonant mispronunciation willtrigger mismatch detection faster than a correctpronunciation.

Vowel mispronunciations. The data from the 34 infantsused in latency analyses for vowel mispronunciationsshowed that infants took an average of 682 ms to switchfrom the distracter to the target image upon hearing cor-rect pronunciations of the target label. Conversely,infants took an average of 851 ms to switch from the dis-tracter to the target image upon hearing vowel mispro-nunciations of the target label. A paired samples t-testcomparing latencies of responses for correct and vowelmispronunciations confirmed that this effect was signifi-cant (M = �168 (CI: �322,�14)).

Age-wise comparisons

The analysis using the data from the 15-month-oldsfound that there was no significant difference in theincrement from the pre- to the post-naming phase

between correct pronunciations and vowel mispronunci-ations. On the basis of these results, it would be possibleto suggest that infants at 15 months are not sensitive tomispronunciations of vowels. On the other hand, vowelmispronunciations did not lead to a significant incre-ment in infants’ preference for the target from the pre-to the post-naming phase, suggesting the opposing viewthat infants are sensitive to vowel mispronunciations infamiliar words—an interpretation supported by thefindings of Experiment 1.

Given the equivocal results of the data from the 15-month-olds when analysed separately, we ran an addi-tional analysis incorporating the data from all the age-groups to see if there were any significant differences inthe performance of the 15-month-olds from the oldergroup of infants, especially with regard to their sensitiv-ity to vowel mispronunciations. The difference betweenthe trials presented to the 15-month-olds (wake-up trialsprior to the main experiment) and the older infants (nowake-up trials prior to the main experiment) was notlarge enough to invalidate such a comparison. We rana mixed model ANOVA with naming and pronunciation(consonant mispronunciations, vowel mispronuncia-tions and correct pronunciations) as within-subject fac-tors and age (15, 18, and 24-month-olds) as a betweensubjects factor. If the 15-month-olds (in contrast to the18 and 24-month-olds) were not sensitive to vowel mis-pronunciations, then we would expect a significant inter-action between naming, pronunciation, and age.However, we would not expect a significant three-wayinteraction if 15-month-olds were as sensitive as theolder infants to both vowel and consonantmispronunciations.

The ANOVA confirmed that there was no main effectof age (F(2,81) = .60; p = .54) or significant interactionbetween naming, age and pronunciation (F(4,162) =.57; p = .68) or significant interactions between ageand any other factors (N * A: F(2,81) = .62; p = .53;P * A: F(4,162) = .27; p = .89). However, there was, asexpected a significant interaction between naming andpronunciation (F(2,80) = 7.88; p = .001). We analysedthe effect of age separately for the three mispronuncia-tions pairs. There was no significant interaction betweennaming, pronunciation and age when comparingvowel mispronunciations and correct pronunciations(F(2,81) = .61; p = .54); consonant mispronunciationsand correct pronunciations (F(2,81) = .61; p = .54) andconsonant mispronunciations and vowel mispronuncia-tions (F(2,81) = .43; p = .65). Similar results wereobtained using the PTL measure. These results indicatethat there was no systematic difference in infants’sensitivity between vowel and consonant mispronuncia-tions at 15, 18 and 24 months of age. However, itshould be noted that this conclusion is based on anull effect and should, therefore, be treated withcaution.


Effects of vocabulary size

We calculated the mean receptive percentile vocabu-lary size of the infants based on parental OCDI reports(15 months: M = 22%; SD = 11%; range = 3—54%;18 and 24 months: M = 47.74%; SD = 28.1%; range:0–100%). We measured the correlation between infants’vocabulary size and their sensitivity to mispronuncia-tions, as measured by the difference in naming effectbetween correct and mispronounced labels. There wasno evidence for any correlation (15 months: vowels:r = �.28, p = .13; consonants: r = .1, p = .3; 18 and

24 months: vowels: r = .10, p > .5; consonants:r = �.07, p > .5), between infants’ sensitivity to mispro-nunciation and size of infants’ receptive vocabulary.

We also calculated the mean productive percentilevocabulary size based on parental OCDI reports(15 months: M = 2%; SD = 2%; range: 0–6%; 18 and

24 months: M = 35.58%; SD = 37.4%; range: 0–100%).We measured the correlation between infants’ produc-tive vocabulary size and their sensitivity to mispronunci-ations, as above. There was no evidence for anycorrelation (15 months: vowels: r = .22, p = 14; conso-nants: r = .23, p = .22; 18 and 24 months: vowels:r = �.059, p = .72, consonants: r = �.27, p = .09).Hence, we found no evidence for a relationship betweenthe phonological specificity of infants’ lexical representa-tions and the size of infants’ productive or receptivevocabulary (as measured by the OCDI), thereby sup-porting the findings of Experiment 1 and of earlierstudies.

Discussion

Experiment 2 evaluated infant sensitivity to voweland consonant mispronunciations of a set of eight famil-iar CVC words. Each consonant mispronunciationinvolved a change along one-dimension (place or voice)of the onset consonant whereas vowel mispronuncia-tions involved a change along one-dimension (heightor backness) of the vowel nucleus. The results of theexperiment showed that 15-, 18- and 24-month-oldswere sensitive to both vowel and consonant mispronun-ciations. The evidence for this conclusion was thatinfants increased their looking, as measured by longestlooks and proportional looking times, towards a targetreferent upon hearing a correct pronunciation of itslabel but they failed to do so when the label was mispro-nounced, either on the onset consonant or the vowelnucleus. This finding replicates the vowel mispronuncia-tion effect reported in Experiment 1 and the consonantmispronunciation effect reported in earlier work (Bailey& Plunkett, 2002; Ballem & Plunkett, 2005; Swingley &Aslin, 2002). It is noteworthy that the vowel and conso-nant mispronunciation effects in Experiment 2 wereachieved without any repetition of trials, suggesting thatneither depends entirely on repetition of stimuli.

Latency measures also supported the conclusion thatinfants at 18 and 24 months are sensitive to mispronun-ciations of familiar labels, though in an unorthodoxfashion as far as consonant mispronunciations are con-cerned. For vowel mispronunciations, the time infantstook to switch their gaze from the distracter to the targetimage was slower than for correct pronunciations of thefamiliar word. This replicates the latency findingsreported by Swingley and Aslin (2000, 2002). However,infants were faster to switch to the target following anonset consonant mispronunciation than a correct pro-nunciation. This effect is opposite to that reported bySwingley and Aslin (2000, 2002) who find a slowerswitch speed for mispronunciations, whether theyinvolve the vowel or the consonant. It will be recalledthat in the current experiments, the target and distracterpairs shared the same onset consonants. This was notgenerally the case in the Swingley and Aslin studies.As noted earlier, switch latency is typically regarded asreflecting the time taken to detect a mismatch betweenthe retrieved semantic category and the currently fixatedpicture. In the current study, infant switch sensitivity isperhaps more accurately interpreted as indexing theinfants’ sensitivity to mismatch between the onset conso-nant and the distracter label.

The results with the 15-month-old infants were moredifficult to interpret. On the one hand, infants showed asignificant effect of naming following only correct pro-nunciations. Vowel and consonant mispronunciationsdid not lead to an increase in infants’ preference forthe target using either LLK or PTL measures. Thismight suggest that infants at 15 months show a symme-try in their sensitivity to vowel and consonant mispro-nunciations. However, there was a significantdifference in 15-month-olds’ looking behaviour onlybetween correct pronunciation and consonant mispro-nunciation trials. There was no significant difference intheir looking behaviour between correct pronunciationand vowel mispronunciation trials. Although there wasno difference in infants’ sensitivity to vowel and conso-nant mispronunciations, the absence of a significant dif-ference in infants’ looking behaviour following correctpronunciations and vowel mispronunciations mightindicate that 15-month-olds experience greater difficultyin establishing vowel identity than consonant identitywhen tested concurrently on their sensitivity to both.However, a comparison of infants’ behaviour acrossall the age-groups revealed that there were no systematicdifferences in infants’ sensitivity to vowel and consonantmispronunciations at 15, 18 and 24-months of age. Theabsence of a consonant–vowel asymmetry found at 18and 24-months also holds at 15 months.

The data also support the suggestion that the size ofthe mispronunciation effect reported for vowels and con-sonants does not differ as young as 15-months of age.This result indicates that infants are just as sensitive to


mispronunciations of the vowel nucleus as they are tothe onset consonant of familiar, monosyllabic CVCwords, at least when assessed using the inter-modal pref-erential looking task. We may conclude that both conso-nants and vowels play an important role in constrainingaccess to infants’ lexical representations during the earlystages of vocabulary development. The importance ofvowels in early lexical representation is reinforced bythe finding that infants exhibited a symmetry in theirsensitivity to vowel and consonant mispronunciations,despite the consonant mispronunciations all being situ-ated in the more salient word-initial position, while thevowel mispronunciations were all word-medial. Indeed,the results of the current study do not support the sug-gestion that consonants have a privileged status overvowels for early lexical recognition processes early inthe second year of life.

General discussion

The two experiments reported here provide a soundempirical foundation for the claim that infants are sensi-tive to mispronunciations of word-medial vowels infamiliar CVC labels when required to identify a targetreferent in an inter-modal preferential looking task.The results of Experiment 1 indicate that infants recog-nise such vowel mispronunciations across a wide rangeof words, certainly by 18-months-old. The performanceof the 15-month-olds was less robust: they only showedsensitivity to vowel mispronunciations during the secondblock of testing. These findings complement and extendthose of Swingley and Aslin (2000, 2002) who reportedinfant sensitivity to vowel mispronunciations of twofamiliar words in a similar task.

The results of Experiment 2 demonstrated thatinfants as young as 15-months of age were equally sen-sitive to vowel-medial and onset consonant mispronun-ciations of familiar, monosyllabic, CVC words whenrequired to identify a target referent, even when the mis-pronunciations involved just a single featural change.This result indicates that the identity of onset conso-nants and vowel nuclei are equally potent in constrain-ing lexical recognition by infants during the secondhalf of the second year of life.

However, the results with the 15-month-olds were,again, less robust. Although these infants did not showa significant effect of naming for vowel mispronuncia-tions, there was no difference in infants’ looking behav-iour following vowel mispronunciations and correctpronunciations. In contrast, consonant changes pro-duced a mispronunciation effect. This contrast pointsto the conclusion that 15-month-olds possess more frag-ile representations of vowels than consonants in familiarwords. However, there was no difference in infants’looking behaviour following consonant mispronuncia-

tions and vowel mispronunciations. Neither was therea difference in infants’ looking behaviour at 15, 18 and24-months of age, providing support for the conclusionthat 15-month-olds are equally sensitive to vowel andconsonant mispronunciations. We suggest that a morecomplete picture of 15-month-olds’ comparative sensi-tivity to vowel and consonant mispronunciationsrequires testing on a number of different issues. Forinstance, the consonant mispronunciations were alwaysword-initial, while the vowel mispronunciations wereword-medial, stacking the deck in favour of infantsbeing sensitive to the word-initial consonant mispronun-ciation. This was a constraint imposed by the relativeinfrequency of vowel-initial words in the infant lexicon.The absence of a significant difference in infants’ lookingbehaviour following vowel mispronunciations and cor-rect pronunciations may have been motivated by theposition of the mispronunciation within the word, ratherthan the identity of the mispronunciation. Importantly,despite the odds being in favour of the consonant mis-pronunciations, infants display sensitivity to vowel mis-pronunciations as early as 15-months of age—providingsupport for the view that vowels and consonants areequally well-specified early in the second year of life.Note, however, that our conclusion of a symmetry in15-month-olds’ sensitivity to vowel and consonantmispronunciations was based on a null effect, hence,our interpretation of this finding is not definitive.

The primary finding of Experiment 2 is that vowelsand consonants constrain lexical access equally early inthe second year of life. Previous research by Nazzi(2005) suggests that consonants have a privileged statusin lexical acquisition. Nazzi found that 20-month-oldFrench infants were able to simultaneously acquiretwo words that differed only by a single consonant, whilenot being able to learn two words that differed by a sin-gle vowel. There are a number of reasons for the behav-ioural differences between the infants in the currentstudy and Nazzi’s experiment. First, Nazzi’s experimentemployed novel words, while the current study presentedinfants with highly familiar words. Earlier experimentshave found that there are differences in the phonologicalspecification of novel and familiar words. For instance,Stager and Werker (1997) found that 14-month-oldshave difficulty learning to associate two phoneticallyminimal novel words with two novel objects (e.g., ‘bih’and ‘dih’)—suggesting underspecification of novelwords. Using a similar habituation task, however, Fen-nell and Werker (2003) found that 14-month-olds aresensitive to violations of object-label pairings of twophonetically similar familiar words (e.g., ‘ball’ and‘doll’)—infants look longer at a target image when theimage does not match the label than when it does. Thesefindings are consistent with the hypothesis that familiarwords may possess a greater level of phonological spec-ificity than novel words. This interpretation provides


one possible explanation for the differences betweenNazzi’s results and those of the current study.

Second, Nazzi employed a novel word-learning taskwhich required infants to learn two very similar novelwords and categorise three objects based on these novellabels. Word-learning tasks can be highly demanding, tosay nothing of the added computational requirements ofthe categorisation task. This may require infants toignore information that is normally accessed when lis-tening to familiar words. For example, Stager and Wer-ker (1997) found that 14-month-old infants could readilylearn two novel words which were not minimal pairs(e.g., ‘lif’ and ‘neem’). In contrast, the current studymerely required infants to notice a match between afamiliar target object and a target label, where a labelcorresponding to the distracter image shared only theonset consonant with the target label. It is conceivablethat the lack of any differences between vowel and con-sonant mispronunciations in the current study was dri-ven by the low processing demands associated with thepreferential looking task: making the task more difficultmight reveal a difference between vowel and consonantsensitivity, even in 18-month-olds.

Third, in his study, Nazzi presented infants with sim-ilar-sounding monosyllabic and bisyllabic word-pairs.His conclusion that vowels are less central then conso-nants in constraining lexical access is based on resultsaveraged across these mono- and bisyllabic words. How-ever, his French infants were successfully able to learntwo monosyllabic novel words differing only in a mini-mal vowel in one of three monosyllabic conditions(p < .001). Conversely, infants performed significantlybelow chance when required to learn two complex bisyl-labic words differing only in the vowel in the initial syl-lable (p < .03). It is possible that learning monosyllabicwords makes it easier for infants to pay more attentionto all the phonological information present in the stim-uli. This would be entirely consistent with the resultsreported in the current study, since infants were pre-sented only with monosyllabic words and were able todistinguish correct and incorrect pronunciations in theservice of identifying a referent.

We have argued that the difference between Nazzi’sresults and those of the current study might be attributedto the novelty of the stimuli and that the complexity of thebisyllabic stimuli may be detracting from vowel sensitiv-ity. However, Ballem and Plunkett (2005) suggest thatnovel word representations may be more phonologicallyspecified than previously assumed. Hence, if novel wordsare not underspecified, the difference between Nazzi’sresults and the current study would appear to implicatethe syllabic complexity of the stimuli tested.

An alternative source of the differences between thecurrent study and Nazzi (2005) might lie in cross-linguisticdifferences between the vowel systems of French andEnglish. The French inventory is larger than English

and possesses a greater number of feature contrasts thanEnglish. Earlier studies have suggested that phonologicalrepresentations of sounds increase in specificity withincrease in the size of the vowel inventory of the language(Bohn, 2004). However, contrary to the differences wehave found between Nazzi (2005) and the current study,this would predict that English infants would be less sen-sitive to vowel mispronunciations than French infants.

On the other hand, Frisch, Pierrehumbert, and Broe(2004) have argued that an increase in the number of con-trastive features shared by the vowels in a language may leadto greater perceived similarity between the vowels. Conse-quently, English infants may be more sensitive to vowelmispronunciations in words than French infants. It shouldbe noted that perceived similarity cannot be calculated with-out careful consideration of the number and nature of thecontrastive features defining the phonological inventory.Hence, this cross-linguistic account must remain speculative.

Conclusion

Early lexical representations of familiar words con-tain adequate information for very young Englishinfants to detect mispronunciations of the vocalicnucleus of a target label. The results from the threeage groups tested indicate that this vowel sensitivity isin place by 15-months-old. These findings constitutethe first experimental evidence that vowel identity con-strains lexical access for a wide range of monosyllabic,familiar words at an early stage of lexical development.We found no evidence of a relationship between vowelsensitivity and vocabulary size and so no support forthe view that the density of lexical neighbourhoods is afactor driving phonological specificity of words.

The similarity in infants’ sensitivity to mispronuncia-tions of vowels and consonants during lexical recogni-tion found in this study offers some qualifications tothat of the recent study by Nazzi (2005) who argues thathis findings might ‘‘be interpreted as the first piece ofevidence for a greater reliance on consonants at the lex-ical level in infancy’’ (p. 28). Our findings rule out thepossibility that consonants play a more pivotal role thanvowels in lexical processing in infancy, at least in recog-nition of familiar words at 18 months of age. Both con-sonants and vowels constrain lexical recognition equallyin the latter half of the second year of life.

There are many questions regarding the role of vow-els in lexical recognition in infancy that the current studyleaves unanswered. Definitive conclusions regarding theimportance of vowels in lexical representation are con-tingent upon further empirical validation of the robust-ness of vowel mispronunciation effects in the recognitionof novel and familiar words, and mono- and bisyllabicwords. Exploration of cross-linguistic factors might alsoinfluence our understanding of the role of vowels in


lexical access. For example, languages like Danish andSwedish have almost twice the number of vowels asEnglish and would provide a test case for exploringthe influence of cross-linguistic differences on vowelmispronunciation effects.

Future studies could also attempt to isolate thedevelopmental onset of vowel mispronunciation effects:all the age-groups tested in the current study showed somesign of sensitivity to vowel mispronunciations. Vowel-spe-cific phonological characteristics may also influence theimportance of vowels in lexical access: some vowel changesmay be more perceptually salient than others. Earlierstudies have suggested that there are perceptual asymme-tries in vowel discrimination such that a vowel change froma less peripheral vowel (in terms of location in formantspace) to a more peripheral vowel is more easily discrimi-nable than a change the other way around (Bohn & Polka,2001; Polka & Bohn, 2003). While not all vowel changeshave been tested for asymmetries, this finding presents atheoretical perspective that can be used to compare theimpact of vowel changes presented to infants. Further-more, the mispronunciations in the current study involvedthe backness, height, and/or roundedness of the vowel.There may be differences in the salience of individualfeatural changes, with changes in one feature being percep-tually (and possibly acoustically) more discriminable thanchanges to another feature.

Finally, there is a caveat to the conclusions drawn bythe current study. We have argued that infants were sen-sitive to changes to the vowels of the words presented tothem. However, there are likely to be vowel–consonantco-articulation effects that might lead to significant dif-ferences in the acoustic characteristics of the consonantbetween the correctly pronounced and mispronouncedwords. Infant performance might have been triggeredby the acoustic differences in the co-articulated conso-nants between the correct and incorrect pronunciationsrather than because of the changes to the vowels them-selves. On this interpretation of the results, sensitivityto mispronunciations would be led by differences inthe acoustic characteristics of the consonant and notthe vowel. While this suggests a less pivotal role for vow-els in lexical representation, there is no empirical evi-dence suggesting that infants’ performance isdependent on the acoustic characteristics of the co-artic-ulated consonant alone.

Acknowledgments

The authors would like to acknowledge the contribu-tion of John Coleman, Janet Werker, Thierry Nazzi andan anonymous reviewer for their comments on earlierversions of the manuscript. This research was supportedby an ESRC grant (# RES-000-23-1322) awarded toKim Plunkett.

Appendix A

Differences between one- and many-dimension

mispronunciations

While it is always difficult to quantify the different degreesof vowel mispronunciations, in this section we attempt to findout if infants’ display any sensitivity to the size of the mispro-nunciation presented to them. Experiment 1 was not specificallydesigned to investigate the impact of the degree of vowel mis-pronunciation. However, infants were presented with differentkinds of mispronunciations, specifically varying between one-dimension changes and many-dimension changes. The threedimensions considered crucial in this study were vowel height,backness, and roundedness. We analysed whether infantsshowed any difference in their sensitivity to mispronunciationsthat involved one-dimension changes to the vowel comparedto mispronunciations that involved more than two- or three-dimension changes to the vowel. Tables 1 and 2 report thenumber of dimensions differentiating the correct and incorrectpronunciations of the words presented to infants.

We only considered those infants who were presented with anequal number of one- and many-dimension mispronunciations(i.e., 1 trial per condition per infant at 15 months and 2 trialsper condition per infant at 18 and 24 months), in order to ensuresimilar variability in each condition. This permitted analysis of 815-month-olds (from Block 2), 12 18-month-olds, and 15 24-month-olds. We then computed the mispronunciation effect sep-arately for one- and many-dimension mispronunciations. Thiswas calculated as the difference in the increment from the pre-to the post-naming phase between correct pronunciation andeither kind of mispronunciation trials (i.e., [(postcorrect �precorrect) � (postmis1� premis1)] for one-dimension mispronuncia-tions and [(postcorrect� precorrect) � (postmis-many� premis-many)]for many-dimension mispronunciations).

15 months

Fifteen-month-olds do not show a significant mispronuncia-tion effect for one-dimension mispronunciations (LLK: 151 ms(CI:�939,1241); PTL: .19 (CI:�.16, .56)), but did show a signif-icant mispronunciation effect for many-dimension mispronunci-ations (LLK: 1425 ms (CI: 491,2371), d = .63; PTL: .38 (CI:.03, .72), d = .54). The difference between these mispronunciationeffects was significant using the LLK measure (M = 1284 (CI:325,2243), d = .54) but not the PTL measure (M = .18 (CI:�.24, .61)). Fifteen-month-olds appear to be more sensitive tomany-dimension mispronunciations than to one-featuremispronunciations.

18 months

Eighteen-month-olds do not show a significant mispronun-ciation effect for one-dimension mispronunciations (LLK:522 ms (CI: �649,1694); PTL: .13 (CI: �.13, .41)) or formany-dimension mispronunciations (LLK: �122 ms (CI:�1364,1119); PTL: �.03 (CI: �.29, .22)). There was no signifi-cant difference between infants’ looking behaviour followingone- and many-feature mispronunciations using the LLK mea-sure (M = �644 (CI: �2241,951)) or the PTL measure


(M = �.17 (CI:�.50, .15)). Hence, 18-month-olds show sensitiv-ity to vowel mispronunciations of familiar words but apparentlydo not recognise differences in the size of the mispronunciation.However, the amount of variance within the small sample(N = 12) might provide some explanation for the failure to sepa-rately find an effect of sensitivity to one-dimension and many-dimension mispronunciations at 18 months. Hence, interpreta-tion of the results of this analysis must remain cautious.

24 months

Twenty-four-month-olds do not show a significant mispro-nunciation effect for one-dimension mispronunciations (LLK:�241 ms (CI: �829,345); PTL: �.07 (CI: �.24, .08)), but showa near-significant mispronunciation effect for many-dimensionmispronunciations (LLK: 327 ms (CI: �67,722); PTL: .06(CI: �.07, .20)). However, there was a significant differencebetween infants’ looking behaviour following one- and many-feature mispronunciations using the LLK measure(M = 569 ms (CI: 72,1465), d = .39) and the PTL measure(M = .14 (CI: .02, .25), d = .35). Hence, at 24 months of age,infants appear to be more sensitive to many-dimension mispro-nunciations than to one-dimension mispronunciations. Whilethe Cohen’s ‘d’ values confirm that the effects reported in thisanalysis are large, there is, however, cause for some concernover the size of the sample in the different age-groups. Noneof the analyses using this reduced dataset yielded a significantmispronunciation effect for one-dimension mispronunciationsat 15, 18 or 24 months. This might suggest that infants are onlysensitive to large mispronunciations of the vowels in familiarwords. However, the absence of a significant mispronunciationeffect for one-dimension mispronunciations might also beexplained by the lack of power in the analysis. Indeed, the lackof comparability between the mean difference in looking timespresented in the current analysis and those in the main analysisadvise against placing too much emphasis on the results of thecurrent analysis.

The results of Experiment 2, where only one-dimensionalchanges were tested, indicate that 15- to 24-month-old infantsare sensitive to small changes to vowels of familiar words. Nev-ertheless, the fact that the small dataset analysed in this appen-dix yields a significant result for large changes but not for smallchanges also indicates that infants are sensitive to the size of themispronunciation with which they are presented.

References

Bailey, T., & Hahn, U. (2005). Phoneme similarity andconfusability. Journal of Memory and Language, 52,339–362.

Bailey, T., & Plunkett, K. (2002). Phonological specificity inearly words. Cognitive Development, 17, 1265–1282.

Ballem, K., & Plunkett, K. (2005). Phonological specificity in14-month-olds. Journal of Child Language, 32, 159–173.

Bohn, O. (2004). How to organize a fairly large vowelinventory: The vowels of Fering (North Frisian). Journal

of the International Phonetic Association, 34, 161–173.Bohn, O., & Polka, L. (2001). Target spectral, dynamic spectral,

and duration cues in infant perception of German stressed

vowels. Journal of the Acoustical Society of America, 110,504–515.

Bond, Z. (1954). Listening to elliptic speech: Pay attention tostressed vowels. Journal of Phonetics, 9, 89–96.

Caramazza, A., Chialant, D., Capasso, R., & Micell, G. (2000).Separable processing of consonants and vowels. Nature,

403, 428–430.Cutler, A., Sebastian-Galles, N., Soler-Vilageliu, O., & van

Ooijen, B. (2000). Constraints of vowels and consonants onlexical selection: Crosslinguistic comparisons. Memory &

Cognition, 28, 746–755.Eimas, P., Siqueland, E., Jusczyk, P., & Vigoriti, J. (1971).

Speech perception in infants. Science, 171, 303–306.Fenson, L., Dale, P. S., Reznick, J. S., Thal, D., Bates, E.,

Hartung, J. P., et al. (1993). MacArthur communicative

development inventories: User’s guide and technical manual.San Diego, CA: Singular Publishing.

Fernald, A., Swingley, D., & Pinto, J. (2001). When half a wordis enough: Infants can recognise spoken words using partialphonetic information. Child Development, 72, 1003–1015.

Fennell, C., & Werker, J. (2003). Early word learners’ ability toaccess phonetic details in well-known words. Language and

Speech, 46, 245–264.Frisch, S. A., Pierrehumbert, J. B., & Broe, M. B. (2004).

Similarity avoidance and the OCP. Natural Language and

Linguistic Theory, 22, 179–228.Gerken, L., Murphy, W. D., & Aslin, R. N. (1995). 3-year olds’

and 4-year-olds’ perceptual confusions for spoken words.Perception & Psychophysics, 57, 475–486.

Hamilton, A., Plunkett, K., & Schafer, G. (2000). Infantvocabulary development assessed with a british communi-cative development inventory. Journal of Child Language,

27, 689–705.Kuhl, P. K. (1987). Perception of speech and sound in early

infancy. In P. Salapatek & L. Cohen (Eds.). Handbook of infant

perception (2, pp. 257–381). New York: Academic Press.Kuhl, P., Williams, K., Lacerda, F., Stevens, K., & Lindblom,

B. (1992). Linguistic experience alters phonetic perception ininfants by 6 months of age. Science, 255, 606–608.

Liberman, A., Delattre, P., Cooper, F., & Gerstman, L. (1954).The role of consonant–vowel transitions in the perception ofthe stop and nasal consonants. Psychological Monographs,

68, 1–13.Maye, J., Werker, J., & Gerken, L. (2002). Infant sensitivity to

distributional information can affect phonetic discrimina-tion. Cognition, 82, B101–B111.

Meints, K., Plunkett, K., & Harris, P. L. (1999). When does anostrich become a bird: The role of prototypes in early wordcomprehension. Developmental Psychology, 35, 1072–1078.

Nazzi, T. (2005). Use of phonetic specificity during theacquisition of new words: Differences between consonantsand vowels. Cognition, 98, 13–30.

Nazzi, T., & Gopnik, A. (2001). Linguistic and cognitiveabilities in infancy: When does language become a tool forcategorisation? Cognition, 80, B11–B20.

Nespor, M., Pena, M., & Mehler, J. (2003). On the different roleof vowels and consonants in speech processing and languageacquisition. Lingue e Linguaggio, 2, 221–247.

Peterson, G., & Barney, H. (1952). Control methods used in astudy of the vowels. Journal of the Acoustical Society of

America, 24, 695–705.


Pisoni, D. B. (1973). Auditory and phonetic memory codes inthe discrimination of consonants and vowels. Perception &

Psychophysics, 13, 253–260.Polka, L., & Bohn, O. (2003). Asymmetries in vowel perception.

Speech Communication, 41, 221–231.Reznick, J. S. (1990). Visual preference as a test of infant word

comprehension. Applied Psycholinguistics, 11, 145–166.Robb, M., & Bleile, K. (1994). Consonant inventories of young

children from 8 to 25 months. Clinical Linguistics and

Phonetics, 8, 295–320.Schafer, G., & Plunkett, K. (1998). Rapid word learning by 15-

month-olds under tightly controlled conditions. Child

Development, 69, 309–320.Selby, J., Robb, M., & Gilbert, H. (2000). Normal vowel

articulations between 15 and 36 months of age. Clinical

Linguistics and Phonetics, 14, 255–265.Smith, B. L. (1978). Temporal aspects of English speech

production: A developmental perspective. Journal of Pho-

netics, 6, 37–67.Southgate, V., & Meints, K. (1999). What typicality

ratings with infants can tell us about the earlylexicon. Paper presented at the International Confer-ence on Cognitive Linguistics, Stockholm, Sweden,August 1999.

Stager, C. L., & Werker, J. F. (1997). Infants listen for morephonetic detail in speech perception than word learningtasks. Nature, 388, 381–382.

Stoel-Gammon, C., & Herrington, P. (1990). Vowel systems ofnormally developing and phonologically disordered chil-dren. Clinical Linguistics and Phonetics, 4, 145–160.

Swingley, D. (2003). Phonetic detail in the developing lexicon.Language and Speech, 46, 245–264.

Swingley, D., & Aslin, R. N. (2000). Spoken word recognitionand lexical representation in very young children. Cognition,

76, 147–166.Swingley, D., & Aslin, R. N. (2002). Lexical neighborhoods and

the word form representations of 14-month-olds. Psycho-

logical Science, 13, 480–484.Swingley, D., Pinto, J. P., & Fernald, A. (1999). Continuous

processing in word recognition at 24 months. Cognition, 71,73–108.

Templin, M. (1957). Certain language skills in children: Their

development and inter-relationships. Institute of Child wel-

fare. Monograph (26). Minneapolis, MN: University ofMinnesota Press.

Van Ooijen, B. (1996). Vowel mutability and lexical selection inEnglish: Evidence from a word reconstruction task. Memory

& Cognition, 24, 573–583.Werker, J., & Tees, R. (1984). Cross-language speech percep-

tion: Evidence for perceptual reorganisation during the firstyear of life. Infant Behaviour and Development, 7, 49–63.

Werker, J. F., & Lalonde, C. E. (1988). Cross-language speechperception: Initial capabilities and developmental change.Developmental Psychology., 24, 672–683.

Phonological specificity of vowels and consonants in early ...

Documents