Content uploaded by Drew Weatherhead
Author content
All content in this area was uploaded by Drew Weatherhead on May 19, 2016
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=hlld20
Download by: [University of Waterloo] Date: 19 May 2016, At: 09:53
Language Learning and Development
ISSN: 1547-5441 (Print) 1547-3341 (Online) Journal homepage: http://www.tandfonline.com/loi/hlld20
He Says Potato, She Says Potahto: Young Infants
Track Talker-Specific Accents
Drew Weatherhead & Katherine S. White
To cite this article: Drew Weatherhead & Katherine S. White (2016) He Says Potato, She Says
Potahto: Young Infants Track Talker-Specific Accents, Language Learning and Development,
12:1, 92-103, DOI: 10.1080/15475441.2015.1024835
To link to this article: http://dx.doi.org/10.1080/15475441.2015.1024835
Published online: 09 Nov 2015.
Submit your article to this journal
Article views: 72
View related articles
View Crossmark data
He Says Potato, She Says Potahto: Young Infants Track
Talker-Specific Accents
Drew Weatherhead and Katherine S. White
Department of Psychology, University of Waterloo
ABSTRACT
One of the most fundamental aspects of learning a language is determin-
ing the mappings between words and referents. An often-overlooked
complication is that infants interact with multiple individuals who may
not produce words in the same way. In the present study, we explored
whether 10- to 12-month-olds can use talker-specific knowledge to infer
the intended referents of novel labels. During exposure, infants heard two
talkers whose front vowels differed; one talker trained them on a word-
referent mapping. At test, infants saw the trained object and a novel
object; they heard a single novel label from both talkers. When the label
had a front vowel (Experiment 1), infants responded differently as a
function of talker, but when it had a back vowel (Experiment 2), they
did not, mapping the novel label to the novel object for both talkers.
These results suggest that infants can track the phonetic properties of two
simultaneously presented talkers and use information about each talker’s
previous productions to guide their referential interpretations.
Introduction
One of the most fundamental aspects of learning a language is determining the mappings between
words and referents. However, learning even the words themselves presents a formidable categor-
ization problem due to rampant variability in the speech signal. In addition to within-speaker
variability, infants interact with multiple individuals, who may not produce words in the same
way. This can be due to physical or idiosyncratic differences across speakers as well as to systematic
language-based differences. For example, consider two individuals from different regions of the
United States: Sarah says “bag”to refer to a sack, while John says “beg”for the same object. More
confusingly, this word, “beg,”is similar to Sarah’s word for pleading. How do we determine when
the same phonetic form (e.g., Sarah’s and John’s“beg”) maps onto different word categories (and,
therefore, meanings) and when different phonetic forms (e.g., Sarah’s“bag”vs. John’s“beg”) map
onto the same word category (and meaning)? One source of information is context (e.g., if the
speaker’s attention is directed toward a sack). However, contextual information is not always
available or unambiguous. For adults, another important source of information is knowledge
about the speaker’s language background. This can come in the form of general knowledge about
the speaker’s language community, which can activate stored information about that community’s
accent (Hay, Nolan, & Drager, 2006), or from direct observation of a speaker’s productions.
Encoding such “talker-specific”information not only allows adults to understand the particular
words someone has produced before, but also to make inferences about other words. If Sarah has
previously heard John say “beg”for a sack, she might infer that “teg”is his pronunciation of her
“tag.”However, if she knows nothing about him, she might treat “teg”as a new word, because
CONTACT Drew Weatherhead deweathe@uwaterloo.ca Department of Psychology, University of Waterloo, 200 University
Avenue West, Waterloo, ON, Canada, N2L 3G1.
© 2016 Taylor & Francis
LANGUAGE LEARNING AND DEVELOPMENT
2016, VOL. 12, NO. 1, 92–103
http://dx.doi.org/10.1080/15475441.2015.1024835
Downloaded by [University of Waterloo] at 09:53 19 May 2016
listeners have a bias to assume one-to-one mappings between phonetic form and meaning. This bias
likely contributes to the difficulty that listeners initially have in understanding a talker with an
unfamiliar accent (e.g., Bradlow & Bent, 2008; Clarke & Garrett, 2004).
The problems posed by this type of accent variability are potentially much more significant for
young language learners. One reason for this is that young learners adhere more strongly than adults
to the assumption that novel phonetic forms should be mapped to novel referents—an adaptive
assumption, given how often new words occur in their environments. A large body of research has
demonstrated that learners as young as 6 months interpret novel wordforms as labels for novel
objects, although the mechanism underlying this mapping preference may change across develop-
ment (e.g., Golinkoff, Mervis, & Hirsh-Pasek; Halberda, 2003; Markman, 1989,1990; Merriman,
Bowman, & MacWhinney, 1989; Shukla, White, & Aslin, 2011). Whether this mapping bias results
from exclusion reasoning (objects have only one label) or a mapping of novelty (of the label) to
novelty (of the referent), accented speech poses a challenge: if an accented pronunciation is judged to
be different from known words, it will be treated as a new word. In other words, learners will posit
wordform-meaning mappings that do not exist, potentially slowing lexical development. Indeed,
children sometimes map mispronunciations of familiar words to novel referents (Mani & Plunkett,
2011; Merriman & Schuster, 1991; White & Morgan, 2008).
Recent work demonstrates that young learners do have some difficulty processing accented
words. Infants are unable to recognize familiarized wordforms across accents until the end of the
first year (Schmale & Seidl, 2009; Schmale, Cristia, Seidl, & Johnson, 2010) and have difficulty
recognizing accented versions of known words even at later ages (Best, Tyler, Gooding, Orlando, &
Quann, 2009), unless they are given sufficient exposure to the accent (van Heugten & Johnson,
2014). Similarly, it is not until 19 months that toddlers recognize, under some conditions, that
accented pronunciations map onto familiar referents (Mulak, Best, Tyler, & Kitamura, 2013;
Schmale, Hollich, & Seidl, 2011; Schmale, Cristia, & Seidl, 2012; White & Aslin, 2011). Even 4-
year-olds have difficulties recognizing familiar words in an unfamiliar dialect (Nathan, Wells, &
Donlan, 1998).
Here, we explore whether infants can encode the phonetic properties of particular individuals, and
use this talker-specific knowledge to overcome the problems posed by accent variability. Just as
children realize that the one-to-one mapping assumption does not operate across languages (Au &
Glusman, 1990), they might also realize that what counts as a novel label depends on a speaker’s
accent. Therefore, if learners can track the differences between two talkers’accents, they may under-
stand that whether two phonetic forms refer to the same or different referents depends on the speaker.
Infants have the ability to link certain types of vocal properties and speakers, at both a group and
an individual level. Infants can match a familiar accent to a face of a familiar race and an unfamiliar
accent to a face of an unfamiliar race (Uttley et al., 2013). Infants are also capable of remembering
links between talkers and the global properties of their speech: they prefer to look at an individual
who previously spoke their native language over one who previously spoke a foreign language
(Kinzler, Dupoux, & Spelke, 2007). However, in the case of native vs. foreign speech, it is enough
to recognize simply that one speaker sounds familiar and the other sounds unfamiliar. Tracking
specific phonetic properties across speakers where there is no familiarity difference is a more
challenging task.
We asked whether 10- to 12-month-olds could learn about the specific properties of two talkers’
accents and, in the absence of any contextual information, use that talker-specific information to
determine the intended referent of novel words. Previous research on early accent processing has
considered whether infants can map a novel accent to their own accent, but not whether they can
track speaker-specific phonetic information. In addition, although it has been shown that exposure
improves infants’and toddlers’recognition of accented speech, which properties of the accent are
learned has remained virtually untested (but see White & Aslin, 2011). We chose to focus on 10- to
12-month-olds because, by the end of the first year, infants have started tuning to the relevant sound
properties of their native language (Houston & Juscyzk, 2000; Singh, White, & Morgan, 2008;
LANGUAGE LEARNING AND DEVELOPMENT 93
Downloaded by [University of Waterloo] at 09:53 19 May 2016
Werker & Tees, 1984), making the kind of accent variability used in the present study (which
involves phonemic category changes) disruptive.
We presented infants with two talkers whose productions systematically differed in the
height of their front vowels –a“Training”Speaker and an “Extension”Speaker. The Extension
Speaker’s front vowels were higher than the Training Speaker’s. We chose to manipulate vowel
height across speakers, as accent differences commonly involve vowels, and English-learning
infants are sensitive to various types of vowel contrasts, including subtle distinctions like [i] vs.
[I] (Swoboda, Morse, & Leavitt, 1976), [a] vs. [ɔ](Kuhl,1983), and [e] vs. [E] (Sundara &
Scutellaro, 2011). In addition, 14-month-olds are sensitive to a range of vowel changes in both
familiar and newly trained words (Mani & Plunkett, 2008;Mani,Mills,&Plunkett,2012).
Following this exposure, infants learned the label for a novel object from the Training Speaker
(tEpu), but did not hear the Extension Speaker label it. In other words, infants were not
directly exposed to the Extension Speaker’s label for the object. In Experiment 1, at test,
infants saw this trained object and an untrained object, and heard each talker use the label
tIpu. If infants are able to track the systematic difference between the speakers, their inter-
pretation of the test label tIpu should differ as a function of the talker’s identity. In Experiment
2, we changed the test label to topu. If infants have learned that the talkers differ specifically in
their front vowels, their interpretation of the test label topu (with only back vowels) should not
differ by talker.
Experiment 1
Participants
Thirty 10- to 12-month-olds were tested (11 females and 19 males; mean age: 324 days; age
range: 297–358 days). Ten additional participants were tested but not included due to non-
completion (2), parental headphone difficulties (3), failure to attend to both objects during the
baseline period (3), or difference scores exceeding 2.5 standard deviations from the mean of
either speaker (2).
Stimuli
Audio stimuli
The stimuli consisted of four pairs of CVCV nonsense words (see Table 1)producedbytwo
female native speakers of English. The pronunciation of the first vowel (always a front vowel)
varied by talker, but the remainder of the word (including the second, back, vowel) did not
differ across talkers. Three of the word pairs (m[I/i]to, d[E/I]lu,andb[I/i]mo
1
)werepresented
during exposure without referents. The word tEpu was used by the Training Speaker during
exposure to label an object. The last word, tIpu, was heard in test. Stimuli were recorded in a
sound-treated booth at a sampling rate of 44100 Hz and were later equated for amplitude in
Praat (Boersma & Weenink, 2009). See Table 2 for acoustic information. The audio stimuli for
exposure were inserted into the videos described below.
Table 1. Audio stimuli used during exposure.
Word type Training speaker Extension speaker
Exposure Pair 1 mIto mito
Exposure Pair 2 dԑlu dIlu
Exposure Pair 3 bImo bimo
Trained object label tԑpu
1
[I] represents the sound in “big”, [i] the sound in “beep”, and [o] (Experiment 2) the sound in “boat.”
94 D. WEATHERHEAD AND K. S. WHITE
Downloaded by [University of Waterloo] at 09:53 19 May 2016
Audiovisual stimuli (exposure phase)
Both talkers, Caucasian females aged approximately 22 years, were recorded against the same
backdrop. They were dressed in different colored t-shirts to provide a salient cue that they were
different people. Each talker recorded three exposure videos, in which a single exposure word was
repeated three times in infant-directed speech approximately two seconds apart. Each talker also
recorded an object presentation event. In the Training Speaker’s object presentation event, she
held and waved the target object while labelling it tEpu three times (this object is hereafter
referred to as the trained object). In the Extension Speaker’s object presentation event, she held
and waved the trained object, but did not label it. Infants were either trained with an unfamiliar
blue object or an unfamiliar yellow object.
Procedure
The participant sat on his/her parent’slap approximately 1.5 ft. from a 36x21-inch plasma screen
television in a sound-treated testing room. A camera under the television recorded the child’s
looking behavior for the entirety of the session. The camera was linked to a monitor and recording
device in the lab area adjacent to the testing room for the experimenter’s viewing purposes and for
later off-line coding. Stimuli were played at approximately 65dB and presented in Psyscope X
(Cohen, MacWhinney, Flatt, & Provost, 1993). Parents were instructed not to interact with their
infants during the session and wore noise-cancelling headphones playing instrumental music to
mask the audio being played to the infant.
The first video pair of the exposure phase involved the object presentation events from both
talkers, to signal to the infants that they were in a word-learning situation. Next, the three
pairs of yoked exposure videos (e.g., mIto-mito) were presented in random order (see Table 1).
These pairs served to highlight the front-vowel difference between the talkers. The object
presentation event pair was repeated twice at the end of the exposure phase. Overall, infants
heard the trained object labeled nine times by the Training Speaker. An attention getter
occurred between the video pairs, with the next pair beginning when the experimenter judged
that the participant was focused on the attention getter. See Figure 1 for a schematic of the
exposure phase.
Thetestphasebeganimmediatelyaftertheexposure. There were two test trials, one for each
talker. Each trial was 10 seconds in length. At the start of each trial, the talker’sfaceand
shoulders appeared alone for 2 seconds, followed by a display with the trained object and a novel
untrained object. The objects remained on the screen for 8 additional seconds, the first 3 seconds
of which was a silent baseline period, followed by an audio recording of the pictured talker
saying the test word (tIpu). The talker in the first test trial and the side on which the trained
Table 2. Acoustic information (first and second formant of the critical vowel in Hz, mean pitch of the word in Hz, and word
duration in seconds) for key tokens used in both experiments. The first column refers to the Training Speaker’s pronunciation of
tEpu, which appears in the object presentation phase of both experiments. The values in this column are a calculated mean across
the 3 tokens used during the object presentation event. The second column refers to each speaker’s pronunciation of the test
word in Experiment 1 (tIpu). The third column refers to each speaker’s pronunciation of the test word in Experiment 2 (topu). For
the test words, each value was calculated for the single token.
Trained label: tEpu
(Experiment 1 & 2)
Test word: tIpu
(Experiment 1)
Test word: topu
(Experiment 2)
Training speaker Training speaker Extension speaker Training speaker Extension speaker
F1 914 539 618 544 508
F2 2357 2599 2620 882 1098
Mean Pitch 276 288 257 289 253
Duration 0.62 0.68 0.64 0.66 0.76
LANGUAGE LEARNING AND DEVELOPMENT 95
Downloaded by [University of Waterloo] at 09:53 19 May 2016
object appeared were counterbalanced across participants (this side assignment remained con-
stant for both test trials). See Figure 2 for a schematic of the test trials.
If infants learned the trained label tEpu from the Training Speaker during the exposure phase,
then the novel label tIpu should be mapped to the untrained object for this talker. If, in addition,
they learned that the two talkers differ in their pronunciations of front vowels and, in particular, that
the Extension Speaker had higher front vowels, then they should interpret tIpu as the Extension
Speaker’s pronunciation of the trained object’s label. In that case, they should look longer to the
trained object for this talker.
Coding of looking times
Looking time during the test phase was coded off-line using in house software (Brown
University), frame-by-frame (1 frame = 33 msec). Looking proportions for the objects were
determined for the baseline period and for the test period, which began 430 msec after test
word onset. This delay corresponded to the time necessary to program an eye movement in
response to the first vowel in tIpu (shifting the analysis window at test is a common practice in
word recognition studies, e.g., Bailey & Plunkett, 2002; Swingley & Aslin, 2002;White&Aslin,
2011). To equate the length of the baseline and test periods, only the first 3 seconds of the test
period were analyzed.
Figure 1. SchematicoftheExposurePhase:Theexposurephasebegins with one Object Presentation Event, followed by
the three Exposure Events, followed by two more Object Presentation Events. In each event, the Training Speaker is seen
first (approximately 8 seconds), followed by the Extension Speaker (approximately 8 seconds). In each event the speaker is
alone on the screen. We present them together in the Figure to highlight the alternation.
96 D. WEATHERHEAD AND K. S. WHITE
Downloaded by [University of Waterloo] at 09:53 19 May 2016
Results
For both the baseline and test periods, the proportion of time infants spent looking at each of the
objects was computed (out of the total 3 seconds for each phase).
2
During the baseline period, there
was no difference in looking to the trained and untrained objects for the Extension Speaker
(proportions of .44 and .43, respectively; t(29) = 0.184, ns)
3
; however, there was an asymmetry for
the Training Speaker (.50 and .37; t(29) = 2.0, p= .054), which is addressed below. Using the
proportions for each period, a difference score was calculated for each trial (proportion object
test
-
proportion object
baseline
). This measure indicates the change in looking towards an object after
labelling. Figure 3 displays the difference scores.
A repeated measures ANOVA on these difference scores with within-subjects factors of Speaker
and Object and a between-subjects factor of test Order revealed no significant main effects of
Speaker (F(1,28) = 0.566, ns) or Object (F(1,28) = 0.091, ns), but a significant interaction between
Speaker and Object (F(1,28) = 6.530, p= .016). There was also a marginal interaction between
Speaker, Object, and test Order (F(1,28) = 4.070, p= .053).
To determine the effect of labeling for each talker separately, one-sample t-tests compared
difference scores for each talker and object against chance (where chance = a difference score of
0). As predicted, following labeling by the Training Speaker, looking significantly increased to the
untrained object (t(29) = 2.594, p= .015). In contrast, for the Extension Speaker, looking to the
trained object significantly increased (t(29) = 2.700, p= .011).
4
Thus, when the Training Speaker said
Figure 2. Schematic of a Test Trial: An image of the speaker appears alone on the screen for two seconds, followed by images of
the trained and untrained object on either side of the screen. Objects are onscreen for 3 seconds before the test word is uttered
(baseline period) and remain on screen for another 4 seconds post-label onset.
2
Note that these proportions are out of the total duration of each phase. Thus, the proportions for each object in a trial do not
necessarily sum to 1. In fact, while infants spent approximately 87% of the total time looking at the objects during the baseline
phase, they spent approximately 95% of the total time looking at the objects during the test phase. The amount of time spent
looking at the screen in each phase was the same for both talkers.
3
All t-tests reported are two-tailed.
4
To ensure that participants with more extreme baseline asymmetries did not affect the overall pattern of results, we also re-
analyzed the data using a weighted difference score, in which trials with larger asymmetries carried less weight. To arrive at this
weighted difference score, we first determined the difference in baseline preference for each object (degree of bias) for each trial
(by participant). The actual difference scores were then multiplied by (1 - the degree of bias). Thus, the larger the bias score, the
less weight the score carried in the overall mean. The pattern of results remained the same (for the Training Speaker, looking
significantly increased to the untrained object t(29) = 2.783, p =.016 and for the Extension Speaker looking significantly
increased to the trained object t(29) = 3.042, p =.009). In addition to this baseline correction, we also re-analyzed the data by
including only trials that had less asymmetric baseline differences, equating baseline scores across the speakers for both objects.
We found the same pattern of results (for the Training Speaker, looking increased to the untrained object t(20) = 1.995, p =.059,
and for the Extension Speaker, looking increased to the trained object t(25) = 2.969, p =.006. Finally, note that although there
was an asymmetry in the baseline for the Training speaker in Experiment 1, this asymmetry was not present in Experiment 2,
where a significant increase in looking to the untrained object was also found.
LANGUAGE LEARNING AND DEVELOPMENT 97
Downloaded by [University of Waterloo] at 09:53 19 May 2016
tIpu, infants increased their looking toward the untrained object, but when the Extension Speaker
said tIpu, they increased their looking toward the trained object. In other words, infants responded
differently to the same test word, depending on which talker produced it. Note from Table 2 that
both talkers’pronunciations of tIpu were distinct from the training word tEpu.
This pattern of results suggests that infants learned the Training Speaker’s label for the training
object (tEpu) and when the same talker used a different label (tIpu), they interpreted it as a label for
the untrained object. The fact that, in contrast, infants increased their looking to the trained object
for the Extension Speaker suggests that they tracked the differences between the two talkers’
pronunciations during the exposure phase.
However, closer analysis revealed that this wasonlytrueofinfantswhowerefirsttestedon
the Training Speaker. For this order, the ANOVA revealed no significant effect of Speaker (F
(1,14) = .0448, ns)orObject(F(1,14) = .202, ns), but a significant Speaker x Object interaction
(F(1,14) = 10.705, p= .006). One-sample t-tests showed that for the Training speaker, there
was a significant increase in looking to the untrained object (t(14) = 2.654, p= .019) and
decreased looking to the trained object (t(14) = -1.101, p= .290). For the Extension speaker,
looking significantly increased to the trained object (t(14) = 2.918, p= .011) and decreased to
the untrained object (t(14) = -1.122, p= .281); see Figure 4). In contrast, those who were first
tested on the Extension Speaker did not reliably change their looking behavior at test. There
was no Speaker x Object order interaction (F(1,14) = 0.141, ns) and looking did not change for
either speaker individually (Training Speaker untrained object: t(14) = 0.846, ns; Training
Speaker trained object: t(14) = 0.358, ns; Extension Speaker untrained object: t(14) = 0.943, ns;
Extension Speaker trained object: t(14) = 0.076, ns).
Given this pattern of results, it is possible that infants did not learn the systematic differ-
ences between the talkers’accents but simply learned that the two talkers pronounced words
differently. If true, when the Training Speaker came first, infants could succeed by determining
her intended referent and then, for the Extension Speaker trial, looking at the other referent. In
the other direction, determining the Extension Speaker’s intended referent would have been
more difficult without the anchor provided by the Training Speaker, thus leading to poorer
overall performance. To investigate whether infants were using only this type of heuristic, a
Figure 3. Difference scores for all participants in Experiment 1: looking proportions during baseline subtracted from looking
proportions during the test period. Positive scores reflect an increase in looking while negative scores reflect a decrease in looking.
* denotes a pvalue less than 0.05. Error bars represent the calculated standard error.
98 D. WEATHERHEAD AND K. S. WHITE
Downloaded by [University of Waterloo] at 09:53 19 May 2016
second experiment was run, in which the expected response was increased looking to the same
object for both talkers.
Experiment 2
In order to determine if the participants in Experiment 1 were tracking the systematic differences
between the talkers’accents, Experiment 2 used a test word, topu, that did not fall into the pattern
learned during exposure. Recall that the difference between the talkers involved the height of front
vowels; critically, the pronunciation of back vowels remained constant. If infants in Experiment 1
simply learned to respond to the two talkers differently, then infants in Experiment 2 should also look
at different objects for the two talkers, even if the test word contains only back vowels. If, however,
infants in Experiment 1 learned that the accent difference was specific to front vowels, then, regardless
of the talker, infants in Experiment 2 should map the word topu to the untrained object.
Participants
Forty 10- to 12-month olds (21 females and 19 males, mean age = 334.85 days, age range = 311-363
days) took part. An additional ten infants were tested but were not included due to fussiness (1), failure
to attend to both objects during familiarization (2) or the screen for the entirety of the test period (6),
or a difference score exceeding 2.5 standard deviations from the mean of either speaker (1).
Stimuli and Procedure
Identical to Experiment 1, except the label t[o]pu was used during the test phase.
Results
Figure 5 displays the difference scores.
5
During the baseline period, there was no significant
difference in looking to the trained versus untrained object for either speaker (proportions of 0.46
and 0.40
6
, respectively, for the Training Speaker: t(39) = 1.459, p= .15; proportions of 0.48 and 0.43,
respectively, for the Extension Speaker: t(39) = 1.191, p= .24). A Repeated-Measures ANOVA was
conducted on the test-baseline difference scores with within-subjects factors of Speaker and
Figure 4. Experiment 1 difference scores for participants who saw the Training Speaker first at test (a), and for participants who
saw the Extension Speaker first at test (b). * denotes a pvalue less than 0.05. Error bars represent the calculated standard error.
5
Infants spent approximately 89% of the total time looking at the objects during the baseline phase, and approximately 92% of
the total time looking at the objects during the test phase.
6
This degree of baseline difference is on the order of those often found when a label for one object is known and for the other
object is unknown (Schafer, Plunkett, & Harris, 1999; White & Morgan, 2008). In such studies, there are still reliable effects of the
type of label (familiar/novel) on looking behavior.
LANGUAGE LEARNING AND DEVELOPMENT 99
Downloaded by [University of Waterloo] at 09:53 19 May 2016
Object and a between-subjects factor of test Order. The ANOVA revealed a significant effect of
Speaker (F(1,39) = 6.123, p= .018) and a significant interaction between Object and Order (F(1,39) =
6.018, p= .019). No other effects reached significance.
Given the interaction involving Order, analyses were conducted for each of the presentation orders
separately. For the participants who saw the Training Speaker first at test, a repeated-measures
ANOVA found significant main effects of Speaker (F(1,19) = 5.009, p= .037) and Object (F(1,19) =
5.242, p= .034), but no interaction (F(1,19) = 1.365, p= .257). One sample t-tests against 0 showed
that, as predicted, infants increased their looking to the untrained object for both the Training Speaker
(t(19) = 2.141, p= .045) and Extension Speaker (t(19) = 1.951, p= .066), and decreased their looking to
the trained object for both speakers (Training Speaker t(19) = -.539, ns; Extension Speaker t(19) =
-2.223, p=.039)(seeFigure 6). However, for participants who saw the Extension Speaker first, the
ANOVA revealed no significant effects (Speaker: F(1,19) = 2.408, p= .137; Object: F(1,19) = 1.414, p=
.249; interaction: F(1,19) = .172, ns). One sample t-tests showed that, unexpectedly, looking increased
to the trained object for the Training Speaker (t(19) = 2.125, p= .047); there was no change for the
untrained object (t(19) = -.130, ns). For the Extension Speaker, there were no significant changes for
either object (trained: t(19) = 0.832, ns;untrained:t(19) = -0.674, ns).
Summarizing these results, infants increased their looking to the untrained object when they
heard either of the two talkers say the test word topu, but only if they saw the Training Speaker first
Figure 5. Experiment 2 difference scores for all participants. Error bars represent the calculated standard error.
Figure 6. Experiment 2 difference scores for participants who saw the Training Speaker first at test (a), and for participants who
saw the Extension Speaker first at test (b). * denotes a pvalue less than 0.05, —denotes a pvalue less than 0.10. Error bars
represent the calculated standard error.
100 D. WEATHERHEAD AND K. S. WHITE
Downloaded by [University of Waterloo] at 09:53 19 May 2016
during the test phase. This suggests that infants in Experiment 1 did not learn only that the two
talkers pronounced words differently. If they had, infants would have looked at different objects for
each of the talkers in Experiment 2 as well. Therefore, infants must have encoded something more
specific about the accent differences. We discuss the implications of these findings below.
Discussion
If infants cannot recognize the equivalence of words that are realized differently due to cross-speaker
variation, they risk positing spurious word-referent mappings that could slow lexical development.
We explored whether 10- to 12-month-olds could overcome the effects of talker-specific variation if
given the chance to determine the relationship between the talkers’accents. Infants were first
exposed to talkers whose front vowels differed. At test, they were presented with a previously
unheard wordform, either tIpu (Experiment 1) or topu (Experiment 2). We predicted that, if infants
were able learn the systematic vowel differences between the talkers and use this talker-specific
information to make inferences about intended referents, their interpretation of tIpu should differ by
speaker, but their interpretation of topu should not. Experiment 1 demonstrated that infants mapped
tIpu to the untrained object for the Training Speaker, but to the trained object for the Extension
Speaker. Experiment 2 ruled out the possibility that infants learned only a heuristic that the two
talkers spoke differently: infants looked longer at the untrained object when both talkers produced
the label topu, at least when the Training Speaker was presented first. Thus, infants appear to have
learned that the difference between the talkers was specific to front vowels. The finding that infants
learned about the relationship between the two talkers’productions is consistent with the fact that
older toddlers can learn about the properties of accents (Schmale et al., 2012; Van Heugten &
Johnson, 2014; White & Aslin, 2011). In those studies, toddlers learned the relationship between a
novel accent and their own. The present work not only extends this ability to younger infants, but
also shows that they can learn a phonetic relationship between two novel talkers that does not
involve comparison to their own accent. This ability to track talker-specific detail parallels adults’
learning of talker-specific properties for multiple speakers (such as voice-onset-time: Allen & Miller,
2004).
7
Infants interpreted novel wordforms as a function of what they had learned about each talker’s
speech. This is consistent with work in other domains demonstrating that infants make person-
specific attributions about certain types of information (e.g., desires, Repacholi & Gopnik, 1997;
action goals, Buresh & Woodward, 2007) and can use person-specific information to guide their
interactions with an individual (e.g., the person’s reliability, Chow, Poulin-Dubois, & Lewis, 2008;
helping and hindering behavior, Hamlin, Wynn, & Bloom, 2007; global aspects of the person’s
accent,Kinzler,Dupoux,&Spelke,2007). The present results suggest that infants can also link
subtle phonetic properties to particular individuals and use that information alone to infer a
talker’sintendedreferent.
In both experiments, infants succeeded only when they saw the Training Speaker first at test. This
suggests that they were using their knowledge of the Training Speaker’s productions to guide their
behavior for the Extension Speaker. However, despite the order effects, infants’pattern of looking
differed between the two experiments, demonstrating that infants were responding to the specifics of
the label in each experiment. The fact that infants succeeded at all in our task is noteworthy. The task
imposed a high processing load on our young participants—in order to succeed, they had to not only
detect and encode the relationship between the talkers’productions, but also learn a new word in the
lab. Even the latter task alone is challenging at this age; only a handful of lab studies have found
7
As pointed out by an anonymous reviewer, an alternative possibility is that infants misattributed the accent difference to
voice (treating [E] and [I] as these speakers’pronunciations of the same vowel, that is, as a within-category difference).
However, under such an interpretation, it is not clear why infants would show differential treatment of ‘tIpu’in Experiment 1,
as the tokens of /I/ are acoustically similar for the two speakers. That said, determining how infants attribute variability to
different sources is an important question for future research.
LANGUAGE LEARNING AND DEVELOPMENT 101
Downloaded by [University of Waterloo] at 09:53 19 May 2016
word learning in this age group from a single talker (e.g., Gogate, 2010; Shukla et al., 2011). Further,
the correct interpretation in some trials was not the mapping of the trained word, but instead
required that the learned mapping be used to map a novel label to the novel object. Thus, our results
also demonstrate precocious use of an exclusion-based or novelty-novelty mapping strategy. In
future work, we plan to further explore developmental changes in this task (e.g., at what point
infants’representations are robust enough to succeed in the opposite order).
In summary, a large body of research has demonstrated that word learners have a strong bias to
map novel labels to novel objects. In the present work, we find that even 10-12-month-olds do so
when labels come from a single talker, but that they do not when the labels come from talkers who
have different accents. Our results also suggest that, like adults, infants can track talker-specific
phonetic properties, and can use information about an individual’s previous language history to
guide their future interactions with that individual. These findings suggest that, from a young age,
infants are equipped with the tools necessary to handle the variable input around them.
Acknowledgments
The authors would like to thank the members of the Lab for Infant Development and Language for
help with participant recruitment, Eiling Yee and Mohinish Shukla for helpful discussion, and all of
the families and infants who participated. This work was funded by an operating grant from the
Natural Sciences and Engineering Research Council of Canada.
References
Allen, J. S., & Miller, J. L. (2004). Listener sensitivity to individual talker differences in voice-onset-time. Journal of the
Acoustical Society of America,115, 3171–3183.
Au, T. K. F., & Glusman, M. (1990). The principle of mutual exclusivity in word learning: to honor or not to honor?
Child Development,61(5), 1474–1490.
Bailey, T. M., & Plunkett, K. (2002). Phonological specificity in early words. Cognitive Development,17(2), 1265–1282.
Best, C. T., Tyler, M. D., Gooding, T. N., Orlando, C. B., & Quann, C. A. (2009). Development of phonological
constancy: Toddlers’perception of native- and Jamaican-accented words. Psychological Science,20(5), 539–542.
doi:10.1111/j.1467-9280.2009.02327.x
Boersma, P., & Weenink, D. (2009). Praat: doing phonetics by computer (Version 5.1. 05) [Computer program].
Bradlow, A. R., & Bent, T. (2008). Perceptual adaptation to non-native speech. Cognition,106(2), 707–729.
doi:10.1016/j.cognition.2007.04.005
Buresh, J. S., & Woodward, A. L. (2007). Infants track action goals within and across agents. Cognition,104(2), 287–
314. doi:10.1016/j.cognition.2006.07.001
Chow, V., Poulin-Dubois, D., & Lewis, J. (2008). To see or not to see: Infants prefer to follow the gaze of a reliable
looker. Developmental Science,11(5), 761–770.
Clarke, C. M., & Garrett, M. F. (2004). Rapid adaptation to foreign-accented English. The Journal of the Acoustical
Society of America,116(6), 3647–3658.
Cohen, J. D., MacWhinney, B., Flatt, M., & Provost, J. (1993). PsyScope: A new graphic interactive environment for
designing psychology experiments. Behavioral Research Methods, Instruments, and Computers,25(2), 257–271.
Retrieved from http://psy.cns.sissa.it
Gogate, L. J. (2010). Learning of syllable–object relations by preverbal infants: The role of temporal synchrony and
syllable distinctiveness. Journal of Experimental Child Psychology,105(3), 178–197.
Golinkoff, R. M., Mervis, C. B., & Hirsh-Pasek, K. (1994). Early object labels: the case for a developmental lexical
principles framework. Journal of Child Language,21, 125–155.
Halberda, J. (2003). The development of a word-learning strategy. Cognition,87, B23–34.
Hay, J., Nolan, A., & Drager, K. (2006). From fush to feesh: Exemplar priming in speech perception. The Linguistic
Review,23(3), 351–379.
Houston, D. M., & Jusczyk, P. W. (2000). The role of talker specific information in word segmentation by infants.
Journal of Experimental Psychology: Human Perception and Performance,26, 1570–1582.
Kinzler, K. D., Dupoux, E., & Spelke, E. S. (2007). The native language of social cognition. Proceedings of the National
Academy of Sciences of the United States of America,104(30), 12577–12580. doi:10.1073/pnas.0705345104
Mani, N., & Plunkett, K. (2008). Fourteen-month-olds pay attention to vowels in novel words. Developmental Science,
11,53–59.
102 D. WEATHERHEAD AND K. S. WHITE
Downloaded by [University of Waterloo] at 09:53 19 May 2016
Mani, N., & Plunkett, K. (2011). Does size matter? Subsegmental cues to vowel mispronunciation detection. Journal of
Child Language,38, 606–627.
Mani, N., Mills, D. L., & Plunkett, K. (2012). Vowels in early words: an event-related potential study. Developmental
Science,15,2–11.
Markman, E. M. (1989). Categorization and naming in children: Problems of induction. Cambridge, MA: MIT Press.
Markman, E. M. (1990). Constraints children place on word meanings. Cognitive Science,14(1), 57–77.
Merriman, W. E., Bowman, L. L., MacWhinney, B. (1989). The mutual exclusivity bias in children’s word learning.
Monographs of the Society for Research in Child Development,54(3/4), 1–129.
Merriman, W. E., & Schuster, J. M. (1991). Young children’s disambiguation of object name reference. Child
Development,62(6), 1288–1301.
Mulak, K. E., Best, C. T., Tyler, M. D., Kitamura, C., & Irwin, J. R. (2013). Development of phonological constancy: 19-
month-olds, but not 15-month-olds, identify words in a non-native regional accent. Child Development,84(6),
2064–2078. doi:10.1111/cdev.12087
Kuhl, P. K. (1983). Perception of auditory equivalence classes for speech in early infancy. Infant Behavior and
Development,6(2), 263–285.
Nathan, L., Wells, B., & Donlan, C. (1998). Children’s comprehension of unfamiliar regional accents: A preliminary
investigation. Journal of Child Language,25, 343–365.
Repacholi, B. M., & Gopnik, A. (1997). Early reasoning about desires: evidence from 14-and 18-month-olds.
Developmental Psychology,33(1), 12–21.
Schafer, G., Plunkett, K., & Harris, P. L. (1999). What’s in a name? Lexical knowledge drives infants’visual preferences
in the absence of referential input. Developmental Science,2, 187–194.
Schmale, R., & Seidl, A. (2009). Accommodating variability in voice and foreign accent: Flexibility of early word
representations. Developmental Science,12(4), 583–601.
Schmale, R., Cristia, A., Seidl, A., & Johnson, E. K. (2010). Developmental changes in infants’ability to cope with
dialect variation in word recognition. Infancy,15(6), 650–662.
Schmale, R., Hollich, G., & Seidl, A. (2011). Contending with foreign accent in early word learning. Journal of Child
Language,38(5), 1096–1108.
Schmale, R., Cristia, A., & Seidl, A. (2012). Toddlers recognize words in an unfamiliar accent after brief exposure.
Developmental Science,15(6), 732–738.
Shukla, M., White, K. S., & Aslin, R. N. (2011). Prosody guides the rapid mapping of auditory word forms onto visual
objects in 6-mo-old infants. Proceedings of the National Academy of Sciences of the United States of America,108
(15), 6038–6043.
Singh, L., White, K. S., & Morgan, J. L. (2008). Building a word-form lexicon in the face of variable input: Influences of
pitch and amplitude on early spoken word recognition. Language Learning and Development,4, 157–178.
Sundara, M., & Scutellaro, A. (2011). Rhythmic distance between languages affects the development of speech
perception in bilingual infants. Journal of Phonetics,39(4), 505–513.
Swingley, D., & Aslin, R. N. (2002). Lexical neighborhoods and the word-form representations of 14-month-olds.
Psychological Science,13(5), 480–484.
Swoboda, P. J., Morse, P. A., & Leavitt, L. A. (1976). Continuous vowel discrimination in normal and at risk infants.
Child Development, 459–465.
Uttley, L., de Boisferon, A. H., Dupierrix, E., Lee, K., Quinn, P. C., Slater, A. M., & Pascalis, O. (2013). Six-month-old
infants match other-race faces with a non-native language. International Journal of Behavioral Development,37(2),
84–89. doi:10.1177/0165025412467583
Van Heugten, M., & Johnson, E. K. (2014). Learning to contend with accents in infancy: Benefits of brief speaker
exposure. Journal of Experimental Psychology: General,143, 340–350.
Werker, J. F., & Tees, R. C. (1984). Cross-language speech perception: Evidence for perceptual reorganization during
the first year of life. Infant Behavior and Development,7,49–63.
White, K. S., & Morgan, J. L. (2008). Sub-segmental detail in early lexical representations. Journal of Memory and
Language,59(1), 114–132.
White, K. S., & Aslin, R. N. (2011). Adaptation to novel accents by toddlers. Developmental Science,14(2), 372–384.
doi:10.1111/j.1467-7687.2010.00986.x
LANGUAGE LEARNING AND DEVELOPMENT 103
Downloaded by [University of Waterloo] at 09:53 19 May 2016
