when word learning heuristics meet cross-situational word ... · ii when word learning heuristics...
TRANSCRIPT
When Word Learning Heuristics Meet Cross-situational Word Learning: A Comparison Between Monolingual and
Bilingual Toddlers
by
Priscilla Pui Yee Fung
A thesis submitted in conformity with the requirements for the degree of Master of Arts
Graduate Department of Psychology University of Toronto
© Copyright by Priscilla P.Y. Fung (2019)
ii
When Word Learning Heuristics Meet Cross-situational Word
Learning: A Comparison Between Monolingual and Bilingual
Toddlers
Priscilla P.Y. Fung
Master of Arts
Graduate Department of Psychology
University of Toronto
2019
Abstract
Monolingual children reportedly rely more heavily on the Mutual Exclusivity Principle (MEP)
than bilinguals in word learning. However, past work has only examined toddlers’ use of MEP
after a single labelling. Here, we examine monolingual and bilingual 2-year-olds’ reliance on the
MEP to learn novel colour labels. Importantly, cross-situational statistics linked a particular label
to a particular colour – but not to a particular object. As a secondary question, we asked whether
the fact that past studies have presented children with atypically-coloured objects (e.g., turquoise
dogs) may have influenced how readily children attached novel labels to colour terms rather than
objects. Preliminary results suggest that monolinguals and bilinguals were equally successful in
learning novel colour labels, but learning only occurred when the objects were atypically-
coloured. We conclude that researchers need to carefully consider the richness of children’s
home learning environments to better understand development in diverse language settings.
iii
Acknowledgments
I would like to thank my advisor, Elizabeth Johnson, for her continuous support and guidance.
Thank you for letting me to explore my ideas. This project will not be possible without your
insights and expertise. I would also like to thank Craig Chambers for his constructive feedback
and helpful comments on the manuscript. A special thanks to Thomas St. Pierre for your help
with data analysis and technical assistance (i.e. with lab equipment).
I would like to extend my gratitude to the Social Sciences and Humanities Research Council of
Canada for their financial contribution to my Master’s thesis. It is challenging to complete this
project within a year. To all the research assistants and our lab manager Lisa Hotson, thank you
for making this happen!
I am grateful to have met such amazing friends in graduate school, Madeleine Yu, Raheleh
Saryazdi, Kay Otsubo, Emma Galarneau, and Prateek Dhamija. Thank you so much for all the
moral support along the way and helping me survive this challenging year. Last, but not least, I
would like to thank my family especially my grandparents. I would not be where I am now
without your love and care!
iv
Table of Contents
Acknowledgments………………………………………………………………………………..iii
Table of Contents ........................................................................................................................... iv
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Literature Review ........................................................................................................................1
1.1 Introduction ..........................................................................................................................1
1.2 Word Learning Heuristics in Monolingual Children ...........................................................1
1.3 Word Learning Heuristics in Bilingual Children .................................................................2
1.4 Cross-Situational Word Learning ........................................................................................3
1.5 Present Study .......................................................................................................................5
Experiment 1 ...............................................................................................................................6
2.1 Introduction ..........................................................................................................................6
2.2 Method .................................................................................................................................7
2.2.1 Participants ...............................................................................................................7
2.2.2 Stimuli and Design ...................................................................................................7
2.2.3 Procedure ...............................................................................................................10
2.2.4 Coding ....................................................................................................................12
2.3 Results and Discussion ......................................................................................................12
Experiment 2 .............................................................................................................................14
3.1 Introduction ........................................................................................................................14
3.2 Method ...............................................................................................................................15
3.2.1 Participants .............................................................................................................15
3.2.2 Stimuli and Design .................................................................................................15
3.2.3 Procedure ...............................................................................................................15
3.2.4 Coding ....................................................................................................................15
v
3.3 Preliminary Results and Discussion...................................................................................17
General Discussion ...................................................................................................................20
References ......................................................................................................................................25
vi
List of Tables
Table 1. Animate and inanimate objects used in Experiment 1 and 2 ……………………............9
Table 2. Percentage of English exposure and other language(s) as reported by caregivers of
bilingual children ……………………..........……………………..........…………..………........16
vii
List of Figures
Figure 1. Sample stimuli used in the training trials for Animate Block …………………….......11
Figure 2. Sample stimuli used in the training trials for Inanimate Block ………………….........11
Figure 3. Sample stimuli used in test trials for Animate Block ……………………....................12
Figure 4. Sample stimuli used in test trials for Inanimate Block ……………………..................12
Figure 5. Windows of analysis used in calculating difference scores …………………..............13
Figure 6. Mean difference scores in Animate Block and Inanimate Block after 6 and 12 training
trials in monolingual children ……………..........……………………..........…………..……….14
Figure 7. Mean difference scores in Animate Block and Inanimate Block after 6 and 12 training
trials in bilingual children ……………..........……………………..........…………..…………...18
Figure 8. Comparison of performance after 12 training trials in monolingual and bilingual
children ……………..........……………………..........…………..…………...............................19
Figure 9. Correlation between CDI scores and performance in Animate Block after 12 training
trials in both monolingual and bilingual children ……………………..........…………………...20
1
Literature Review
1.1 Introduction
During the second year of life, toddlers can learn new words at a remarkable rate. They do this
despite the fact that there are an infinite number of possible referents in the environment for
every label. There are two well-studied word learning heuristics that children use to reduce
referential ambiguity, namely the Mutual Exclusivity Principle (MEP) and Whole-Object
Assumption. However, our understanding of these word learning strategies has been based
almost entirely on work with monolingual children (e.g., Hall, Waxman, & Hurwitz, 1993;
Liittschwager & Markman, 1994; Markman, 1990). Given that half of the world’s population is
multilingual (e.g., U. S. Census Bureau, 2010), basing our theories of early word learning on
work solely with monolinguals is risky because those theories may not generalize to multilingual
children. Indeed, recent work has suggested that bilingual children may rely much less on the
MEP than monolingual children (e.g. Davidson & Tell, 2005; Houston-Price, Caloghiris &
Raviglione, 2010). And since strong reliance on the MEP has played a major part in classic
approaches to explain how children learn second-order attributes such as colour, this begs the
question of how bilinguals learn adjectives like colour names. Moreover, other well-established
word learning strategies, such as the tracking of cross-situational statistics (e.g., Smith & Yu,
2008; Suanda, Mugwanya, & Namy, 2014; Yu & Smith, 2007), have never been explored in
bilingual toddlers. In this thesis, we examine monolingual and bilingual two-year-olds’ use of the
MEP and cross-situational statistics to learn that a novel label refers to a colour (but not to the
object displaying the colour). As a secondary question, we look at the attachment of colour
attributes to animate versus inanimate objects. We aim to not only compare colour label learning
in monolingual versus bilingual children, but also to begin to better understand how in general,
the design of laboratory experiments can impact children’s reliance on different types of word
learning strategies.
1.2 Word Learning Heuristics in Monolingual Children
Although monolingual and bilingual infants experience a different language environment, they
both face the same challenge of having to determine word meanings from the continuous speech
2
stream that they are hearing. It has been suggested that children rely on word learning heuristics
to narrow the search space for possible word-referent mappings (Clark, 1988; Markman and
Wachtel, 1988; Markman, 1994; Landau et al., 1998). Children use the Whole-Object
Assumption to predict a novel label refers to an object as a whole but not to its parts, colour, or
other properties (e.g., Markman & Wachtel, 1988; Soja, Carey, & Spelke, 1991). For example,
upon hearing the word “rabbit”, children will assume the word refers to the whole rabbit that
they are seeing, rather than its ears or the colour white. In a monolingual environment, words and
objects largely follow a one-to-one mapping, with most object kinds only have one label.
Therefore, it has been argued that it is adaptive for monolingual children to follow the MEP to
assume one object has only one basic-level label. This helps them to override the Whole-Object
Assumption and consider the possibility that the novel label refers to properties of the object, for
example “white” refers to the colour of the rabbit instead of the whole rabbit.
1.3 Word Learning Heuristics in Bilingual Children
More recent studies have started to extend the study of word learning biases to multilingual
children. These children differ from monolinguals because they routinely encounter situations
that violate the MEP (e.g. dog in English and chien in French). Therefore, compared to
monolinguals, bilingual children might be less likely to use MEP in acquiring word meanings
because they might perceive it as less useful. Indeed, studies have consistently shown that
bilingual children rely less MEP than monolinguals (e.g. Davidson & Tell, 2005; Houston-Price,
Caloghiris & Raviglione, 2010). Particularly, Byers-Heinlein and Werker (2009) found that, at
18 months, monolinguals showed the strongest use, bilinguals showed marginal use, and
trilinguals showed no use at all. Kandhadai, Hall, and Werker (2017) further asked whether
bilinguals were simply confused in these novel word learning situations or in fact they were able
to interpret the novel word systematically as a second object category label for the familiar
object. In their experiment, 18-month-old monolinguals and bilinguals heard a novel label for a
familiar object that had a salient colour (e.g. an aqua-coloured dog). They were then tested
whether they interpreted the novel label as a second category label for the object (e.g. another
label for dog) or as a label for its salient property (e.g. aqua). They found that bilinguals did not
rely on MEP; instead, they systematically interpreted the novel word as a second label for the
familiar object. Monolinguals, on the contrary, rejected the novel word as a second label and
3
showed a tendency to interpret it as a property term for the familiar object (i.e. colour label).
They concluded that bilinguals have clear expectations that guide their mappings even though
they do not adhere to MEP. A recent study by Byers-Heinlein (2017) explored the question of
whether monolingual and bilingual children initially follow similar word learning heuristics and
then diverge later in development after they have acquired more knowledge of word-object
relations, or, in fact, whether bilingual language experience changes their expectations about
how words refer to object kinds early in life. She examined the use of MEP in 9- to 10-month-
old monolingual and bilingual infants and found that they did not share the same expectation
even before they had acquired a sizable vocabulary. Taken together, these studies show that
language experience affects how strictly children follow the MEP from early lexical
development. While these findings are interesting, they raise more questions than they answer –
if bilingual children are less reliant on MEP, then how do they overcome the Whole-object
Assumption and learn labels for object properties?
1.4 Cross-Situational Word Learning
Both monolingual and bilingual children can clearly learn labels for object properties such as
colour names. Note that this does not conflict with the claim that bilinguals are less reliant on
MEP than monolinguals because MEP is not the only way for them to learn adjectives. In the
real world, there is much more information available in the environment to help children
determine word meanings than in a typical lab setting. For example, in the studies outlined
above, children’s tendency to use MEP to resolve ambiguity in word-referent mappings was only
tested once after a single object was labelled. However, there are typically many words and
many potential referents present in the same environment, where word-to-object pairing is not
always as apparent as the experimental settings. For example, when a mother asks her child
during dinner, “Do you want more water?” On the table there are food, water, plates, forks, and
napkins. How can the child identify the correct referent for water among all the irrelevant ones?
One way for children to learn this is by keeping track the statistical evidence of which particular
words and objects co-occurred across multiple situations. For example, on the first night, the
child may have used word learning heuristics to identify three potential referents for “water” –
plates, water, napkins. On the second night, when the mother asks the same question again, the
child may identify another three potential referents – water, apple, forks. Because water is the
4
only object common in these two instances, the child will then be able to identify it as the correct
referent even when the word-object pairing is ambiguous in each individual instance. Indeed, a
growing body of literature suggests that both infants and adults are able to learn word-to-object
mappings over multiple instances by tracking statistical evidence (e.g., Smith & Yu, 2008;
Suanda, Mugwanya, & Namy, 2014; Yu & Smith, 2007), which is known as cross-situational
word learning. In a classic cross-situational word learning paradigm (e.g., Smith & Yu, 2008),
participants have to learn novel words through a series of individually ambiguous trials. Within
each trial, they will hear two or more novel words and see their potential referents in a random
order so that they will not know which word refers to which object. However, since the novel
words and their referents consistently co-occurred across trials, they would be able to figure out
the word-object pairings by accumulating and evaluating the statistical information over multiple
trials. It has been shown that both 12- and 14-month-olds are sensitive to these cross-situational
co-occurrence patterns and are capable to use this knowledge to acquire new word-to-object
mappings (Smith & Yu, 2008). This finding indicates that, even under referential uncertainty,
children can attach some information about potential referents to the novel words that they heard.
They can then retrieve and update this information across trials to identify the correct word-
object pairings.
Apart from object labels, it has also been found that toddlers can use cross-situational
consistency to learn verbs and adjectives. In Scott and Fisher (2012), the authors showed 2.5-
year-olds two novel action events simultaneously along with two novel verbs (e.g., Look, she is
pimming. And she is nading.). Each action had multiple tokens performed by different actors.
The only source of information about the intended referent of each verb is the consistency
between the verb and its accompanied action. The authors found that children were able to
abstract across actions performed by multiple actors and learn the verb through statistical
consistency.
Compared to nouns and verbs, the acquisition of adjectives emerges later in development
(Fenson et al., 1991). One proposed reason is that adjectives refer to selected properties and
children need to selectively attend to those properties and segregate them from the whole object
(Smith, Gasser, & Sandhofer, 1997; Gasser & Smith, 1998). Akathar and Montague (1999)
examined whether 2-, 3-, and 4-year-olds could use cross-situational statistical consistency to
learn labels for shape and texture. In particular, they showed children novel objects that differed
5
systematically in shape and texture. Then they labelled one set of shape with the same novel
adjective in the shape condition but one set of texture in the texture condition (e.g., This is a
modi one). They found that all three groups of children were able to learn the novel adjective by
attending to the consistent element across naming contexts and extend the novel adjectives to
new exemplars.
To date, no study has examined cross-situational word learning in bilingual children. However,
recent studies from monolingual and bilingual adults have provided some initial evidence that
bilinguals might be better than monolinguals at this task (Chen, Gershkoff-Stowe, Wu, Cheung,
& Yu, 2017; Escudero, Mulak, Fu, & Singh, 2016). It has been argued that the advantage for
bilinguals stems from their constant need to extract patterns from complex dual language input
and, as a result, that bilinguals are more capable in tracking multiple regularities and structures
simultaneously. This is evidenced in Antovich and Graf Estes (2017), where the authors
presented 14-month-old monolinguals and bilinguals with two interleaved speech streams in
artificial languages, which mimics code-switching in bilingual speech. They found that
bilinguals, but not monolinguals, were able to learn the structure of the two interleaved
languages and segment words from the speech stream using transitional probabilities alone.
Similarly, Kovács and Mehler (2009) found that 12-month-old bilinguals were able to learn and
generalize the mutually inconsistent regularities of two structures that were presented
simultaneously, whereas their monolingual peers learned only one of them. These results reveal
that the cognitive systems of bilingual infants are adapted to track multiple regularities in their
environment, and thus they are more flexible at learning speech structures.
1.5 Present Study
The present study sought to examine monolingual (Experiment 1) and bilingual children’s
(Experiment 2) use of the MEP in learning labels for object properties over the course of
multiple labelling instances. In Experiment 1, we aim to replicate earlier work demonstrating that
monolingual children can learn object property labels such as colour names with the help of
MEP. As a secondary question, we examine whether the learnability of colour names through the
MEP might be conditioned by factors such as colour typicality effects. We include this question
because past studies examining the use of the MEP to learn colour labels have used colour
pairings not typically seen in nature (e.g., turquoise dogs and purple elephants). These unusual
6
pairings may have drawn children’s attention to colour information in a way that facilitated the
learning of a novel label as referring to the colour. As a result, children’s reliance on the MEP
might have been overestimated if the relationship between the colour and the coloured object is
less salient (e.g., turquoise mittens). We therefore compare the ease of learning if the colour is
typical for the object or not. In Experiment 2, we use the same materials and methodology to
examine how bilingual children use cross-situational statistics to overcome the Whole-Object
Assumption and learn colour labels.
In both of these experiments, the word-referent pairing in each individual trial was ambiguous
such that the novel word could either be interpreted as a colour label or as an object label.
However, the pairings were consistent over the trials so that a child could in principle deduce
that the novel word referred to a colour label. We asked whether monolingual and bilingual
children could learn the intended referent of the novel label by tracking consistently co-occurring
pairings across trials, and whether there would be a difference if the relationship between the
colour and coloured object is salient or not. All of the objects were in a colour that does not map
onto any primary colour (i.e., non-focal colour) because children tend to learn primary colours
first. We manipulated the saliency by dividing our stimuli into animate and inanimate objects.
Although both types of objects have the same colours, it would be more salient in animate than
in inanimate objects because these colours are atypical in animate objects (e.g., a turquoise dog)
but possible in inanimate objects (e.g., turquoise mittens). We test 22- to 26-month-olds in the
current study because at this age, most toddlers (1) have acquired productive vocabularies of at
least 50 words, (2) have begun to map words to object properties, and (3) are more likely to
know the noun labels for all the objects used in the experiment.
Experiment 1
2.1 Introduction
In Experiment 1, we examine whether monolingual children can learn a novel word as a colour
label by tracking cross-situational statistics. Specifically, 2-year-olds heard a novel word (e.g.,
wug) and saw a familiar object in a non-focal colour in each trial, such that the novel word could
either be interpreted as a second label for the object or as a colour label. In one block of trials,
children saw pairings that were atypical in nature (e.g., an animate object such as a dog paired
7
with an unlikely color such as turquoise). In the other block of trials, children saw pairings that
were less atypical (e.g., an inanimate object such as a mitten paired with a possible color such as
turquoise). Children were presented with four novel words over the trials: two for animate
objects (the Animate Block, always with atypical colors) and two for inanimate objects (the
Inanimate Block, always with possible colors). In each block, the two novel words (e.g., wug;
teek) were paired with different familiar objects but of the same colour (e.g., magenta-coloured
cat, magenta-coloured turtle; coral-coloured bed, coral-coloured shirt). Note that although the
word-referent pairing was ambiguous in each of these individual trials, the consistency of the
pairing across trials would in principle allow the novel word to be reliably paired with one of the
two colours. To test children’s understanding on the trained novel words, we pitted the two
colours against each other as children heard the label (e.g. magenta-coloured elephant, coral-
coloured elephant). If they successfully learned the novel word (e.g. wug) as a colour label, they
should look longer at the object that was matched in colour (e.g. magenta-coloured elephant).
We predicted 2-year-olds would learn colour labels for both animate and inanimate objects, but
more readily for animate objects because the atypical colours in animate objects are more likely
to draw their attention.
2.2 Method
2.2.1 Participants
Thirty-two 22- months to 26-month-olds (MAge = 701 days, range = 663−804 days; 14 females)
monolingual English-learning children from middle- to upper middle-class families in the
Greater Toronto Area were tested. All monolinguals received at least 90% English language
input (M = 97% English) and English vocabulary size was measured by parental report (CDI
percentile score: M = 47.8%). Fifteen additional children were tested but were excluded from the
study prior to coding due to fussiness (12), parental influence (1), or reportedly not being
familiar with the names of all the objects used in the study (2).
2.2.2 Stimuli and Design
The experiment was divided into two blocks. One block consisted of animate objects and one
block consisted of inanimate objects (see Table 1). The order of the two blocks were
8
counterbalanced across children. Children learned two novel words in each block, with a total of
four novel words − wug, teek, pog, and gaf. In the training trials, each of the novel words was
paired with different objects, but of the same colour − namely coral, magenta, turquoise, or
periwinkle. The word-colour pairings were counterbalanced across participants. Each colour was
labelled six times, with a total of 12 training trials in each block. We chose six labelling
instances for each colour because toddlers were able to learn a novel adjective after six pairings
in Scott and Fisher (2012). Trial order was pseudo-randomized, with the same colour appeared
no more than twice in a row. The visual stimulus consisted of an image in the centre presented
on a white background. This was accompanied by a recording in which the colour of the object
was labelled. Each novel word was preceded by an attention-getting speech sound (‘Look!’ or
‘Wow!’). All auditory stimuli were produced in an infant-directed manner by a female native
English speaker. To facilitate learning, all images were animated with a zoom effect that was
synchronized with the audio (Gogate, Bolzani, & Betancourt, 2006; Gogate, Maganti, & Bahrick,
2015).
In each test trial, two images were presented side-by-side on a white background along with an
audio labelling the colour (e.g., “Look, wug! Can you see it?”). The side which the object was on
was counterbalanced across children. Each word was tested twice – once after six training trials,
and once after an additional six training trials. The order of which word was tested first was
counterbalanced. We included test trials after both six and 12 training trials because we were
uncertain how quickly children would learn the labels. In this way, we could maximize the
likelihood of observing any potential differences in how the labels would be learned across
conditions.
9
Table 1. Animate and inanimate objects used in Experiment 1 and 2.
Animate Objects Inanimate Objects
Horse Bed
Frog Car
Pig Ball
Duck Mittens
Chicken Table
Giraffe Chair
Lion Hat
Mouse T-shirt
Rabbit Pants
Turtle Shoes
Monkey Balloon
Elephant Box
Bear Socks
Dog Book
Sheep Cup
Cat House
10
2.2.3 Procedure
The study was conducted in a sound-attenuated booth, where children watched the video on a TV
screen while sitting on their caregiver’s lap. Their caregivers wore headphones and listened to
masking music to prevent them from biasing their child’s responses. The masking music
consisted of continuous music mixed with multiple overlaid tracks of the speech stimuli used in
the study (Pinto, Fernald, McRoberts, & Cole, 1998). Caregivers were instructed not to point to
the screen unless it was to direct the children’s attention. The experiment started with the first six
training trials in Block 1. Each trial was 5 seconds long. A novel word was played twice, once at
2 s into the trial and once at 4 s into the trial. A familiar object in a non-focal colour was shown
on the screen. The novel words consistently occurred with the colour (see Figures 1 and 2). After
the first six training trials, children were presented with two test trials in which the two colours
used in the experiment were pitted against each other (see Figures 3 and 4). A 2 s flashing white
star on a black screen was inserted before each test trial to attract children’s attention to the
center of the screen. Each trial was 10 seconds long and each of the two labels was tested once.
Children were then presented with the next six training trials and two test trials in Block 1. After
that, they completed Block 2 with the same procedure. A 4-second animated clip was inserted
after each testing block in order to keep children engaged in the video. The entire procedure was
videotaped for offline coding.
At the end of the experiment, caregivers were asked to fill out a vocabulary questionnaire to
assess whether children could both understand and say the labels of all the objects used in the
experiment, either in English or in another language.
11
Figure 1. Sample stimuli used in the training trials for Animate Block. In this example, wug
refers to the colour periwinkle, whereas teek refers to the colour magenta.
Figure 2. Sample stimuli used in the training trials for Inanimate Block. In this example, pog
refers to the colour coral, whereas gaf refers to the colour turquoise.
12
Figure 3. Sample stimuli used in test trials for Animate Block.
Figure 4. Sample stimuli used in test trials for Inanimate Block.
2.2.4 Coding
Children’s eye movements were hand-coded frame-by-frame from silenced videos using
SuperCoder (Hollich, 2005). Each 33 ms frame was coded as a look to the left image, right
image, or away.
2.3 Results and Discussion
We first established children’s baseline looking preferences in the test trials before any labels
were provided. The baseline target proportion score was computed by dividing the time spent
13
looking at the target by the total time spent looking at both object before the label onset (between
0 ms to 1500 ms; see Figure 5). The effect of labelling was then calculated using a difference
score that compares looking to the target object before and after labelling. The proportion of
fixations to the target object after labelling was computed by dividing the number of fixations to
the target by the total time spent looking at both objects in the in a time window that began 500
ms following word onset and ended 3000 ms later.
Figure 5. Difference scores were calculated by subtracting the proportion of fixations to the
target object during the pre-labelling baseline window from the proportion of fixations to target
during the post-labelling window.
We predicted that, if children could learn the novel words as a colour label, the mean difference
score would be significantly higher than zero. Following Kandhadai, Hall, and Werker (2017),
we conducted one-sample t-tests for both animate and inanimate object test trials (see Figure 6).
For animate objects, there was no effect of labeling after six training trials (M = 0.01, SE = 0.05,
t(31) = 0.24, p = .41), but we found a positive difference score after 12 training trials, indicating
a significant increase in the proportion of looking to the colour-matched object (M = 0.11, SE =
0.04 , t(31) = 2.86, p = .004). For inanimate objects, however, the difference scores were not
significantly different from zero after either six (M = -0.11, SE = 0.06 , t(31) = - 1.83, p = .96) or
12 training trials (M = 0.01, SE = 0.05 , t(31) = 0.10, p = .46). Next, we conducted a two-sample
t-test to compare the performance in Animate Block and Inanimate Block. Overall, colour labels
in Animate Block (M = 0.06, SE = 0.03) were easier to learn than in Inanimate Block (M = -0.05,
SE = 0.04, t(117) = 2.18, p = .015).
14
These results only provide evidence for the learning of colour labels in the Animate Block; and
even then, evidence for learning was only apparent after 12 training trials. Thus, colour label
learning appeared to be a difficult task for monolinguals in our study.
Figure 6. Mean difference scores in Animate Block and Inanimate Block after 6 and 12 training
trials in monolingual children. Error bars indicate the standard error.
Experiment 2
3.1 Introduction
In Experiment 1, we showed that monolinguals can use learn colour labels in this paradigm, but
only when colour terms were used with animate objects and only after 12 (rather than just six)
training trials. Presumably, the animacy effect was driven by colour typicality. We return to this
issue of animacy and colour typicality in the General Discussion. In Experiment 2, we repeat
Experiment 1 with only one modification: we test bilingual children instead of monolingual
children. We anticipate two likely possible outcomes: (1) Bilingual children will be less
successful than monolinguals tested in Experiment 1 because bilinguals rely less on the MEP to
15
learn labels for object properties, or (2) Bilingual children will outperform monolinguals in view
of the evidence that adult bilinguals are better than monolinguals at cross-situational learning. Of
course, it is also possible that these two opposing effects will cancel each other out, leading to no
differences between the performance of monolinguals in Experiment 1 and bilinguals in
Experiment 2.
3.2 Method
3.2.1 Participants
Sixteen 22- months to 26-month-olds (MAge = 722 days, range = 670−765 days; 9 females)
bilingual children from middle- to upper middle-class families in the Greater Toronto Area were
tested. All bilinguals received a range of 30% to 70% exposure to English, and the rest from at
least one other language (M = 56% English; see Table 2). English vocabulary size was measured
by parental report (CDI percentile score: M = 32%). The differences between the vocabulary size
in bilingual and monolingual in Experiment 1 was marginally significant (Monolingual: M =
303, SE = 0.88; Bilingual: M = 190, SE = 1.12, t(28) = 1.82, p = .08), with bilingual had a lower
vocabulary score than monolingual. Five additional children were tested but were excluded from
the study prior to coding because of fussiness (4) or reportedly not being familiar with the names
of all the objects used in the study in (1).
3.2.2 Stimuli and Design
The stimuli were the same as in Experiment 1.
3.2.3 Procedure
The procedure was the same as in Experiment 1.
3.2.4 Coding
We used the same coding procedure as in Experiment 1.
16
Table 2. Percentage of English exposure and other language(s) as reported by caregivers of
bilingual children.
Children English Exposure Exposure to other language(s)
1 50 50% Urdu
2 60 30% Vietnamese, 10% Tagalog
3 40 50% Hindi, 10% Punjabi
4 70 15% Italian, 10% Polish, 5% Portuguese
5 70 30% Romanian
6 70 30% Cantonese
7 60 40% Cantonese
8 60 20% Portuguese, 20% Italian
9 40 60% Cantonese
10 40 40% Albanian, 20% French
11 60 40% Portuguese
12 60 20% Spanish, 20% Italian
13 60 40% Portuguese
14 60 40% Mandarin
15 50 50% Polish
16 35 45% Cantonese, 15% Mandarin, 5% Taiwanese
17
3.3 Preliminary Results and Discussion
The effect of labeling was again measured using difference scores, calculated in the same
manner as in Experiment 1. As in Experiment 1, we conducted one-sample t-tests for both
animate and inanimate object test trials (see Figure 7). For animate objects, learning did not
occur after six training trials (M = -0.06, SE = 0.07 , t(15) = - 0.77, p = .77) but only after 12
training trials, with a difference score significantly higher than zero (M = 0.19, SE = 0.05 , t(15)
= 1.90, p = .03). For inanimate objects, bilingual children did not learn the novel words as a
colour label after either six (M = -0.002, SE = 0.08 , t(15) = -0.04, p = .52) or 12 training trials
(M = 0.08, SE = 0.10 , t(15) = 0.80, p = .22). Next, we conducted a two-sample t-test to compare
the overall performance in Animate Block (M = 0.11, SE = 0.05) and Inanimate Block (M = 0.03,
SE = 0.04; t(58) = 1.20, p = .05). The results show that colour labels in Animate Block were
easier to learn than in Inanimate Block.
Our results revealed that bilingual children were only able to learn colour labels for animate
objects – and the task appeared to be difficult for bilinguals since they only learned the colour
label mappings after 12 training trials. Thus, based on this preliminary analysis, we fail to find
reliable evidence that monolinguals and bilinguals differ in this task, but we do find additional
support for our prediction that colour labels are easier to learn in animate than in inanimate
objects.
18
Figure 7. Mean difference scores in Animate Block and Inanimate Block after 6 and 12 training
trials in bilingual children. Error bars indicate the standard error.
Since learning only occurred after 12 training trials for both monolingual and bilingual children,
we compared performance in the monolinguals and bilinguals after 12 training trials (see Figure
8). Two-sample t-tests were conducted to compare their performance in animate object and
inanimate object test trials. Based on this preliminary data set with only 16 bilinguals tested,
bilingual and monolingual children’s performance was not significantly different for either
animate (t(19) = 0.71, p = .48) or inanimate objects (t(23) = 0.87, p = .39). However, the mean
difference scores of bilinguals are consistently higher than monolinguals, which suggests that
bilinguals might be better than monolinguals in learning colour labels using this paradigm. We
will be able to draw a firmer conclusion after collecting additional bilingual data.
19
Figure 8. Comparison of performance after 12 training trials in monolingual and bilingual
children.
In an exploratory analysis, we examined if the potential difference in monolingual and bilingual
children’s performance is due to their differences in language experience or vocabulary size. We
conducted a correlation analysis between their vocabulary size and performances in Animate
Block after 12 training trials by collapsing monolingual and bilingual children (see Figure 9).
Results of the Pearson correlation indicated that there was no significant association between
CDI scores and performance (r(45) = -0.03, p = .97). This suggests that the vocabulary size
differences between monolinguals and bilinguals do not seem to affect performance on this task.
However, one reason could be our final sample only include those who could both understand
and say all the object labels that we used in the experiment and so there was no difference in
their word knowledge on our stimuli.
20
Figure 9. Correlation between CDI scores and performance in Animate Block after 12 training
trials in both monolingual and bilingual children.
General Discussion
Children learn words remarkably fast, regardless of whether they are learning one language or
more than one language. But recent work has reported that monolingual and bilingual toddlers
differ in how heavily they rely on well-established word learning heuristics, like the MEP (e.g.
Davidson & Tell, 2005; Houston-Price, Caloghiris & Raviglione, 2010). In one study,
monolingual toddlers tended to interpret a novel word for a familiar object as a second order
label (e.g., colour names; Kandhadai, Hall, & Werker, 2017). Unlike monolinguals, bilinguals
appeared to more readily accept multiple labels for individual referents, leading them to fail to
map novel labels onto second order properties like an object’s colour. In the real world, however,
bilinguals can clearly learn labels for second order properties. In the current study, we examined
the acquisition of novel colour labels by monolingual and bilingual 2-year-olds. Our results
suggest that when provided with appropriate contextual information (i.e., cross-situational
21
statistics), all children were able to acquire novel colour labels. However, the task was not easy.
Evidence for learning was only observed after 12 (but not six) training trials, and only highly
salient atypical colour-object pairings (e.g. turquoise dogs and magenta turtles) triggered the
mapping of novel labels to colour properties in either group.
In line with our predictions, labels for colours were easier to learn in atypical colour-object
pairings (i.e., in animate objects) than in typical colour-object pairings (i.e., in inanimate objects)
for both monolingual and bilingual children. In fact, contrary to our prediction, neither
monolinguals nor bilinguals showed any evidence of learning colour labels in inanimate objects.
This suggests that colour typicality might play a role in how easily children could learn a colour
label. The atypical colours in animate objects might have drawn their attention more, and thus
facilitating a mapping between the adjective and the perceptual property. In contrast, the more
probable colour-object pairings in inanimate objects are less salient, it is possible that children at
this age might adhere to the Whole-Object Assumption if the perceptual property is not as
apparent. Another possibility is that children tend to pay more attention to objects with faces than
those without (Fantz, Fagan, & Miranda, 1975). Children might simply pay more attention
during the animate object block than the inanimate object block, leading to the higher
performance with animate objects. To tease apart whether the pattern is due to the colour
typicality or an attentional bias, we plan to run a follow-up study with colours that are probable
in both animate and inanimate objects, such as yellow or grey. If the advantage of animate
objects disappears, this would mean colour typicality influences how easy children can map a
novel adjective to a colour. An alternative explanation is that all animate objects we used in the
experiment belong to the same superordinate category – animal, whereas the inanimate objects
we used do not share a same superordinate category. A familiar inclusive label for the animate
objects might have facilitated the mapping between the novel adjective and the colour, and
subsequently allowed them to extend the adjective to another member within the same
superordinate category. However, we found this explanation to be unlikely because another line
of research has shown that toddlers and young children were only able to extend novel adjectives
to the test objects when all objects were drawn from the same basic level category, but failed to
do so if the objects were drawn from different basic level categories (e.g., Klibanoff & Waxman,
2000; Waxman & Markow, 1998), even when superordinate labels were explicitly provided to
22
them (Graham, Cameron & Welder, 2005). This suggests that having all objects in the same
superordinate category does not fully explain the learning advantage in animate objects.
Another important finding is that monolinguals did not perform better than bilinguals in this task;
they were only able to learn the colour labels in animate objects, and were only successful after
12 training trials. This implies monolingual children do not, as previous studies suggested,
reliably use MEP in learning labels for second order properties. Instead, the tendency to use
MEP might depend on the specific paradigm and specific properties being tested. Akathar and
Montague (1999) found that children could use cross-situational consistency to learn labels for
shape and texture, so why did children in our experiment not reliably learn colour labels? A
possible explanation comes from another body of literature that has repeatedly demonstrated
toddlers and young children attend to shape when classifying inanimate objects but instead
attend to both shape and texture when classifying animate objects because these dimensions are
critical to lexical category membership (e.g., Booth & Waxman, 2002; Jones & Smith, 2002;
Jones, Smith, and Landau, 1999) . Particularly, Graham and Poulin-Dubois (1999) found that 16-
to 20-month-olds generalized novel labels to same shape referents but not to the same colour
referents in both animate and inanimate objects. This means that children might bias their
attention more towards shape and texture than colour when searching for commonalities between
the objects used in the task, which helps them to more easily identify the potential referent as
shape or texture as opposed to colour.
Our results also demonstrate that both monolingual and bilingual children are able to use cross-
situational consistencies to learn second order property labels, but perhaps to a different degree.
Virtually all studies in the past used syntactic cues (e.g., This is a blick-ish one) to indicate the
adjectival status of the novel word in order to facilitate learning because infants as young as 14
months were found to be sensitive to these grammatical categories and are able to make relevant
inferences about meanings (Hall, Waxman, and Hurwitz, 1993). Therefore, since the
grammatical category of our novel words is ambiguous, and could either be interpreted as a noun
or an adjective, children, especially bilinguals, would need to rely on cross-situational
consistency in order to accurately map them to the object property. Based on our preliminary
results in Experiment 2, although it did not reach a significant level, bilinguals showed some
evidence that they might be outperforming monolinguals because of their consistently higher
mean scores. If their difference is indeed significant, this would provide the first evidence that
23
bilingual children are better than monolingual children in tracking cross-situational statistics. To
be successful in this task, children would need to track the two word-to-referent pairings
simultaneously. Bilinguals might have benefited in this task due to their constant need to track
multiple regularities and structures in their environment. Another possible explanation is that,
counterintuitively, the smaller vocabulary size in bilinguals might be beneficial to adjective
learning without a strong syntactic cue indicating the adjectival status. Sandhofer and Smith
(2007) compared 24-month-olds who had high CDI scores to those who had low CDI scores.
They found that those who had a smaller noun inventory learned adjectives better without strong
syntactic cues; In contrast, those who had a larger noun inventory learned adjectives better with
strong syntactic cues. This is because children with a larger vocabulary size are more sensitive to
the distinction between nouns and adjectives, and are therefore more likely to use word class
information in learning new adjectives. However, this type of explanation seems to be unlikely
given our sample because there is no significant correlation between performance and CDI
scores, in either monolinguals or bilinguals.
This study provides three important contributions. First, we found that bilinguals do not, as
previously suggested, always suspend the MEP or systematically infer a novel word as a second
label for a familiar object. Instead, when given more contextual information, they are able to take
other cues in the environment into account in resolving referential ambiguity. Therefore, future
studies should focus more on how children utilize or weigh the different cues in the environment,
instead of focusing on whether they adhere to or suspend a single type of cue. Second, neither
monolinguals nor bilinguals made use of the MEP equally well in all conditions. In particular,
based on our findings, children appeared to successfully use the MEP only when the colour-
object pairings were atypical, which suggests that colour typicality affects the ease of learning
novel colour labels. Third, this study also adds to the growing literature that young children can
make use of cross-situational statistics in word learning. Thus far, there is only a handful of
studies of cross-situational word learning in children and no study has been done on bilingual
children. In fact, word-to-object pairings are often learned across multiple situations in everyday
contexts due to the many potential referents in the environment. Future studies can directly
compare monolingual and bilingual children’s abilities in tracking cross-situational statistics.
This strategy could help tease apart possible explanations for the results in current study, such as
whether children are using both the MEP and statistical information, or only statistical
24
information in learning colour names, if indeed any difference between monolinguals and
bilinguals emerges.
To conclude, this study highlights the importance of considering the ecological validity as well
as the richness of the word learner’s home environment. It also highlights the importance of
linguistic background in shaping children’s word learning strategies, which, in turn, shape lexical
development.
25
References
Akhtar, N. & Montague, L. (1999). Early lexical acquisition: the role of cross-situational
learning. First Language, 19, 347–358.
Antovich, D. M., & Graf Estes, K. (2018). Learning across languages: Bilingual experience
supports dual language statistical word segmentation. Developmental Science, 21(2), 1-
11.
Booth, A. E., & Waxman, S. R. (2002). Word learning is “smart”: Evidence that conceptual
information affects preschoolers’ extension of novel words. Cognition, 84(1), 11–22.
https://doi.org/10.1016/S0010-0277(02)00015-X
Bunce, J. P., & Scott, R. M. (2017). Finding meaning in a noisy world: Exploring the effects of
referential ambiguity & competition on 2·5-year-olds’ cross-situational word learning.
Journal of Child Language, 44(3), 650–676. https://doi.org/10.1017/S0305000916000180
Byers-Heinlein, K. (2017). Bilingualism affects 9-month-old infants’ expectations about how
words refer to kinds. Developmental Science, 20(1), 1–10.
https://doi.org/10.1111/desc.12486
Byers-Heinlein, K., & Werker, J. F. (2009). Monolingual, bilingual, trilingual: Infants’ language
experience influences the development of a word-learning heuristic. Developmental
Science, 12(5), 815–823. https://doi.org/10.1111/j.1467-7687.2009.00902.x
Chen, C. H., Gershkoff-Stowe, L., Wu, C. Y., Cheung, H., & Yu, C. (2017). Tracking Multiple
Statistics: Simultaneous Learning of Object Names and Categories in English and Mandarin
Speakers. Cognitive Science, 41(6), 1485–1509. https://doi.org/10.1111/cogs.12417
Clark, E. V. (1988). On the logic of contrast. Journal of Child Language, 15(2), 317-335.
Davidson, D., & Tell, D. (2005). Monolingual and bilingual children’s use of mutual exclusivity
in the naming of whole objects. Journal of Experimental Child Psychology, 92(1), 25–45.
https://doi.org/10.1016/j.jecp.2005.03.007
26
Escudero, P., Mulak, K. E., Fu, C. S. L., & Singh, L. (2016). More limitations to
monolingualism: Bilinguals outperform monolinguals in implicit word learning. Frontiers
in Psychology, 7, 1–13. https://doi.org/10.3389/fpsyg.2016.01218
Fantz, R. L., Fagan, J. F., & Miranda, S. B. Early perceptual development as shown by visual
discrimination, selectivity, and memory with varying stimulus and population parameters.
In L. Cohen & P. Salapatek (Eds.), Infant perception: from sensation to cognition. Vol. 1.
Basic visual processes. New York: Academic Press, 1975
Fenson, L., Dale, P. S., Reznick, J. S., Thai, D., Bates, E., Hartung, J. P., Pethick, S., & Reilly,
J.S. (1991). Technical manualfor the MacArthur Communicative Development
Inventories. San Diego, CA: San Diego State University.
Graham, S. A., Cameron, C. L., & Welder, A. N. (2005). Preschoolers’ extension of familiar
adjectives. Journal of Experimental Child Psychology, 91(3), 205–226.
https://doi.org/10.1016/j.jecp.2005.03.001
Graham, S. A., & Poulin-Dubois, D. (1999). Infants’ reliance on shape to generalize novel labels
to animate and inanimate objects. Journal of Child Language, 26(2), 295–320.
https://doi.org/10.1017/S0305000999003815
Gasser, M., & Smith, L. B. (1998). Learning nouns and adjectives: A connectionist account.
Language and Cognitive Processes, 13(2), 269-306.
Hall, D. G., Waxman, S. R., & Hurwitz, W. M. (1993). How Two‐ and Four‐Year‐Old Children
Interpret Adjectives and Count Nouns. Child Development, 64(6), 1651–1664.
https://doi.org/10.1111/j.1467-8624.1993.tb04205.x
Hansen, M. B., & Markman, E. M. (2009). Children’s Use of Mutual Exclusivity to Learn Labels
for Parts of Objects. Developmental Psychology, 45(2), 592–596.
https://doi.org/10.1037/a0014838
Hollich, G. (2005). Supercoder: A program for coding preferential looking (Version 1.5).
[Computer Software]. West Lafayette: Purdue University.
27
Houston-Price, C., Caloghiris, Z., & Raviglione, E. (2010). Language experience shapes the
development of the mutual exclusivity bias. Infancy, 15(2), 125–150.
https://doi.org/10.1111/j.1532-7078.2009.00009.x
Jones, S. S., & Smith, L. B. (2002). How children know the relevant properties for generalizing
object names. Developmental Science, 5(2), 219–232. https://doi.org/10.1111/1467-
7687.00224
Jones, S., Smith, L., & Landau, B. (1991). Object Properties and Knowledge in Early Word
Learning. Child Development, 62, 499–516.
Kandhadai, P., Hall, D. G., & Werker, J. F. (2017). Second label learning in bilingual and
monolingual infants. Developmental Science, 20(1), 1–14.
https://doi.org/10.1111/desc.12429
Klibanoff, R. S., & Waxman, S. R. (2000). Basic level object categories support the acquisition
of novel adjectives: Evidence from preschool-aged children. Child Development, 71(3),
649–659. https://doi.org/10.1111/1467-8624.00173
Kovács, Á. M., & Mehler, J. (2009). Flexible Learning of Multiple Speech. Science, 325(5940),
611. https://doi.org/10.1126/science.1173947
Landau, B., Smith, L., & Jones, S. (1998). Object shape, object function, and object
name. Journal of Memory and Language, 38(1), 1-27.
Liittschwager, J. C., & Markman, E. M. (1994). 16-Month-Old and 24-Month-Old - Use of
Mutual Exclusivity as a Default Assumption in 2nd-Label Learning. Developmental
Psychology, 30(6), 955–968.
Markman, E. M. (1991). The whole-object, taxonomic, and Mutual Exclusivity Assumptions As
Initial Constraints on Word Meanings. In J. P. Gerlman, S.A., Byrnes (Ed.), Perspectives on
language and thought: Interrelations in development; perspectives on language and
thought: Interrelations in development (pp. 72–106). New York, NY: Cambridge University
Press.
28
Markman, E. M. (1990). Constraints children place on word meanings. Cognitive Science, 14(1),
57–77. https://doi.org/10.1016/0364-0213(90)90026-S
Markham, E.M. (1994). Constraints on word meaning in early language learning. Language, 92,
199-277
Markman, E. M., & Wachtel, G. F. (1988). Children’s use of mutual exclusivity to constrain the
meanings of words. Cognitive Psychology, 20(2), 121–157. https://doi.org/10.1016/0010-
0285(88)90017-5
Pinto, J. P., Fernald, A., McRoberts, G. W., & Cole, S. (1998). Reliability and validity in infant
auditory preference procedures. In C. Rovee-Collier, L. P. Lipsitt, & H. Hayes (Eds.),
Advances in infancy research (pp. 278–297). Norwood, NJ: Ablex
Sandhofer, C., & Smith, L. B. (2007). Learning Adjectives in the Real World: How Learning
Nouns Impedes Learning Adjectives. Language Learning and Development, 3(3), 233–267.
https://doi.org/10.1080/15475440701360465
Scott, R. M., & Fisher, C. (2012). 2.5-year-olds use cross-situational consistency to learn verbs
under referential uncertainty. Cognition, 122(2), 163–180.
https://doi.org/10.1016/j.cognition.2011.10.010.2.5-year-olds
Smith, L. B., Gasser, M., & Sandhofer, C. M. (1997). Learning to talk about the properties of
objects: A network model of the development of dimensions. In R. L. Goldstone, D. L.
Medin, & P. G. Schyns (Eds.), The psychology of learning and motivation, Vol. 36.
Perceptual learning (pp. 219-255). San Diego, CA, US: Academic Press.
Smith, L., & Yu, C. (2008). Infants rapidly learn word-referent mappings via cross-situational
statistics. Cognition, 106(3), 1558–1568. https://doi.org/10.1016/j.cognition.2007.06.010
Soja, N.N., Carey, S., & Spelke, E.S. (1991). Ontological categories guide young children's
inductions of word meaning: object terms and substance terms. Cognition, 38(2), 179-
211.
29
Suanda, S. H., Mugwanya, N., & Namy, L. L. (2014). Cross-situational statistical word learning
in young children. Journal of Experimental Child Psychology, 126, 395–411.
https://doi.org/10.1016/j.jecp.2014.06.003
U. S. Census Bureau. (2010). The 2011 Statistical Abstract. Languages Spoken at Home by
Language: 2008, Table 53. Retrieved July 4, 2019 from http://www.census.
gov/compendia/statab/cats/population/ancestry_language_spoken_at_home.html
Waxman, S.R., Markow, D. B. (1998). Object Properties and Object Kind: Twenty-One-Month-
Old Infants’ Extension of Novel Adjectives.pdf. Child Development, 69, 1313–1329.
Yu, C., & Smith, L. B. (2007). Rapid word learning under uncertainty via cross-situational
statistics. Psychological Science, 18(5), 414–420. https://doi.org/10.1111/j.1467-
9280.2007.01915.x