Cross-situational word learning in Dutch infants
Comparing modes of word presentation.
- Master thesis of Carola M. Werner -
Internship period: Student: Student number: Master: Track: Organization: Unit: Daily supervisor: Co-assessor: Amount of ECTs: Jan. – Dec. 2014 C.M. Werner BSc 10529462
MSc Brain and Cognitive Sciences Cognitive Science
University of Amsterdam Developmental Psychology Mw. Dr. C.M.M. Junge
Mw. Prof. Dr. R.E.M. Raijmakers 26
2
Abstract
Word learning is difficult. One of the factors that could influence infants’ success of word learning is auditory context; words may be presented either in isolation or in supportive carrier-phrases. Whereas parents tend to present novel words embedded in supportive, familiar sentence frames, experimental research studying novel word learning in the lab often present novel words in isolation. In the current study, we compared both modes of auditory context in a more ecologically valid word learning paradigm: cross-situational word learning. Within this paradigm infants have to keep track of object-word pairings in order to learn the correct mappings. To examine which auditory context promotes word learning better, the current study aimed to teach 18-month-olds six novel words which were presented in two different auditory contexts; isolation versus utterances. The results of this experiment showed that monolingual 18-month-olds learn equally well from both modes of presentation, although infants participating in the isolated words condition showed larger learning effects, which suggests that infants find words in isolation easier to learn.
3
Inhoud
Abstract ... 2 Introduction ... 4 Methods ... 6 Participants ... 6 Stimuli ... 6 Apparatus ... 8 Procedure ... 8 Data-analysis ... 9 Results ... 10 Mode of presentation ... 10Individual differences in word learning ... 11
Discussion ... 12
Main results ... 12
Mode of presentation ... 12
Individual differences in word learning ... 13
Future directions ... 14
Conclusions ... 14
References ... 15
Appendix 1: Cross-situational word learning of multi-lingual infants ... 18
4
Introduction
Early word learning already starts around the age of six months with the understanding of several common words (Bergelson & Swingley, 2012). Although word learning of infants mostly evolves without many problems, acquiring this word learning skill is not as easy as it seems when infants start to learn words. In fact, word learning is complicated, because infants who learn novel words have to face several challenges. First, there is the problem of how infants find words in a continuous speech stream, given that they hear many multi-word utterances (Johnson, Lahey, Ernestus, & Cutler, 2013). Recognising separate words from these utterances is difficult, because in a spoken sentence there are no clear pauses to identify single words. Consequently, infants need to segment the spoken sentence into word-like units. When they are able to recognise the word from a multi-word utterance from one speaker, they also need to learn to recognise it from other speakers who use for example different intonation patterns (i.e. acoustic variance) (Singh, White, & Morgan, 2008). Second, infants need to recognise the object that the word refers to. Third, when the word and object are correctly recognised, infants have to make the right connection between word and object (Waxman & Lidz, 2006).
Successful word learning is also dependent on mode of presentation. Words can be presented in two different ways: in isolation (e.g. “Cup”) or embedded in multi-word utterances (e.g. “Here is your cup”). Some studies suggest that word learning is easier when words are presented in isolation, because infants do not need to segment the word from multi-word utterances (Brent & Siskind, 2001; Junge, Kooijman, Hagoort, & Cutler, 2012). On the other hand, parents tend to present words in carrier phrases when a natural word learning environment is simulated. A recent corpus study of word learning data showed that in 72% of the cases parents taught novel words via supportive carrier phrases (Johnson et al., 2013). Other studies report that even 88% to 93% of all daily life Child-Directed Speech (CDS) consists of words embedded in supportive sentences (van de Weijer, 1999). In accordance with these findings, several studies suggest that at 18 months of age the speaker’s referential behaviour becomes increasingly important for infants (i.e. the infants will attach more value to sentential context) (Hollich et al., 2000; Moore, Angelopoulos, & Bennett, 1999). Hence, the current study examines whether word learning is easier when words are presented in isolation (e.g. no segmentation required) or in supportive carrier phrases (e.g. it facilitates the interpretation of the communicative intent (Tomasello, 2000)).
The effect of mode of presentation was investigated in 18-month-old infants in a study that aimed to see when infants recognised familiar object names faster: presented in isolation (“Doggy”) or embedded in a familiar carrier-phrase (e.g. “Look at the doggy!”). The results showed that infants were 120 milliseconds slower when the words were presented in isolation compared to the supportive carrier phrases (Fernald & Hurtado, 2006). Accordingly, this suggests that word recognition is easier when words are presented in a familiar carrier phrase; infants can anticipate when the target word will be presented in the utterance. A more recent study taught novel words to 14-month-olds. The results of this study showed that success of word learning is dependent on the information that is present in the referential context; 14-month-olds need some familiar referential cues (e.g. familiar carrier phrases like: “Look at the X!”) in order to make the right word-referent pairings (Fennell & Waxman, 2010). There is also research demonstrating that infants find it easier to recognize words that are presented in isolation. In a speech segmentation study, most 10-month-olds were able to recognize words presented in isolation, but fewer infants were also able to
5 recognize then when words were first presented within an utterance (Junge et al., 2012). Note, however, that speech segmentation studies typically do not require infants to understand the sentences in which the (often low-frequency) target words are presented: they only test infants’ recognition of word forms. Hence, such words occur in sentences that generally lack clear communicative intent. In other words, it could be that infants might equally succeed in learning words presented in continuous speech that more closely follows the type of sentences that parents typically use to present novel words. Presenting novel words in typical-naming situations (“frequent frames”; Mintz, 2003) might even boost word learning, not only because such frequent frames highlight the teaching situation (Tomasello, 2009), but also because they decrease the speech segmentation problem described above (Bortfeld, Morgan, Golinkoff, & Rathbun, 2005).
Another clear difference between studies finding either an advantage or disadvantage for sentential context is the age of the infants: for younger infants words presented in utterances proves more difficult than for older infants. It is likely that with increasing experience of listening to their native language, older infants have eventually mastered the speech segmentation problem. In a study comparing two age groups (15-month-olds and 18-month-olds), a natural learning situation was simulated by teaching infants novel words. Previous studies on mode of presentation focussed on word recognition; they used familiar objects. In this study however, infants had to learn the names of two novel objects. Learning novel words resembles a natural word learning situation in that sense that infants have to learn many novel objects to increase their vocabulary. Infants were tested in a within-subjects design on their ability to learn novel words presented either in isolation or within sentences. To test this, two conditions were used. One condition presented the names in isolation, while in the other condition the names were presented in familiar carrier phrases. Each trial consisted of 1 object on the screen and repeated the name of this object three times, either in isolation or in a simple carrier phrase. For younger infants (15 months) boys performed better in a training with isolated words whereas girls performed better in a training with words embedded in sentences. The older infants (18 months) acquired the names of the objects in both contexts equally well (Trehub & Shenfield, 2007).
The earlier described studies all share one disadvantage; they only showed one object per trial and repeated its name many times. This is not how learning naturally proceeds. In a natural learning environment there are many objects visible at the same time. The child needs to resolve out of a variety of these objects which is the one that a word refers to. Infants need to keep track to object-referent pairings in order to learn the name of a novel object (Yu & Smith, 2010; Yu, Zhong, & Fricker, 2012). For example, in the first situation a parent talks about a cup and tells the infant that the cup is placed on the table. In the next situation, the cup has been taken to the kitchen. The parent names the table again, this time in combination with a plate. The infant links these two situations and concludes that the only object that was present both times was the table, because the cup and the plate varied. Therefore, the unfamiliar object has to be the table. This kind of learning is called cross-situational learning. Cross-cross-situational learning approaches a natural learning situation more closely allowing us to investigate word learning in a laboratory setting. In a cross-situational learning paradigm there are always two objects presented in different configurations on the computer screen, while the names of these objects are presented via sound clips. Recent studies using a cross-situational learning paradigm show that 12- and 14- month-old infants already can do this (Smith & Yu, 2008; Yu & Smith, 2010). Although these studies examined word learning in a cross-situational learning paradigm in order to approach a more ecologically valid visual learning situation, the
6 auditory context in these studies was limited; words were only presented in isolation. However, parents tend to present novel words more often in supportive carrier-phrases (van de Weijer, 1999; Johnson et al., 2013). Therefore, the present study will also create a more natural auditory context, in addition to an ecologically valid visual learning situation. We will train 18-month-old infants to map six non-existing Dutch-sounding words to the correct referents, with always two objects in varying configurations present on the screen. Hence, the aim of this study is to compare both modes of presentation: do children learn more words when they are presented in isolation, or when they are presented in a supporting sentence context? In addition, we will examine individual learning differences.
Methods
Participants
The 69 participants in this research were all month-old infants. They were healthy, full-term 18-month-old children, all recruited from the Amsterdam Babylab Database. All infants were from monolingual Dutch families with no history of neurological or language impairments. In total, 20 infants had to be excluded due to too few valid test trials (i.e. less than 2 out of 12 possible trials; at least 1 in the first 6 trials and 1 in the second 6 trials). The 49 remaining infants had an average age of 18.2 (SD = 0.44) months. The infants were divided over two conditions; 22 infants participated in the utterance condition (7 females) and 27 infants participated in the isolated words condition (14 females). Every parent signed an informed consent approved by the Ethical Committee. All infants and their parents received a book as a reward after participating in the experiment.
Stimuli
Visual Stimuli
The visual stimuli consisted of six pictures of unfamiliar stuffed animal toys in different bright colours (Fig. 1). None of the infants had seen these toys before. In each trial two pictures, edited with Photoshop, were presented simultaneously on a dark grey background. Each object was roughly 20 x 20 cm on the screen, with an average distance between objects (i.e. the center of each object is on the same spot on the screen, resulting in varying differences between the objects).
Fig. 1. These are the used stuffed toys: Gemer, Kaven, Pola,
7
Verbal Stimuli
The names of these animals were six non-existing words following the phonotactic probabilities of Dutch (i.e. foenie, kaven, gemer, pola, miekel & buiming). There were six types of carrier sentences for the familiarization:
1. Hier is een X en daar een X. (Here is a X and there a X.)
2. Kijk, dit is een X en dit is een X. (Look, this is a X and this is a X.)
3. Wat leuk, weer de X en de X. (How nice, there are de X and the X again.) 4. Daar is de X, samen met de X. (There is de X, together with the X.) 5. Hier is de X, samen met de X. (Here is the X, together with the X.) 6. Kijk, dit zijn de X en de X. (Look, here are the X and the X.)
In the familiarization phase the text referred to both animals shown on the screen, whereas in the test phase the infant was asked to look at one of the animals specifically using three types of carrier sentences:
1. Zie je de X? X. (Do you see the X? X.) 2. Waar is de X? X. (Where is the X? X.) 3. Kijk naar de X. X. (Look at the X. X.)
A native Dutch female speaker recorded the stimuli in an animated child-directed manner, in a sound-attenuating booth. Stimuli were sampled to disk at 44.1 kHz mono. The mean duration of target words in the familiarization phase was 624 ms (range: 405 ms – 945 ms). Sentences lasted on average 3374 ms (range: 2190 ms – 4556 ms), with a mean interval between target words of 1110 ms (range: 330 ms – 2503 ms).
We labeled the onset of target words in sentences based on auditory and visual inspection using PRAAT software (Boersma & Weenink, 2014). For the experimental group who listened to words in isolation, all other words than the target words from the familiarization sentences were replaced by silence. In this way acoustic properties and timing of the target words in the familiarization sentences were kept constant across experimental manipulations (See Fig. 2). For the test phase of both conditions timing was manipulated such that target started at 2500 ms from onset of the visual stimulus.
A.
B.
Fig. 2. Visual patterns in PRAAT of verbal stimuli. A. Words embedded in sentences. B. Words presented in
8
Design
Infants either participated in the sentence condition or the isolated word condition (familiarization phase consisting of the same auditory targets spliced from sentences). The test phase was identical for both conditions. For each condition (sentences; isolation) we created eight versions of the experiment to rule out order effects. We also created two different name-object pairings, using two varying mappings. For example, the yellow object is called the Buiming in version A, but Pola in version B. This makes it possible to exclude any preference effects (see Table 1). Half of the versions used pairing A. The other half used pairing B. For each version we used different naming sentences, counter-balancing which object was named first as well as the order of the trials.
Table 1. Varying name-object pairings across versions.
Version A Version B
A Buiming Yellow A B Buiming Green F
A Foenie Orange B B Foenie Red E
A Gemer White C B Gemer Blue D
A Kaven Blue D B Kaven White C
A Miekel Red E B Miekel Orange B
A Pola Green F B Pola Yellow A
Sentence condition
In this condition the names of the animals were presented in supporting sentences. These sentences were often used utterances in child direct speech (e.g., “Look, there are the Foenie and the Gemer again”; “There is the Kaven, together with the Gemer”). Each sound file consisted of two short sentences with a conjunction in the middle. Target words were always placed at the end of the sentences.
Words condition
In this condition the names of the animals were presented in isolation during the familiarization phase. The sound files were created with the computer software Praat. With Praat the exact start and end time of the target words were selected. The remaining parts of the sentence were silenced, but not deleted (see Fig. 2.). Therefore the words in this condition were said at exactly the same time and in exactly the same pitch as in the sentence condition. The test phase of this condition was exactly the same as the test phase of the sentence condition.
Apparatus
The infants’ eyes were tracked using an EyeLink eye tracker: the Remote 1000. This eye tracker was calibrated with a 3-point calibration. The stimuli (20x20 cm) were presented on a 17 inch monitor that was on a distance of 60 cm of the infants’ eye.
Procedure
When the participants arrived they were welcomed in our Babylab. There was some time for the infants to get used to their new environment as we first asked the parent(s) some questions about the development of the child and informed them about the procedure. The parent signed an informed consent form. Subsequently the experiment started. The experiment took place in the
9 adjacent test room. Here the child was put in a car seat, while the parent took place behind the child. The light in the room was dimmed, so there was as less distraction as possible. Subsequently, the Eyelink eye tracker was calibrated. Following Smith & Yu (2008) and Yu & Smith (2010) we had a familiarization phase of 30 trials. Each trial consisted of a picture presenting two objects for five seconds during which a sound file labeling these objects was played, either presenting the words in supporting utterances or in isolation. Each object was presented ten times in combination with one of the other objects. After 30 familiarization trials the test phase began. The test phase consisted of 12 test trials. In each trial two objects were presented on the screen for five seconds, but in contrast to the familiarization phase only the target word was mentioned instead of both objects. Calibration, familiarization and testing took about 10 minutes. Afterwards the parent filled in a questionnaire about the vocabulary of the child (N-CDI) to measure comprehensive and productive vocabulary. At the end of the session the experimenter discussed the test with the parents. All infants and their parents were rewarded with a book.
Data-analysis
Around each target object an area of interest (with an equal size for all objects) was designed to identify where the infant looked at in order to calculate the total looking times for each object. From these total looking times the Proportional Target Looking (PTL) and Latency Longest Look (LLK) were calculated. For the PTL the total looking time to the target object was divided by the total looking time in that trial. For the LLK the longest focus on the target object, without switching to the distractor, was calculated. Next, the longest look for the distractor was calculated and subtracted from the target look. This resulted in the LLK data.
These PTL and LLK data were analysed with a Repeated Measures ANOVA, with naming effect (Pre naming vs. Post naming) as within-subjects factor and mode of presentation (Utterances vs. Isolated words) and gender (male vs. female) as between-subjects factors. The pre naming phase is the time period of the start of the trial until the target is named and the post naming phase is the time period from the moment of naming the target until the end of the trial. We looked at the test phase as a whole (trial 1-12) as well as the effect of the different blocks to see whether there is a learning effect within the test phase (trial 1-6 vs. trial 7-12). We also looked at gender influences, size of vocabulary and correlations with the score in the familiarization phase. Additionally we looked at the average amount of words that is learned for each condition.
10
Results
Mode of presentation
PTL
The PTL data of the test phase is analysed with a Repeated Measures ANOVA. The between subjects factors are mode of presentation (Utterances vs. Isolated Words) and gender (male vs. female). The within subjects factor is the naming effect (Pre naming phase vs. Post naming phase). The PTL data of the pre and post naming phases are summarized in Figure 3. The naming effect is calculated by subtracting the PTL of the Pre naming phase from the PTL of the Post naming phase. When the naming effect equals 50%, infants
perform at chance. An overall ANOVA shows a marginal naming effect of 3.6% (F(1,45) = 3.360, p = 0.073, η2p = 0.069), but there is no interaction with mode of presentation (F(1,45) = 0.473, p = 0.495, η2p = 0.010), nor with gender (F(1,45) = 0.116, p = 0.735, η2p = 0.003). The naming effect of the words condition (i.e. 4.5%) was nominally larger than the naming effect of the utterance condition (i.e. 2.3%), but a ttest did not show a significant difference between the two conditions (t(47) = -0.642, p = 0.524) (see Fig. 3).
LLK
The LLK data of the test phase is also analysed with a Repeated Measures ANOVA. The between subjects factors are mode of presentation (Utterances vs. Isolated Words) and gender (male vs. female). The within subjects factor is again naming effect (Pre naming vs. Post naming phase). An overall ANOVA shows a naming effect of 180.12 ms, i.e. their longest look on the target was on average 180.12 ms longer after naming, compared to before the target was named (F(1,45) = 4.787, p = 0.034, η2p = 0.096). As in the PTL data, we observed no interaction with mode of presentation (F(1,45) = 0.338, p = 0.564, η2p = 0.007), nor with gender (F(1,45) = 0.453, p = 0.505, η2p = 0.010) .
Learning in the test phase
To see whether the learning process still continues in the test phase, the test trials are divided in two blocks of 6 trials (1-6 vs. 7-12). Below, the results of this analysis are described for the PTL- and LLK-data.
Fig. 3. Mean PTL-scores of Pre naming phase vs. Post naming phase of the
11 PTL
An overall ANOVA shows no naming effect (F(1,45) = 2.502, p = 0.121, η2p = 0.053). There is no block effect (F(1,45) = 0.011, p = 0.917, η2p = 0.000) either, nor any interaction of mode of presentation with naming (F(1,45) = 0.098, p = 0.756, η2p = 0.002), gender with naming (F(1,45) = 0.128, p = 0.723, η2p = 0.003), mode of presentation with block effect (F(1,45) = 0.836, p = 0.365, η2p = 0.018), gender with block effect (F(1,45) = 0.097, p = 0.757, η2p = 0.002), or any interaction between naming and block effect (F(1,45) = 1.246, p = 0.270, η2p = 0.027).
LLK
A marginal naming effect is found when LLK-data is analysed with an overall ANOVA (F(1,45) = 3.044, p = 0.088, η2p = 0.063). Across blocks, the LLK after naming was 136.26 ms (min = -601.6 ms, max = 1359.80 ms, SD = 383.89 ms), whereas it was -43.86 ms (min = -590.67 ms, max = 642.00 ms, SD = 249.16 ms) in the pre-naming phase. There is no block effect (F(1,45) = 0.002, p = 0.968, η2p = 0.000), nor any interaction of mode of presentation with naming (F(1,45) = 0.502, p = 0.482, η2p = 0.011), gender with naming (F(1,45) = 0.723, p = 0.400, η2p = 0.016), mode of presentation with block effect (F(1,45) = 0.071, p = 0.790, η2p = 0.002), gender with block effect (F(1,45) = 0.243, p = 0.625, η2p = 0.005), or any interaction between naming and block effect (F(1,45) = 0.302, p = 0.585, η2p = 0.007).
Learned name-object pairings
To see whether some names were easier to learn in comparison to the others, several t-tests are performed. For each infant the delta PTL scores (i.e. Post naming PTL – Pre naming PTL) are calculated per name. These scores are derived from the 12 test trials, so for each name there are two delta PTL scores and these are averaged to get an average delta PTL score for each stuffed animal. One sample t-tests show for version A that the name of the Buiming (yellow object) is learned by a significant amount of infants, t(22) = 2.232, p = 0.036. For version B the name of the Pola (yellow object) is learned by a significant amount of infants, t(20) = 2.779, p = 0.012. This means that the yellow object is learned best across pairings.
Individual differences in word learning
Correlation with familiarization phase
The PTLs of the familiarization phase and of the test phase are normally distributed. The PTL of the last 15 trials familiarization phase has a skewness of -0.148 (SE = 0.340) and a kurtosis of 1.130 (SE = 0.668) and the PTL of the test phase has a skewness of 0.964 (SE = 0.340) and a kurtosis of 1.662 (SE = 0.668). Therefore we performed a Pearson’s correlation test to test whether the performance in the familiarization phase can predict the performance in the test phase.
Fig 4. Correlation between the last 15 trials of the familiarization phase
12 Pearson’s correlation test shows that there is a negative correlation between the performance in the last 15 trials of the familiarization phase and the performance in the test phase of the experiment (r(49) = -0.484, p < 0.001) (see Fig. 4).
Correlation with productive vocabulary (CDI)
For each individual child data is collected about its vocabulary. In contrast with the PTL data of the test phase, the productive vocabulary is non-normally distributed with skewness of 1.964 (SE = 0.340) and kurtosis of 4.519 (SE = 0.668). Therefore a Spearman’s correlation test is performed. This test shows that there is no correlation between the amount of words the infants says and the performance in the test phase of the experiment (r(49) = -0.150, p = 0.305).
Discussion
Main results
The aim of this study was to examine whether success of word learning was dependent of mode of presentation. A second aim was to relate word learning success to individual differences, by linking performance in the test phase to performance in the training phase. To these purposes we trained 49 18-month-olds to map six non-existing Dutch-sounding words to the correct referents. To see whether mode of presentation can influence the success of word learning, we presented the names of the object in two different ways: embedded in a supportive sentence, or in isolation. The participants were divided into two groups and each group participated in one of the conditions. Our data shows that 18-month-old infants are able to map the meanings to words, but that mode of presentation does not matter. Our data also suggests that the process of word learning has its onset in the training phase, but this does not extend to the test phase. During the training phase the word learning process was inconsistent; the majority learned only one out of the six novel words. At the individual level we found a large variation in the process of word learning. In order to explain this large variation in the test phase, we looked at variation in the familiarization phase, gender and productive vocabulary. First, performances in the training and test phase were negatively correlated. Second, we did not find any gender specific differences in our data, nor a correlation with productive vocabulary. Below, we will further discuss our main findings in more detail.
Mode of presentation
Our main finding is that there is a naming effect regardless of mode of presentation; 18-month-olds learn the names of novel objects. Although the effect is small (e.g. increase of 3.6%), it is comparable with the results of Smith & Yu (2008) (e.g. increase of 7%). There was a large variation in the sizes of individual naming effects; a small majority had a positive naming effect (28 out of 49).
Clearly, mode of presentation does not matter. On the one hand, presenting words in isolation might increase successful learning, because there is no segmentation required. On the other hand, supportive carrier phrases increase segmentation problems, but also provide more information about the communicative intent (Tomasello, 2000), hence more referential cues. Besides, parents tend to teach novel words embedded in supportive sentences more than in isolation in general (Johnson et al., 2013; van de Weijer, 1998). We tested infants at the age of 18 months, because at this age referential cues start to become more important for word learning (Moore et al., 1999; Tomasello, Strosberg, & Akhtar, 1996). Therefore, we hypothesized that the infants in the utterance
13 condition would perform better. Although we found no differences in performance between the modes of presentation, there was a nominal difference: infants in the isolated words condition performed slightly better (i.e. increase of 4.5 % vs. increase of 2.3 % in utterance condition). This suggests that it is easier for 18-month-olds to learn novel words if presented in isolation; in that case there is no segmentation problem. Although segmentation made the learning process more complicated in the utterance condition, the infants still show a learning effect, suggesting they are able to segment the carrier-phrases. It could be that segmentation was so hard, because we used six different types of carrier-phrases. Perhaps, success of word learning could be improved if we had used the same syntactic structure consistently. This keeps the communicative intent constant, but decreases the word segmentation problem. Another way to decrease the segmentation load is to familiarize infants with the target words first (Brent & Siskind, 2001; Swingley, 2007).
The process of learning is a continuously ongoing process. Although we divided the experiment in a training and test phase, it could be that learning still goes on in the test phase. However, our results do not support this hypothesis; during the training phase the learning effect was not very robust, whereas during the test phase there was no learning at all. Besides, there was not much consistent learning. Whereas Smith & Yu (2008) found that most of their participating infants learned four out of six novel words on average, the infants in our study only learned one out of six words. In version A the name-object pairing of the Buiming was learned by all infants, while in version B the Pola was learned by all infants. In both conditions the Buiming (version A) and the Pola (version B) were yellow. This suggests that the yellow object was easier to learn than the other objects, but also shows that learning occurred very inconsistently.
Apparently, there is much individual variation in learning; although a small majority showed naming effects, a substantial number did not show evidence of learning. Moreover, there was no learning during the test phase and additional analyses showed very inconsistent learning. We examined what factors could explain this variation in word learning success. Below, we will further elaborate on these individual learning differences.
Individual differences in word learning
We observed many individual differences during the test phase of the experiment. Here, we studied the relation between performances in the different phases of the experiment, gender differences and the influence of productive vocabulary size.
First, we hypothesized that the performances in the familiarization phase and test phase would correlate positively (e.g. high score in familiarization phase equals high score in test phase). Recent work suggests that infants learn the name of an object if it is named at the moment that they make this object visually dominant, but not when naming occurs during a less visually selective moment (Yu & Smith, 2012). So, in our study, infants who showed a larger learning effect might also be the infants who were interested in the target object, thus making it visually dominant when naming occurred, resulting in a correct word-referent mapping. On the opposite, others were maybe interested in the distractor object, making the wrong word-referent mapping. However, our results show a negative correlation between performance in the last block of the familiarization phase and performance in the test phase (see Fig. 3). Possibly, infants who learn better during the familiarization phase build up an expectation about the switch of objects; in the familiarization phase both objects are named in contrast with the test phase where only the target object is named. These infants already anticipate by switching to the distractor, but we only ask for the target in the test
14 phase. Hence, they have shorter looking times to the target object, resulting in a smaller, often even negative, naming effect. The infants who showed less successful word learning in the familiarization phase did not learn yet that there is a switch of objects. Therefore, they perform better in the test phase, where they possibly still show a learning curve.
Second, we looked at gender specific differences. Although the majority of studies did not find gender specific differences for the way of word learning (e.g. Houston-Price, Plunkett, & Harris, 2005; Tan & Schafer, 2005), some report age-related gender differences. A recent study on mode of presentation (isolation vs. supportive sentences) found a gender effect at the age of 15 months (boys preferred isolation vs. girls supportive sentences), but no effect at the age of 18 months (Trehub & Shenfield, 2007). Based on these results we expected to find no gender specific preferences for mode of presentation. Our results indeed confirmed that 18-month-olds do not show a gender specific preference for mode of presentation.
Third, the size of the productive vocabulary of the learner, measured with parental reports, might show a relation with performance in the test phase. Although we expected that infants with a larger vocabulary would perform better in the test phase, because they are more experienced learners, we did not find a correlation of productive vocabulary with performance in the test phase. Note, however, that many other studies on word learning also do not find this pattern. This in contrast to studies on known word processing. Moreover, there has also been some criticism on the use of parental reports, because they are prone to parental biases (Houston-Price, Mather, & Sakkalou, 2007; Tomasello & Mervis, 1994). Indeed, in our study productive vocabulary was non-normally distributed, suggesting parental biases. This might be the cause of the missing correlation between productive vocabulary and performance in the test phase.
Future directions
The present study design had a few shortcomings. A substantial problem of our design is that the segmentation load for the utterance condition might be too high because we used a variety of sentences. Future studies should examine how to solve this problem. Possibly, the segmentation load would be less when the same syntactic structure is used. Another solution for the segmentation problem could be to familiarize infants with the target words first. These two options should be further investigated using a within-subjects design instead of a between subjects design.
Moreover, there is a large educational bias in the present study; most infants are raised with a high Social Economic Status (SES). Recent work revealed that infants from higher SES-families have an advantage in language-processing compared to infants from lower SES-families (Fernald, Marchman, & Weisleder, 2013; Hoff, 2003). Future studies need to investigate whether this also is the case for cross-situational word learning of 18-month-olds.
Conclusions
In sum, we can conclude that there is no effect of mode of presentation. Infants will learn the name-object parings in both ways. Therefore mode of presentation does not matter when presenting words to 18-month-olds, for example in treatment for word learning problems. Furthermore, a great variation in word learning is present due to individual differences; each infant has its own way of novel word learning.
15
References
Bergelson, E., & Swingley, D. (2012). At 6-9 months, human infants know the meanings of many common nouns. Proceedings of the National Academy of Sciences of the United States of America, 109(9), 3253–8. doi:10.1073/pnas.1113380109
Boersma, P.P.G., & Weenink, D.J.M. (2014). Praat: doing phonetics by computer [Computer program].Version 5.3.61, retrieved 20 January 2014 from http://www.praat.org/
Bortfeld, H., Morgan, J. L., Golinkoff, R. M., & Rathbun, K. (2005). Mommy and me: familiar names help launch babies into speech-stream segmentation. Psychological Science, 16(4), 298–304. doi:10.1111/j.0956-7976.2005.01531.x
Brent, M. R., & Siskind, J. M. (2001). The role of exposure to isolated words in early vocabulary development. Cognition, 81(2), B33–B44. doi:10.1016/S0010-0277(01)00122-6
Fennell, C. T., & Waxman, S. R. (2010). What paradox? Referential cues allow for infant use of phonetic detail in word learning. Child development, 81(5), 1376-1383.
Fernald, A., & Hurtado, N. (2006). Names in frames: infants interpret words in sentence frames faster than words in isolation. Developmental Science, 9(3), F33–40.
doi:10.1111/j.1467-7687.2006.00482.x
Fernald, A., Marchman, V. A., & Weisleder, A. (2013). SES differences in language processing skill and vocabulary are evident at 18 months. Developmental Science, 16(2), 234–248.
doi:10.1111/desc.12019
Hoff, E. (2003). The Specificity of Environmental Influence: Socioeconomic Status Affects Early Vocabulary Development Via Maternal Speech. Child Development, 74(5), 1368–1378. doi:10.1111/1467-8624.00612
Hollich, G. J., Hirsh-Pasek, K., Golinkoff, R. M., Brand, R. J., Brown, E., Chung, H. L., … Rocroi, C. (2000). Breaking the language barrier: an emergentist coalition model for the origins of word learning. Monographs of the Society for Research in Child Development, 65(3), i–vi, 1–123. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/12467096
Houston-Price, C., Mather, E., & Sakkalou, E. (2007). Discrepancy between parental reports of infants’ receptive vocabulary and infants' behaviour in a preferential looking task. Journal of Child Language, 34(4), 701–24. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/18062356 Houston-Price, C., Plunkett, K., & Harris, P. (2005). “Word-learning wizardry” at 1;6. Journal of Child
Language, 32(1), 175–89. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/15779882 Johnson, E. K., Lahey, M., Ernestus, M., & Cutler, A. (2013). A multimodal corpus of speech to infant
and adult listeners. The Journal of the Acoustical Society of America, 134(6), EL534. doi:10.1121/1.4828977
Junge, C., Kooijman, V., Hagoort, P., & Cutler, A. (2012). Rapid recognition at 10 months as a
predictor of language development. Developmental Science, 15(4), 463–73. doi:10.1111/j.1467-7687.2012.1144.x
16 Mintz, T. H. (2003). Frequent frames as a cue for grammatical categories in child directed speech.
Cognition, 90(1), 91–117. doi:10.1016/S0010-0277(03)00140-9
Moore, C., Angelopoulos, M., & Bennett, P. (1999). Word learning in the context of referential and salience cues. Developmental Psychology, 35(1), 60–8. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/9923464
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(2), 157–178. doi:10.1080/15475440801922131
Smith, L., & Yu, C. (2008). Infants rapidly learn word-referent mappings via cross-situational statistics. Cognition, 106(3), 1558–68. doi:10.1016/j.cognition.2007.06.010
Swingley, D. (2007). Lexical exposure and word-form encoding in 1.5-year-olds. Developmental Psychology, 43(2), 454–64. doi:10.1037/0012-1649.43.2.454
Tan, S. H., & Schafer, G. (2005). Toddlers’ novel word learning: Effects of phonological representation, vocabulary size and parents' ostensive behaviour. First Language, 25(2), 131–155.
doi:10.1177/0142723705050338
Tomasello. (2000). The item-based nature of children’s early syntactic development. Trends in Cognitive Sciences, 4(4), 156–163. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/10740280
Tomasello, M. (2009). Constructing a Language: A Usage-Based Theory of Language Acquisition. Retrieved from http://books.google.com/books?hl=nl&lr=&id=7M_lSEfzTQoC&pgis=1
Tomasello, M., & Mervis, C. B. (1994). The Instrument Is Great, but Measuring Comprehension Is Still a Problem. Monographs of the Society for Research in Child Development, 59(5), 174–79. Retrieved from http://eric.ed.gov/?id=EJ493709
Tomasello, M., Strosberg, R., & Akhtar, N. (1996). Eighteen-month-old children learn words in non-ostensive contexts. Journal of Child Language, 23(01), 157–176.
doi:10.1017/S0305000900010138
Trehub, S. E., & Shenfield, T. (2007). Acquisition of early words from single-word and sentential contexts. Developmental Science, 10(2), 190–8. doi:10.1111/j.1467-7687.2007.00545.x Van de Weijer, J. (1999). Language input for word discovery. PhD Thesis, Radboud University
Nijmegen, Nijmegen.
Waxman, S. R., & Lidz, J. L. (2006). Early Word Learning. In W. Damon & R. M. Lerner (Eds.), Handbook of Child Psychology (6th ed.). Hoboken, NJ, USA: John Wiley & Sons, Inc. doi:10.1002/9780470147658.chpsy0207
Yu, C., & Smith, L. B. (2010). What you learn is what you see: using eye movements to study infant cross-situational word learning. Developmental Science, 14(2), 165–180. doi:10.1111/j.1467-7687.2010.00958.x
17 Yu, C., & Smith, L. B. (2012). Embodied attention and word learning by toddlers. Cognition, 125(2),
244–62. doi:10.1016/j.cognition.2012.06.016
Yu, C., Zhong, Y., & Fricker, D. (2012). Selective attention in cross-situational statistical learning: evidence from eye tracking. Frontiers in Psychology, 3, 148. doi:10.3389/fpsyg.2012.00148
18
Appendix 1:
19
Introduction
As stated earlier, word learning is difficult. Infants have to face several challenges in order to achieve successful word learning (Singh et al., 2008; Waxman & Lidz, 2006). Most studies to early word learning and modes of presentation mainly focus on monolingual infants (Smith & Yu, 2008; Yu & Smith, 2010; Trehub & Shenfield, 2007; Fernald & Hurtado, 2006). However, multilingual infants have to face an extra challenge on top of the normal difficulties in word learning, because they need to learn the name of each object in multiple languages. In the present study we also tested a group of participants that was multilingual. In this appendix their results will be discussed.
Methods
The stimuli, design, apparatus, procedure and data-analysis were exactly the same as in the original study. The only thing that differed from the original study was the group of participants. In this part of the study 21 multi-lingual 18-month-olds participated (13 females). They were also recruited from the Amsterdam Babylab Database and were healthy, full-term infants with an average age of 18.2 (SD = 0.40) months and had no history of neurological or language impairments. In the utterance condition 11 infants participated (6 females) and in the words condition 10 infants participated (7 females). Every parent signed an informed consent form approved by the Ethical Committee. At the end of the session the participant received a book as reward.
Results PTL
The PTL data of the test phase is analysed with a Repeated Measures ANOVA. Between subjects factors are mode of presentation (Utterances vs. Isolated Words) and gender (male vs. female). The within subjects factor is the naming effect (Pre naming vs. Post naming phase). An overall ANOVA did not show a naming effect (F(1,15) = 0.408, p = 0.533, η2p = 0.026), nor an interaction with gender (F(1,15) = 1.089, p = 0.313, η2p = 0.068), or with mode of presentation (F(1,15) = 0.440, p = 0.517, η2p = 0.028).
LLK
The LLK data is also analysed with a Repeated Measures ANOVA. Again, between subjects factors are mode of presentation (Utterances vs. Isolated Words) and gender (male vs. female). The within subjects factor is the naming effect (Pre naming vs. Post naming phase). An overall ANOVA does not show a naming effect (F(1,15) = 0.027, p = 0.871, η2p = 0.002). There was also no interaction with gender (F(1,15) = 0.523, p = 0.481, η2p = 0.034), nor with mode of presentation (F(1,15) = 0.557, p = 0.467, η2p = 0.036). However, there was a marginal three-way interaction between naming effect, gender and mode of presentation (F(1,15) = 4.422, p = 0.053, η2p = 0.228).
Correlation with familiarization phase
To test whether there is a correlation between the performance in the familiarization phase and the test phase a correlation test is performed. Because the PTL data of the last 15 trials of the
familiarization phase is normally distributed with skewness of 0.798 (SD = 0.501) and kurtosis of 0.347 (SD = 0.972) and the PTL data of the test phase is also normally distributed, a Pearson correlation test is performed. No correlation is found between the score in the last 15 trials of familiarization phase and the score in the test phase (r(17) = 0.093, p = 0.722).
20 Correlation with vocabulary (CDI)
For the multi-lingual children data is collected about their vocabulary as well. Their productive vocabulary is non-normally distributed with skewness of 2.768 (SD = 0.512) and kurtosis of 8.616 (SD = 0.992). However, the PTL data of the test phase is normally distributed with skewness of -0.235 (SD = 0.524) and kurtosis of -0.225 (SD = 1.014). Therefore, a Spearman’s correlation test is performed. This test shows that there is no correlation between productive vocabulary and performance in the test phase of the experiment (r(18) = -0.108, p = 0.668).
Discussion
The aim of this additional analyses was to see whether the fact that word learning is more complicated for multilingual infants (i.e. they have to learn the name of an object in multiple languages) affects their success in word learning. Previous work suggests that multilingualism does not interfere with the process of early word learning (Byers-Heinlein, Fennell, & Werker, 2012). Therefore, we hypothesized that multilingual infants would learn the presented words as well as the mono-lingual infants. However, the results of our analysis show that multilingual infants do not learn equally well as the mono-lingual infants, because no naming effect is found in this group. They do not make the correct name-object pairings regardless of mode of presentation. Possibly, this difference can be explained by the multilingualism of these infants. Future studies need to investigate this with a larger group of participant. It might also be interesting to study whether multilingual infants still perform differently on word learning tasks in a later stage of word learning, e.g. at the age of 24 months.
References
Byers-Heinlein, K., Fennell, C. T., & Werker, J. F. (2012). The development of associative word learning in monolingual and bilingual infants. Bilingualism: Language and Cognition, 16(01), 198–205. doi:10.1017/S1366728912000417
Fernald, A., & Hurtado, N. (2006). Names in frames: infants interpret words in sentence frames faster than words in isolation. Developmental Science, 9(3), F33–40.
doi:10.1111/j.1467-7687.2006.00482.x
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(2), 157–178. doi:10.1080/15475440801922131
Smith, L., & Yu, C. (2008). Infants rapidly learn word-referent mappings via cross-situational statistics. Cognition, 106(3), 1558–68. doi:10.1016/j.cognition.2007.06.010
Trehub, S. E., & Shenfield, T. (2007). Acquisition of early words from single-word and sentential contexts. Developmental Science, 10(2), 190–8. doi:10.1111/j.1467-7687.2007.00545.x Waxman, S. R., & Lidz, J. L. (2006). Early Word Learning. In W. Damon & R. M. Lerner (Eds.),
Handbook of Child Psychology (6th ed.). Hoboken, NJ, USA: John Wiley & Sons, Inc. doi:10.1002/9780470147658.chpsy0207
Yu, C., & Smith, L. B. (2010). What you learn is what you see: using eye movements to study infant cross-situational word learning. Developmental Science, 14(2), 165–180. doi:10.1111/j.1467-7687.2010.00958.x
21
