Dual language input and its impact on lexical development 

Get Complete Project Material File(s) Now! »

Early language discrimination

It has been argued that early language discrimination is an essential step to successful bilingual language acquisition (Bosch & Sebastian-Galles, 1997; Nazzi, Bertoncini, & Mehler, 1998; Werker et al., 2009). Indeed, if throughout their development infants failed to ever notice the presence of two languages in their input, they could end up learning a hybrid language composed of a mix of their linguistic properties. For a long time, this was thought to be the case, leading some researchers to propose that young bilinguals had a uni ed language system for the rst two or three years of life, and that languages were only gradually discriminated later on (Leopold, 1970; Redlinger & Park, 1980; Volterra & Taeschner, 1978). However, an increasing body of evidence has emerged over the past 30 years, drawing a di erent picture; while young bilingual children may sometimes mix both languages in their production, they are generally capable of discerning them and learning distinct properties of each language (Genesee, 1989; Hammer et al., 2014; Werker & Byers-Heinlein, 2008). Exactly how young bilinguals discover their two languages is not yet fully understood, but experimental evidence suggests that some perceptual sensitivities available from early on may help them to solve this task.
Humans are born with great sensitivity to acoustic properties of speech (for a review, see Gervain and Werker, 2008). Mehler et al. (1988) were the rst to investigate whether some of these early perceptual abilities would allow newborn infants to detect di erences between languages. Using a high-amplitude sucking procedure (Jusczyk, 1985), they habituated French newborns to either French or Russian phrases spoken by a pro cient bilingual. After habituation, half of the infants heard new utterances spoken in the same language (control condition), and the other half heard utterances spoken by the same speaker in the other language (switch condition). Their results show an increase in newborns’ attention during the test phase (as indicated by a rise in their sucking rate) which was signi cantly larger after a language switch than in the control condition, suggesting that they successfully detected the language change. Additionally, they observed an asymmetry in their dishabituation patterns in-dicating a preference for their native language, French. In a subsequent experiment, the researchers exposed newborns to a low-pass ltered version of the same utterances, showing again evidence of dis-crimination (Mehler et al., 1988). These results suggested for the rst time that prosodic information, such as rhythm and intonation, are salient properties that newborn infants can use to discriminate between languages.
A number of follow-up studies using various language pairs, experimental paradigms and speech manipulations (e.g., natural, ltered, or resynthesized speech) have since replicated and extended these original results (Byers-Heinlein, Burns, & Werker, 2010; Mehler & Christophe, 1995; Mehler et al., 1988; Moon, Cooper, & Fifer, 1993; Nazzi et al., 1998; Ramus, 2002b). A summary of these studies is shown in Table 1.1. Taken together, the available evidence indicates that newborns, regardless of the language heard in the womb, are generally able to discriminate language pairs that di er in their rhythmic properties, such as French and Russian, while they fail at discriminating close language pairs, such as English and Dutch, or Spanish and Catalan. Furthermore, the discrimination of distant language pairs remains intact when using ltered speech or resynthesized speech in which only prosody and some phonotactic properties are preserved. This ability has also been observed in newborns whose mothers spoke two languages during gestation (Byers-Heinlein et al., 2010).
These results have been interpreted as evidence for an innate rhythmic-based discrimination of lan-guages in agreement with some theories of language rhythm typology (Ladefoged, 1975; Ramus, Nes-por, & Mehler, 1999). According to the rhythmic class hypothesis, languages can be classi ed into three main groups depending on their characteristic rhythmic patterns: syllable-timed (including many Ro-mance languages, such as French, Spanish and Italian), stress-timed (including Germanic languages such as English, Dutch and German) and mora-timed (including languages such as Japanese and Tamil). Indeed, experimental data from newborns seems to re ect a rhythmic classi cation. However, it is not entirely clear whether we can conclude from these experiments that infants use rhythm to classify the languages in their input, for several reasons. First, (to the best of our knowledge) the only evidence showing that newborns fail at discriminating within-class language pairs comes from studies that have used ltered or resynthesized speech (indeed, these studies were meant to test new-borns’ sensitivity to rhythmic properties); it thus remains to be proven that they cannot discriminate them when hearing natural, un ltered speech. Second, the fact that infants can discriminate certain language pairs when hearing ltered speech does not necessarily mean that rhythm is su cient for language discrimination. Ramus (2002b) made an interesting observation when testing newborns on their discrimination of Dutch and Japanese, two languages belonging to distinct rhythmic classes. When presented with un ltered speech from multiple monolingual speakers, infants failed to detect the language change, while they succeeded when presented with resynthesized speech. Ramus argued that the presence of multiple speakers – in contrast with a single bilingual speaker as in Mehler et al.’s (1988) study – introduced additional variability to the signal, hiding the relevant contrast. This experiment suggests that variability in other acoustic dimensions besides prosody may compete for infants’ attention, eventually blurring otherwise salient distinctions.
This leads us to a third point, which is the fact that most experiments so far have controlled in one form or another the amount of variability in the experimental stimuli, either by using a bilingual speaker, by ltering or resynthesizing the signal, or by simply selecting utterances that were matched in length and number of syllables. These types of manipulations are common in infant experimental research, and are necessary to isolate the factors of interest. However, these cases might end up being far removed from the real-world situations in which infants are immersed, where speech comes from several speakers, some of which may be monolinguals and some not, where phrases come in all shapes and sizes, and where speakers’ mood, speech register and even ambient noise may a ect the way language sounds to the infant’s ear. Further research is thus needed to understand how language and speaker distance, as well as other sources of variability, are re ected in early perceptual representations of speech. All that being said, the studies mentioned above have shown strong evidence that newborns possess sensitivity to rhythmic properties of languages, which might be used as a stepping stone for language discrimination.
Further research with infants aged 2- to 5-months-old has shown that, as infants grow up and be-come tuned to their native language, discrimination becomes more sensitive but at the same time dependent on infants’ familiarity with one (or both) of the languages (Bahrick & Pickens, 1988; Bosch & Sebastian-Galles, 1997, 2001; Christophe & Morton, 1998; Dehaene-Lambertz & Houston, 1998; Mehler et al., 1988; Molnar, Gervain, & Carreiras, 2014; Nazzi, Jusczyk, & Johnson, 2000). That is, infants may fail at discriminating a language pair, even across rhythmic class, if they are unfamiliar with both languages. For instance, American 2-month-olds fail at discriminating French from Rus-sian, but succeed at discriminating English from Italian (Mehler et al., 1988). On the other hand, infants newly developed sensitivity to their native language allows them to discriminate it from other languages within the same rhythmic class, an ability that was not observed in newborns (Bosch & Sebastian-Galles, 2001; Nazzi et al., 2000). For example, Bosch and Sebastian-Galles (2001) showed that Spanish and Catalan 4-month-old monolinguals could discriminate these two languages. Most interestingly, they found that Spanish-Catalan bilinguals also succeed in discriminating their two lan-guages. Thus, before the end of their rst semester, infants seem to have accumulated su cient knowledge about their native language(s) to allow discrimination, with bilinguals being no exception. This distinction might emerge naturally as young bilinguals learn the regularities of their input. As pointed out by Byers-Heinlein (2014), \English words are likely to be adjacent to other English words. They are composed of English sounds, follow English phonotactic rules, take English morphological endings, and are heard in sentences with English prosody ». These regularities could thus eventually produce clusters in the infants’ mental representations. This idea is supported by PRIMIR (Curtin, Byers-Heinlein, & Werker, 2011; Werker & Curtin, 2005), a theoretical framework that describes how infants (both monolinguals and bilinguals) learn language from their input using a combination of perceptual biases, general learning mechanisms and directed attention. However, no explicit account of how this process may unfold has yet been proposed.
Besides acoustic and linguistic cues to language discrimination, infants may additionally be sensitive to other sources of information available in their environment. One such cue is visual information (Sebastian-Galles, Albareda-Castellot, Weikum, & Werker, 2012; Weikum et al., 2007). Weikum et al. (2007) studied infants’ ability to discriminate languages by the facial gestures produced while talking. Young monolingual English and bilingual French-English infants were exposed to silent videos of bilingual speakers talking in one of these two languages. After habituation, infants saw new videos of the same speakers either talking in the same language as in habituation, or talking in the other language. They found that while monolinguals discriminated the languages at 4 and 6 months old, they failed at 8 months old. On the other hand, bilingual infants still succeeded in the task at the latest age, indicating that they had retained their sensitivity to this cue for longer than their monolingual peers. Similar results were observed in Spanish, Catalan and Spanish-Catalan 8-month-olds tested on the same French and English stimuli (Sebastian-Galles et al., 2012), suggesting that bilingual’ ability to discriminate languages visually did not depend exclusively on familiarity with one of the two languages. Like facial gestures, bilingual infants might be able to detect and exploit many other sources of information that happen to co-occur with each language, such as contextual and social cues. For instance, if speakers in the bilingual’s environment adopt a one-person-one-language approach, infants may be able to detect this regularity. We will further discuss the role of environmental cues in bilingual language acquisition in Chapters 2 and 3.

Phonological development

It is by now well established that babies are born with a language-general sensitivity to speech sounds. Evidence from a vast amount of studies on monolingual infants shows that, throughout their rst year of life, infants’ perception becomes tuned to the sounds of their native language. That is, they gradually perfect their ability to recognise the phonemes (i.e., the sound categories) that are used to distinguish words, while they simultaneously lose sensitivity to contrasts that are unused in their language (for a review, see Maurer and Werker, 2014). Vowel contrasts begin to stabilize around 6 months of age, while consonants take slightly longer, with most contrasts being in place by the end of the rst year of life. Phonological development not only involves learning the sounds of one’s native language, but also the rules that dictate how these sounds can be combined to form syllables and words { i.e., its phonotactics { and how the pronunciation of certain sounds or words may change depending on the phonological context. In monolinguals, sensitivity to the phonotactic rules of their native language emerges around the age of 9 months (Jusczyk, Luce, & Charles-Luce, 1994). Other phonological rules may take longer to acquire, such as the French liaison, which may take years to master (Chevrot, Dugua, & Fayol, 2009). Research on phonological development in children growing up with two languages is much more scarce, with only a handful of studies exploring phonological perception, principally in infants and toddlers, and another handful focusing on phonological production, mainly in preschoolers. In the following subsections we will review both lines of research.

Phonological perception in bilinguals

Evidence on phonological perception suggests that, like monolinguals, bilingual infants tune to the phonemic categories of their native languages during the rst year of life (for a review, see Werker, 2012 and Hammer et al., 2014). However, their developmental trajectories are not yet well understood, with studies showing di erent patterns depending on the experimental method, the phonological contrast, and the language pair. Such is the case of the previously discussed vowel contrast /e/-/E/. Bosch and Sebastian-Galles (2003b) rst investigated the perception of this contrast in di erent age groups of Spanish-Catalan bilingual infants, as well as in Spanish and Catalan monolinguals, using a familiarization-preference procedure. Their results con rmed, rst of all, that at 4 months all three groups of infants could discriminate this contrast, regardless of the language they were exposed to. Furthermore, by the time they were 8 months old, both monolingual groups had learnt their native categories, that is, Catalan infants maintained their discrimination of /e/-/E/, while Spanish monolinguals (for whom this contrast is not phonemic) had already lost sensitivity to it. Interestingly, bilingual infants who had been exposed to both languages since birth (and hence to both phonological categories) failed at discriminating these vowels at 8 months, but succeeded again at 12-months-old, suggesting that their phonological development follows a \U-shaped » pattern: they go from language-general perception, to a temporary collapse of the two categories, and nally a recovery of the phonemic distinction. A similar U-shaped acquisition has been attested in Spanish-Catalan bilinguals for the consonant contrast /s/-/z/, which also exists in Catalan but not in Spanish (Bosch & Sebastian-Galles, 2003a), and the /o/-/u/ vowel contrast, which exists in both languages (Bosch & Sebastian-Galles, 2005), but not in a more distant vowel contrast, /e/-/u/ (Bosch & Sebastian-Galles, 2005).
A subsequent study revisited the /e/-/E/ contrast with a di erent experimental paradigm (an an-ticipatory looking task), and found evidence that Spanish-Catalan bilingual infants can discriminate this contrast at 8-months-old (Albareda-Castellot, Pons, & Sebastian-Galles, 2011), suggesting that speci c task demands may in uence their ability to attend to this contrast. Moreover, Sundara and Scutellaro (2011) studied Spanish-English bilinguals on this same contrast, which also exists in English. They showed, even using a similar experimental paradigm to the one used by Bosch and Sebastian-Galles (2003b), that Spanish-English bilingual 8-month-olds had retained this distinction. The authors suggested that a possible explanation for Spanish-English bilinguals’ success (while Spanish-Catalan infants had failed at this age) is that Spanish and English are easier to discriminate due to the fact that they belong to two di erent rhythmic classes, as previously discussed in Section 1.2. Thus, Spanish-English infants’ ability to sort out their input by language may have facilitated the separation of these phonemic categories.
Other contrasts seem to be less problematic. For instance, French-English bilingual infants retain their discrimination of a dental vs alveolar contrast distinguishing the French and English realisations of /d/, while monolingual French and English infants lose it by 10 months of age (Sundara, Polka, & Molnar, 2008). In another study with French-English bilinguals, Burns, Yoshida, Hill, and Werker (2007) explored infants’ ability to discriminate /b/ from /p/. An interesting property of this contrast is that the voice onset time (VOT) boundary separating these two categories is di erent in French and English, thus leaving an intermediate range of VOT values in which French adults usually perceive a /p/, while English adults hear a /b/. Three age groups (6-8 months, 10-12 months and 14-20 months) of French-English bilingual and English monolingual infants were tested on their discrimination of this intermediate category against samples from the two extremes of the /b/-/p/ spectrum. Their results showed that, while both monolinguals and bilinguals behaved similarly in the youngest age group, by 10 to 12-months-old each group had tuned to their native language(s). That is, monolingual English infants treated the intermediate category as a /b/ (i.e., they discriminated it only against /p/), while bilinguals discriminated both contrasts, thus showing phonological knowledge of both of their languages. Bilinguals’ success in these contrasts may have been due to the fact that their target languages (French and English) belong to di erent rhythmic classes, but more evidence of similar contrasts across di erent language pairs would be needed before we could understand the role of language distance on phonological development.
Some recent studies have begun to explore the e ects of the speci c language pair in a more systematic way, but are still scarce (Havy, Bouchon, & Nazzi, 2016; Liu & Kager, 2015). For instance, Liu and Kager (2015) explored the perception of this same /b/-/p/ intermediate contrast in bilingual infants learning Dutch and an additional language which either shares the same VOT values as Dutch for these two consonants (French, Spanish), or has a di erent VOT boundary (Chinese, English, German). Their results showed that the perception of these contrasts depended on the language pair (with those sharing similar realisations of these consonants showing a stable contrast discrimination at 11-months-old, although with noisy results at 8-9 months), as well as on their language dominance, a ecting principally those infants learning a language pair with di ering realisations of these sounds. More studies of this kind are needed in order to form a full picture of the developmental trajectory of phonemic categories in young bilinguals.
Beyond the development of sound categories, and past the rst two years of life, not much is known about young bilinguals’ phonological perception. Indeed, a language’s phonological system is not limited to its composing sounds. Languages may di er in the possible combinations of phonemes within a syllable – that is, their phonotactics – as well as on phonological rules that may alter the surface form of a word when pronounced within a sentence. Sebastian-Galles and Bosch (2002) explored Spanish-Catalan bilinguals’ perception of Catalan phonotactic rules. Ten-month-old bilinguals, as well as Spanish and Catalan monolinguals, were tested on their discrimination of legal versus illegal consonant clusters. Their results showed that Catalan-dominant bilinguals, as Catalan monolinguals, succeeded in discriminating these two types of clusters, while Spanish-dominant bilinguals and Spanish monolinguals failed. To the best of our knowledge, no other study has explored the perception of phonological rules in young bilinguals, leaving a big gap in our understanding of their phonological development.

Phonological production in bilinguals

Many studies on phonological production have focused on preschoolers (Fabiano-Smith & Barlow, 2010; Fabiano-Smith & Goldstein, 2010; Fabiano-Smith, Oglivie, Maiefski, & Schertz, 2015; Goldstein, Fabiano, & Washington, 2005; Goldstein & Washington, 2001; MacLeod & Fabiano-Smith, 2015; Munro, Ball, Muller, Duckworth, & Lyddy, 2005; Nicoladis & Paradis, 2011), while some others have looked at infants and toddlers, although most often through case studies (Deuchar & Clark, 1996; C. E. Johnson & Lancaster, 1998; Kehoe, 2002; Kehoe, Lleo, & Rakow, 2004; Maneva & Genesee, 2002; Paradis, 2001; Schnitzer & Krasinski, 1994, 1996). In a study of a French-English bilingual infant, Maneva and Genesee (2002) found evidence of some language-speci c phonological features in the infant’s babbling, depending on the language of the parent that the child was interacting with. For instance, the child produced more stop + vowel syllables when interacting with his English-speaking parent, and more approximant + vowel syllables when interacting with his French-speaking parent. This suggests that language di erentiation may take place from the onset of speech production.
Once children start producing words, most speech production studies have focused on the development of phonemic categories. Some have conducted longitudinal studies on a small number of children, analysing their natural productions (e.g., C. E. Johnson and Lancaster, 1998; Kehoe, 2002; Kehoe et al., 2004; Schnitzer and Krasinski, 1994, 1996). For instance, Schnitzer and Krasinski (1994, 1996) analysed the productions of two children acquiring Spanish and English, and found evidence of a merged consonant system in one of them, but not in the other. Others have used elicited speech or picture naming tasks to collect production data in speci c age groups (e.g., Fabiano-Smith and Goldstein, 2010; Fabiano-Smith et al., 2015; Goldstein et al., 2005; Goldstein and Washington, 2001). For example, Goldstein and Washington (2001) used a picture naming task to assess the phonological production of Spanish-English bilingual preschoolers. They observed that half of the children had a complete consonant repertoire in at least one of their two languages, and all of them produced the full vowel repertoire of both languages. Furthermore, they found some evidence (although rare) of cross-linguistic e ects re ected in consonant substitutions. In a similar study with 3 to 4-year-old Spanish-English bilinguals, Fabiano-Smith and Goldstein (2010) found that their phonological accuracy in word production was higher in sounds that were shared by both languages. Overall, the available evidence from production studies suggests that bilingual children begin to develop phonological categories of both of their languages from early on, yet their dual phonological systems may sometimes interact.
Furthermore, a couple of studies have analysed other phonological properties of young bilinguals’ productions. Paradis (2001) studied French-English 2-year-olds patterns of truncation during a nonce-word repetition task. Truncations (e.g., nana for \banana ») are typical of toddlers’ productions and are known to be in uenced by language-speci c word-prosodic properties (Allen & Hawkins, 1980). While the truncation patterns of the bilingual toddlers in each of their languages were found to generally match those of the corresponding monolingual French and English toddlers, there was also some evidence of cross-linguistic transfer. Thus, their phonological systems appeared to be di erentiated, yet not fully independent. In another study using a picture naming task, Nicoladis and Paradis (2011) explored French-English bilingual children’s production of liaison, a complex phonological rule in French that causes word- nal silent consonants to be pronounced before a vowel-initial word (e.g., petit [p@ti] \small » is pronounced [p@tit] in petit ours \small bear »). They found that, in general, young children’s French vocabulary | but not their age -{ correlated positively with their production of liaison. Interestingly, when matched by vocabulary, bilinguals applied liaison less often than their monolingual peers’, but only in low-frequency collocation frames. However, the sample size of this matched comparison was very small (6 monolinguals and 6 bilinguals between the ages of 3 and 5), making it di cult to draw conclusions.
In summary, while current evidence from perception and production studies suggests that young bilinguals start acquiring phonological properties of both of their languages from early on, their de-velopment does not seem to be equivalent to that of two monolinguals in one. Complex patterns of phonological acquisition emerge depending on the speci c language pairs and phonological properties that BFL learners set out to discover, sometimes showing a short-lived delay in the development of a phonological feature of one of the two languages, and sometimes showing cross-linguistic in uences. However, the number of studies conducted so far remains very limited, leaving many open questions regarding their phonological development.

Table of contents :

1 Introduction 
1.1 Dening bilingualism in early childhood
1.1.1 The challenge of bilingual rst language acquisition
1.2 Early language discrimination
1.3 Phonological development
1.3.1 Phonological perception in bilinguals
1.3.2 Phonological production in bilinguals
1.4 Lexical development
1.4.1 Bilinguals’ milestones and learning strategies
1.4.2 Vocabulary size and composition in bilinguals
1.5 Thesis overview
2 Modelling early language discrimination 
2.1 Introduction
2.2 I-vector pipeline
2.3 Computational experiments
2.3.1 Experiment 1: A proof of concept (Article 1)
2.3.2 Experiment 2: Generalizing to other language pairs
2.3.3 Experiment 3: Language discrimination with ltered speech
2.3.4 Experiment 4: Language discrimination across speakers
2.3.5 Experiment 5: The role of the background model
2.3.6 Experiment 6: The impact of a bilingual background (Article 2)
2.4 General discussion and future directions
2.A Appendix A: Pipeline details
2.A.1 Feature extraction
2.A.2 Background Model
2.A.3 Total Variability space
2.A.4 I-vector extraction
2.B Appendix B: Expectation-Maximization algorithm for GMMs
2.C Appendix C: P-values from simulations in Exp. 4
3 Dual language input and its impact on lexical development 
3.1 Introduction
3.2 Methods
3.2.1 Subjects
3.2.2 Materials and procedures
3.2.3 Coding and pre-processing
3.3 Results and discussion
3.3.1 Language Environment Questionnaire
3.3.2 Language Diaries
3.3.3 Vocabulary scores
3.3.4 Eects of Language Exposure on Vocabulary Scores
3.4 Discussion
3.A Appendix A: Language Diary sample page
3.B Appendix B: Language Environment Questionnaire
3.C Appendix C: Comparison of weighting options
3.D Appendix D: Correlation results for indirect input measures
4 Perception of language-specic phonological rules 
4.1 Article
4.2 Additional studies
4.2.1 Pilot experiment 1
4.2.2 Pilot experiment 2
4.2.3 Pilot experiment 3
4.2.4 Pilot experiment 4
4.2.5 Pilot experiment 5
4.2.6 General discussion
5 General discussion 
5.1 Language separation in the rst year of life
5.2 Separation of phonological rules
5.3 Conclusion
Bibliography

GET THE COMPLETE PROJECT