1 Running Head: Bilingual Toddlers’ Lexicon Vocabulary of 2-Year-Olds Learning English and an Additional Language: Norms and Effects of Linguistic Distance Caroline Floccia (1) Thomas D. Sambrook (2) Claire Delle Luche (3) Rosa Kwok (4) Jeremy Goslin (1) Laurence White (1) Allegra Cattani (1) Emily Sullivan (1) Kirsten Abbot-Smith (5) Andrea Krott (6) Debbie Mills (7) Caroline Rowland (8) Judit Gervain (9) Kim Plunkett (10) (1) University of Plymouth, (2) University of East Anglia, (3) University of Essex, (4) University of Coventry, (5) University of Kent, (6) University of Birmingham, (7) University of Bangor, (8) Max Planck Institute for Psycholinguistics & University of Liverpool, (9) Laboratoire Psychologie de la Perception, UMR 8242 CNRS & Université Paris Descartes, Sorbonne Paris Cité, Paris, (10) University of Oxford 2 Abstract The majority of the world’s children grow up learning two or more languages. The study of early bilingualism is central to current psycholinguistics, offering insights into issues such as transfer and interference in development. From an applied perspective, it poses a universal challenge to language assessment practices throughout childhood, as typically-developing bilingual children usually underperform relative to monolingual norms when assessed in one language only. We measured vocabulary with Communicative Development Inventories for 372 24-month-old toddlers learning British English and one Additional Language out of a diverse set of 13 (Bengali, Cantonese, Dutch, French, German, Greek, Hindi-Urdu, Italian, Mandarin, Polish, Portuguese, Spanish and Welsh). We furthered theoretical understanding of bilingual development by showing, for the first time, that linguistic distance between the child’s two languages predicts vocabulary outcome, with phonological overlap related to expressive vocabulary, and word order typology and morphological complexity related to receptive vocabulary, in the Additional Language. Our study also has crucial clinical implications: we have developed the first bilingual norms for expressive and receptive vocabulary for 24-month-olds learning British English and an Additional Language. These norms were derived from factors identified as uniquely predicting CDI vocabulary measures: the relative amount of English versus the Additional Language in child-directed input and parental overheard speech, and infant gender. The resulting UKBTAT tool was able to accurately predict the English vocabulary of an additional group of 58 bilinguals learning an Additional Language outside our target range. This offers a pragmatic method for the assessment of children in the majority language when no tool exists in the Additional Language. Our findings also suggest that the effect of linguistic distance might extend beyond bilinguals’ acquisition of early vocabulary to encompass broader cognitive processes, and could constitute a key factor in the study of the debated bilingual advantage. 3 Chapter 1 – Introduction Caroline Floccia, Thomas D. Sambrook, Claire Delle Luche, Rosa Kwok, Jeremy Goslin, Laurence White, Allegra Cattani, Emily Sullivan, Kirsten Abbot-Smith, Andrea Krott, Debbie Mills, Caroline Rowland, Judit Gervain, Kim Plunkett Corresponding author: Caroline Floccia, [email protected] Parents eagerly await the moment their one-year-old infant produces their first word, and professionals working with young children track the appearance of the two-word combination stage at around age 2. To developmental psycholinguists, these two milestones in healthy language acquisition are supported by an impressive range of achievements, from the attuning of perceptual abilities (Werker, Yeung, & Yoshida, 2012), the development of word segmentation skills (Bergmann & Cristia, 2016) and the retrieval of word meaning (Stevens, Gleitman, Trueswell, & Yang, 2017), to the discovery of syntactic, morphological and conversational rules (Gleason & Ratner, 2017; Hoff, 2013). The complexity of the task that young children naturally solve places the study of language development at the heart of the debate about the nature of the human mind (Pinker, 1995; Tomasello, 2009). In addition, the co-occurrence - and interdependence - of language development and that of other domains such as motor coordination (Iverson, 2010), object perception (Jones & Smith, 2005), or social skills (Carpenter, Nagell, Tomasello, Butterworth, & Moore, 1998), indicates that the study of language learning is fundamental to the understanding of child development as a whole. Even more impressive than an infant acquiring her maternal language is an infant learning two or more languages, an achievement of the majority of the world’s children (Kohnert, 2010), including an increasing proportion in the United Kingdom (UK) (17.5% in primary schools; NALDIC, 2012). Although knowledge about language development is mostly built around the monolingual child model, the study of early bilingualism has become a central issue in psycholinguistics. This is because it allows us to address theoretical key 4 questions such as specificity, differentiation, transfer and interference in development, but also because, from an applied perspective, it poses a universal challenge to language assessment practices throughout childhood. Indeed, language development in bilinguals is notoriously difficult to predict, due to a variety of situational factors related to the proportion and properties of dual-language exposure, making the use of monolingual norms largely inadequate for this population (Cattani et al., 2014). This is further complicated by the variety of language pairs being spoken in the world. For example, in the United States (US), whilst Spanish constitutes the most common other-than-English language spoken at home (62% speakers), the remaining 38% speakers share up to at least 350 languages, mostly in large metropolitan areas (United States Census 2020). Given the mounting evidence that children develop their two languages in separate, yet interfering, systems (Hoff, 2013), the degree of influence between two languages should theoretically depend upon the linguistic distance between the two languages. The central aim of this paper is to investigate the role of linguistic distance in early language development, and through this, we address two key questions: first, we bring new theoretical knowledge by directly testing hypotheses regarding the existence of language-to- language influence in early bilingual development; second, we offer a pragmatic solution to the widespread problem of how to assess bilingual toddlers’ language skills when no tools are available in the home language. Although the study was conducted in the UK, the theoretical conclusions we reached about the role of linguistic distance are not community-specific, and the principles behind the assessment tool we have designed can be exported to any other community. We collected data about vocabulary development and contextual variables in a cohort of 372 bilingual two-year-old toddlers learning British English and one of 13 Additional Languages. We investigated how vocabulary development is modulated by linguistic distance 5 between British English and the Additional Language, bringing new evidence about language-to-language influence in early acquisition, and the first demonstration to date of an effect of structural similarity between languages on early vocabulary development (Study 1). In addition, we generated a vocabulary model able to predict the lexicon size of 2-year-old bilinguals in English and one the 13 target Additional Languages, which fed into a new assessment tool, the UKBTAT (UK Bilingual Toddler Assessment Tool) (Study 2). Finally we showed that this model could reliably predict the English scores of a new cohort of 58 British-English 2-year-olds learning a new, non-target Additional Language (Study 3). In what follows, we review the literature pertaining to the theoretical (Study 1) and applied contributions of this work (Studies 2 and 3). In Chapter 2 we present the methods for the cohort data collection, followed by the analyses pertaining to Study 1 in Chapter 3, Studies 2 and 3 in Chapter 4, ending with Chapter 5 where we discuss how linguistic distance shapes language development and the limiting factors of this work. Study 1: Understanding the Role of Linguistic Distance in Bilingual Development Most bilingual studies are conducted with homogeneous samples, with Spanish- American English bilinguals constituting the largest cohort (e.g., Marchman, Fernald, & Hurtado, 2010; Place & Hoff, 2011), followed by Canadian English-French (e.g., Paradis, Crago, Genesee, & Rice, 2003), Barcelona Catalan-Spanish (e.g., Bosch & Sebastián-Gallés, 2001) and Welsh-British English (e.g,. Gathercole & Thomas, 2009). Although it is often reasonable to generalise results obtained with one language pairing to different language pairs, is it safe to assume that bilingual children from different backgrounds are confronted with the same linguistic problems to solve, or that they are able to solve them within the same learning span? There is growing evidence that, according to the language they acquire, monolinguals’ learning paths can differ, including for early lexical prosodic processing (Adam & Bat-El, 6 2009), word segmentation (across dialects: Floccia et al., 2016; Nazzi, Mersad, Sundara, Iakimova, & Polka, 2014), and phonological processing (Nazzi, Floccia, Moquet, & Butler, 2009; Mani & Plunkett, 2007; Delle Luche, Floccia, Granjon, & Nazzi, 2016; Bouchon, Floccia, Fux, Adda‐Decker, & Nazzi, 2015), culminating in differences in vocabulary growth (Bleses et al., 2008; Thordardottir, 2005). Bilingualism is likely to exacerbate these language- specific differences, adding not only a new language but also the complexity of interactions between languages. Why should we expect interactions between the two language systems in a bilingual child? In adult bilinguals, evidence for automatic cross-language activation in visual word recognition (Lauro & Schwartz, 2017), and in spoken word recognition (Mercier, Pivneva, & Titone, 2014; Spivey & Marian, 1999) strongly point to the interdependence of the two language systems (see French & Jacquet, 2004). A common position nowadays, following the original proposal of a primary, undifferentiated language system (Volterra & Taeschner, 1978), is that early bilinguals develop two independent language systems (Genesee, 1989; Genesee, Nicoladis, & Paradis, 1995). However, accumulating evidence suggest early influences from one language system onto the other, mirroring the structure of the adult system. Interaction between the two language systems is perhaps most obvious in production, where intra-sentential code-mixing, i.e., including elements of each language at the sentence level, is frequently observed in toddlers at the phonological, lexical and morphosyntactic level (e.g., Gildersleeve-Neumann, Kester, Davis, & Peña, 2008; Paradis & Genesee, 1996). More convincing however is non-selective lexical access in comprehension, that is, the fact that speech presented in one language activates word recognition in the two languages, as demonstrated in adults (e.g., Dijkstra, 2005), and observed at least as young as from three years of age (e.g., Poulin-Dubois, Bialystok, Blaye, Polonia, & Yott, 2013; Von Holzen & Mani, 2012; see the review by DeAnda, Poulin-Dubois, Zesiger, & Friend, 2016). However, 7 even if there were an initial separation of lexicons, bilinguals would still demonstrate language-specific differences from the parallel learning of two language systems. For example, French infants rely more on consonants than vowels for lexical processing from the age of 8 months (Nishibayashi & Nazzi, 2016), whereas British English learners process consonants and vowels equally (Mani & Plunkett, 2007; Floccia, Nazzi, Delle Luche, Poltrock, & Goslin, 2014), at least until the age of 30 months (Nazzi et al., 2009), and Danish children rely more on vowels than consonants at 20 months of age (Højen & Nazzi, 2016). How do such differences translate to the case of bilingual learners? Will language-specific routes for vowel-consonant processing be delayed until the onset of separate language processing, or will one linguistic strategy be adopted, at an efficiency cost to the other language? In sum, each language pairing will necessarily produce a different linguistic learning problem for bilingual infants to solve, which is likely to translate to variable delays and/or adapted pathways (see Polka, Orena, Sundara, & Worrall, 2017, for word segmentation outcomes differing in bilingual and monolingual 8-month-olds). In this project we conducted the first systematic evaluation of the impact of differences between languages, as measured through metrics of linguistic distance, on vocabulary acquisition in both British English and the Additional Language. Measuring linguistic distance. As adults, second language learning seems easier if the language is intuitively similar to our own (e.g., English/German vs English/Cantonese), which is supported by studies in second language learning for both adults and school-aged children (e.g., Lado, 1957; Lindgren & Muñoz, 2012; Van der Slik, 2010). For example Lindgren and Muñoz (2012) showed that a cognate-based measure of language distance is the most important predictive factor for formal second language learning in schools, above differences in the exposure of the languages at home. These results support the idea that in 8 second-language learning, the knowledge and structure of L1, i.e. the maternal language, provide scaffolding for the acquisition of L2, i.e. the Additional Language (see also the literature on cross-linguistic transfers in second language reading acquisition, e.g., Genesee, Geva, Dressler, & Kamil, 2006). However, in early simultaneous bilinguals the effects of language distance are more complex as the languages are acquired in parallel from infancy. While similarities between the two languages may reinforce phonological, lexical and syntactic acquisition across the two languages, it would also reduce the perceptual separation between languages. Knowledge of the interaction between reinforcement and separation is crucial to our understanding of bilingual acquisition, but the complexity of multi-dimensional representations of language means that unitary measures of the seemingly intuitive notion of ‘linguistic distance’ are difficult to evaluate. This complexity is reflected in the many distance metrics which have been proposed, including cognate distance (e.g., Dyen, Kruskal, & Black, 1992), genetic linguistic distance (Harding & Sokal, 1988; Ruhlen, 1987), phonetic distance (Nerbonne et al., 1996), distance in terms of linguistic rhythm (Ling, Grabe, & Nolan, 2000; Ramus, Nespor & Mehler, 1999) and second language learnability (Chiswick & Miller, 2005). Of all linguistic distance measures, cognate distance is probably the metric that has the widest currency, at least at the lexical level. Traditionally, this refers to the proportion of translation equivalents sharing common historical origins, such as lait in French and leche in Spanish (milk, sharing the Latin root lac). In an influential cognate database (e.g., Dyen et al., 1992, adapted by McMahon & McMahon, 2005), the index of linguistic cognate distance is obtained from the compilation of 200 frequent culture-neutral words in 84 Indo-European languages and dialects, and for each language pairing. However, the definition of cognates in Dyen et al.’s database makes it difficult to generalise to languages without a clear common history. Approaches based on automatic methods have been proposed to refine the definition 9 of cognates, for example by using intra-language similarity (Ellison & Kirby, 2006) or cross- language orthographic similarity measures (Serva & Petroni, 2008). While some of these metrics may be suitable for the adult speech environment, child-directed speech differs in lexical, prosodic and pragmatic content from adult-directed speech (e.g., Thiessen, Hill, & Saffran, 2005). Importantly, infants do not share adults’ meta-linguistic and orthographic knowledge. Given the young age of our participants it was necessary to base our distance metrics on a set of child-familiar basal words and to consider phonetic, phonological and metrical similarities rather than historical origins or orthographic properties. To this end we developed a measure of linguistic distance which focussed upon corpora related to toddlers’ speech environment. We used the Oxford Communicative Developmental Inventory of Oxford CDI (Hamilton et al., 2000), which supplied us with a list of words that should be known to British English children of our target age group. Here a short introduction to the CDIs is necessary, as we will use these tools not only to develop a metric of phonological overlap, but also to collect vocabulary data from toddlers. The first Communicative Developmental Inventories, which are parental reports of their children’s vocabulary on a checklist of familiar words, were developed for American English children (Fenson et al, 1994), with norms published in two separate forms for different age ranges (8-18 months for the Words and Gestures form, and 16-30 months for the Words and Sentences form). An updated norming sample of 2,550 US children was produced later for the two forms (MacArthur-Bates CDI, Fenson et al., 2007). Crucially, the normed CDI parent reports have been adapted (not translated) in a multitude of languages, with the purpose of mirroring the structure of the reference language as much as possible. The availability of CDIs in many languages has created new opportunities for cross-linguistic studies of language development (see e.g., CLEX database now called Wordbank: Jørgensen, 10 Dale, Bleses, & Fenson, 2009; Frank, Braginsky, Yurovsky, & Marchman, 2017) and for bilingual studies (e.g., Armon-Lotem, & Ohana, 2017; Cattani et al., 2014; Gatt, 2017; Pearson, Fernandez, & Oller, 1993; O’Toole et al., 2017). The reliability and validity of CDIs is long established for use in research (Mancilla-Martinez, Gamez, Vagh, & Lesaux, 2016; Marchman, Thal, Dale, & Reznick, 2006) and for clinical assessment (e.g., Charman, Drew, Baird, & Baird, 2003; Heilmann, Weismer, Evans, & Hollar, 2005; Thal, DesJardin, & Eisenberg, 2007). To produce vocabularies for the 13 target Additional Languages, we could have used the words listed in each language-specific CDI as a proxy of toddler vocabulary for each Additional Language. However, these CDI vocabularies would largely reflect the cultural and physical environment in which the language was predominantly spoken. As our bilinguals all live in the UK it is likely that both their British English and Additional Language vocabulary would reflect a British English environment, rather than the environment of the monolingual Additional Language CDI. As such, we believe that the bilingual Additional Language lexicon would be better represented by translation equivalents of the words of the Oxford CDI. This approach also has the advantage that it is unaffected by variations in the methodologies used to construct Additional Language CDIs, which can result in wide differences in CDI word counts. Details of these toddler-centric metrics can be found in the method section (Chapter 2). Note that the use of a common CDI inventory across all language groups was restricted to the calculation of a metric of phonological overlap, and not to collect vocabulary knowledge from toddlers, which was performed with language-specific inventories. With this first metric based on lexical phonological overlap, we expected phonological similarity between languages to facilitate the acquisition of words in each language. This hypothesis is supported by results from Bosch and Ramon-Casas (2014)