One language or two? Navigating cross‐language conflict in statistical word segmentation

Abstract

Despite the prevalence of bilingualism (Crystal, 2004; Grosjean, 2010), there is limited work investigating how infants detect words in two languages. Infants with bilingual exposure may encounter conflicting prosodic, lexical, and co-occurrence cues to word boundaries across their languages, which in turn could disrupt word segmentation. For example, an infant learning both French and English would encounter opposing prosodic cues; the predominant stress pattern in English is trochaic (strong-weak), whereas French words tend to follow an iambic (weak-strong) pattern. Language-specific patterns that support segmentation within one language may interfere with segmentation in another (Polka, Orena, Sundara, & Worrall, 2017). However, effective separation of the languages' properties and efficient shifting between language contexts may support accurate word segmentation and allow infants to learn from the rich information present in dual speech streams. Early exposure to diverse regularities across languages may allow infants' language knowledge to self-organize into linguistic clusters over time (e.g. PRIMIR; Curtin, Byers-Heinlein, & Werker, 2011). This process would improve as infants learn core features that differentiate between languages (e.g. prosody, phonology, etc.; Byers-Heinlein, Burns, & Werker, 2010), allowing infants to monitor for language changes and to integrate new information by language, reducing cross-linguistic interference. This is critical, as the language in use in bilingual environments may change between, or even within, interactions and speakers, and failure to anticipate language switches can incur processing costs for both infants and adults (Byers-Heinlein, Morin-Lessard, & Lew-Williams, 2017), suggesting that change detection is important for efficient bilingual processing. Once bilinguals acquire tools to track information for each language separately, they can more effectively learn language-specific

Statistical word segmentation: Anchoring learning across contexts

“…The authors found that 8‐month‐old monolingual infants exhibited no evidence of learning sequentially presented artificial languages when they contained overlapping syllables that formed different words in each language, but infants could learn the languages when presented separately. This finding aligns with Antovich and Graf Estes' finding (2020) that monolingual infants failed to demonstrate learning of interspersed artificial languages that contained conflicting syllable patterns. As a whole, these lines of work suggest that infants' learning of sound sequences can interact across speech streams.…”

“…This aligns with the work from Lany et al. (2018) and Antovich and Graf Estes (2020), who found associations between laboratory statistical learning tasks and language processing in 15‐month‐olds and between statistical learning and bilingual input in 16‐month‐olds, respectively. These connections suggest that even early in development, there is a relationship between lab‐based statistical learning tasks and real‐world language processing abilities.…”

“…To produce the language stimuli, a female speaker recorded isolated three‐syllable sequences that represented all possible co‐articulation contexts in the language (following the method used by Antovich & Graf Estes, 2020; Graf Estes et al., 2007; Graf Estes & Lew‐Williams, 2015). She was coached to produce the syllables with pitch and stress that were as consistent as possible, to try to avoid the introduction of salient lexical stress patterns.…”

Is there a bilingual disadvantage for word segmentation? A computational modeling approach

“…A recent analysis suggests that bilingual infants exposed to more frequent language mixing (including within-sentence switching) are more successful in segmenting words out of both of their languages (Orena & Polka, 2019). Moreover, monolinguals presented with a “bilingual” input in an artificial language setting failed to segment words from those languages (Antovich & Graf Estes, 2018, 2020), suggesting that in natural language acquisition, infants may lose the capacity to keep track of two sets of cues if exposed to a predominantly consistent input; or perhaps that multilinguals gain the capacity to track multiple sets of cues when exposed to variable input (see particularly follow-up analyses on frequency of mixing, in Orena & Polka, 2019; and on proportion of bilingual speakers in the environment, in Antovich & Graf Estes, 2020). Although many of these studies report significant preferences, it should be noted that the direction of preference is sometimes toward familiarity (as in most previous word segmentation studies on monolinguals) and sometimes towards novelty (i.e., longer looking to the test stimuli with a foil than with the familiarized word).…”

“…This strategy has been studied extensively in infancy (e.g., Aslin, Saffran & Newport, 1998; Pelucchi, Hay & Saffran, 2009; a review in Lany & Saffran, 2010; Saffran, Aslin & Newport, 1996), often (but not always) using artificial languages where all other cues (including phrase breaks) have been neutralised (a meta-analysis in Black & Bergmann, 2017). Importantly, these cues are also employed by bilingual infants: who can keep track of two sets of statistics presented in interleaved utterances better than monolingual infants (Antovich & Graf Estes, 2018, 2020).…”