PhD thesis

When does a word end and another one begin?

Language acquisition is one of the most complex tasks that humans face. The difficulty of this task can be appreciated if we consider, as adults, how puzzling it is to hear a conversation in a totally unknown foreign language. Since speech is a continuous flow of sounds, swe have no idea how many words have been spoken. Spoken words are rarely produced in isolation, and contrary to written words, they are not separated by silences but are more commonly integrated within sentences. How can young children succeed at extracting words from the running speech? How do they tackle this challenging task and build their lexicon? 

From early on, infants have been shown to use different strategies to successfully identify word boundaries. Their experiences are crucial, since segmentation cues are complex and language dependent. With Rushen Shi, I have investigated how French-learning infants and adults resolve lexical ambiguities when word boundaries are unclear due to a potential liaison. 

Part of my work has investigated how learners deal with conflicting cues to word boundaries when encountering novel words in liaison contexts. I found that, before the age of two, infants are guided by a syllabic-alignment bias, interpreting the liaison consonant as the onset of a novel word (e.g. gros /z/onches "big onches"-> infants misinterpret 'zonche' as the word; Babineau & Shi, 2014; 2011). Soon after, sensitivity to the distributional patterns of liaison develops; enabling infants to parse the correct form from variable liaison contexts (e.g., parse the novel word onche after hearing it in various contexts such as un /n/onche, petit /t/onche, ses /z/onches). When exposed to regular minimal pair words, 24-month-olds do not parse a vowel-initial form, e.g. when the variable consonants surfacing as onsets are not linked to liaison (e.g. guonche, vonche, chonche) or when the syntactic context does not trigger these particular liaison consonants (e.g. un zonche, gros nonche, premier tonche; Babineau, Emond, & Shi, 2023). This shows that through extensive exposure to liaison, and with the help of a growing lexicon, infants become aware of the fact that a small group of liaison consonants surface in front of vowel-initial words (e.g. zarbre/narbre for “tree”, zours/nours for “bear”). These consonants gain a special status. Yet, children have a careful approach: they take into account the larger context in which these special consonants appear when they interpret the phonological form of novel words. 

When do children master liaison? This is a hard question, as there are a lot of interindividual variations. Despite of this, we have found that the process of building a representation for a new word is more advanced around 30 months of age, when toddlers can clearly take into account its syntactic context (Babineau & Shi, 2013; 2016). Around the age of three, toddlers have more precise expectations regarding the surfacing of certain liaison consonants (e.g. the singular determiner un "a" has an underlying /n/). Nonetheless, revisiting the phonological representation of words takes time. In Babineau, Legrand & Shi (2021), we found that 30-month-old French-learning children’s lexicons contain multiple variants for familiar words subject to liaison, especially frequent variants (e.g., the word elephant can be accessed through different forms such as éléphant, téléphant, zéléphant). Liaison-onset variants, which were most likely encoded at an early age, remain in their lexicon for some time. 

What about adults? Can they tell apart homophonous sequences linked to liaison (e.g., un /n/air "a melody" vs un nerf "a nerve") or interpret novel words correctly? In Babineau, Shi and Achim (2017;  see also Babineau & Shi, 2013), we found that segmentation of liaison ambiguous phrases remains challenging for adult native speakers of French, who are influenced by both the syntactic context (determiner vs adjective) as well as the frequency of the onset consonant as a liaison.