Differences in language learnability across ages – LangAge

To achieve our goals, we rely on an interdisciplinary approach in which we combine corpora studies to describe the input, computational modelling to study the learning mechanisms and effects of learning biases in perfectly controlled situations, and psychological laboratory experiments on both young children and adults to assess whether the putative learning mechanisms and biases we attribute to learners are actually in operation.
Corpus data consist of cross-linguistic and socio-culturally varied corpora gathered using unobtrusive daylong recorders, which have already been transcribed by native listeners of each language. This allows for a unique glimpse onto children’s natural language environment, which typically includes both speech directed to them, and speech directed to others around them, including adults. We describe and compare the input to children and adults, and use machine learning to explore learnability of phonological rules and morphosyntactically-defined words, as well as study the impact of a specific property of child-directed speech, its ‘burstiness’ or repetitivity, on learnability.
Our laboratory experiments use a variety of methods, including high-variability auditory training (to assess adults’ learning of L2 sound categories), implicit and explicit teaching of novel words and novel morphosyntactic contexts (to compare toddlers’ and adults’ acquisition mechanisms), and artificial language learning paradigms (to compare toddlers’ and adults’ learning biases).
These are techniques with which we have ample experience, and all necessary resources are in place. Our adult experimental platform contains five sound-attenuated testing booths, equipped with an eye-tracker. Our babylab contains three sound-attenuated testing booths, equipped with 3 eye-trackers, and set-ups for several behavioral techniques (habituation, preferential looking, anticipatory looking). We also have contacts with local kindergartens, where we can test three- to six-year-old children individually in a small room, as well as with day nurseries, where we can test younger children (12-36 months). For this purpose, depending on experimental protocols we use a portable eye-tracker (EyeLink 1000), a tablet with a touchscreen (iPads ®), or we simply present stimuli on a portable computer (with the children’s pointing or oral responses being video-recorded for off-line blind coding).

Until now (Sept. 2022), many results have already been obtained. Some highlights :

- Children as young as 20 months of age can learn the co-occurrence relationships between known content words and novel function words (ko, ka, in e.g. ‘ko pig’, ‘ko turtle’, ‘ka book’, ‘ka bottle’), then exploit these to infer the probable meaning of novel content words (e.g. ‘ko bamoule’: ‘bamoule’ is more likely to refer to a novel animal, while in ‘ka doripe’, ‘doripe’ is more likely to refer to an inanimate object).

- Explicit teaching of vocabulary translations (e.g., 'neko' means cat) affects learning differently at different ages and literacy levels: direct translations are only useful for learning once the learner is literate (for adults and literate 5-6-year-olds, but not for pre-literate 5-6-year-olds). However, all learners feel more confident when they learn through direct translations (although their performance does not benefit from it for pre-literate 5-6-year-olds).

- Infants aged months understand negative sentences.This ability to understand negative sentences so early might support language acquisition, providing infants with a tool to constrain the space of possibilities for word meanings.

- An analysis of a French corpus of infant-directed speech shows that an optional phonological rule (i.e. word-final liquid deletion, as in ‘tab’ carrée’ for ‘table carrée’ «square table«), is applied in more than half of the possible cases, and that pairs of tokens of the same word with and without the rule application tend to cluster together (with a median distance of 49 seconds of speech). This clustering could be a powerful cue to acquire the rule.

- Auditory training enhances adult L2 learners’ perception of non-native sounds not only at the prelexical but also the lexical level, thus providing real-life benefit for word recognition.

- A phonetically natural rule of vowel harmony is easier to learn (by adults) than its unnatural disharmony counterpart.

- A meta-analysis of behavioral observations suggests that children in urban, industrialized communities are talked to 3 times more than children growing up in rural, small-scale ones, which leads to the prediction that there should be vast differences in outcomes across such cultures according to most, but not all, theories.

- Citizen scientists can help us extract useful information from long recordings (this result was highlighted by CNRS (https://insb.cnrs.fr/fr/cnrsinfo/lorsque-les-bebes-beneficient-de-la-sagesse-de-la-foule).

This project compares children's native language acquisition and adults' second language acquisition, and thus its main impact is in the domain of cognitive psychology. In addition, the project has a potential longer-term societal impact, in the following two directions. First, a better understanding of the normal course of early language development has implications for our understanding of pathological development. Specifically, it may enable researchers to develop more appropriate diagnostic tools that will allow for detection of abnormalities of language development at the earliest possible stage, as well as specific remediation tools. Second, the project includes studies on adult language learning strategies, which may lead to the development of improved teaching techniques for second-language learners.

Until now (Sept. 2022), this project has given rise to more than 30 articles in peer-reviewed international journals, including Developmental Science, Cognition, Infancy, Phonology, PLOS One, et Psychological Science.

Some examples :

Babineau, M., de Carvalho, A., Trueswell, J. & Christophe, A. (2021). Familiar words can serve as a semantic seed for syntactic bootstrapping. Developmental Science, 24, e13010.

de Carvalho, A., Dautriche, I., Fiévet, A.-C. & Christophe A. (2021). Toddlers exploit referential and syntactic cues to flexibly adapt their interpretation of novel verb meanings. Journal of Experimental Child Psychology, 203, 105017.

Havron, N., Scaff, C., Carbajal, J., Linzen, T., Barrault, A. & Christophe, A. (2020). Priming syntactic ambiguity resolution in children and adults. Language Cognition and Neuroscience, 35, 1445-1455.

Babineau, M., Shi, R. & Christophe, A. (2020). 14-month-olds exploit verbs’ syntactic contexts to build expectations about novel words. Infancy, 25, 719-733.

Carbajal, J. & S. Peperkamp (2020). Dual language input and the impact of language separation on early lexical development. Infancy 25, 22-45.

Martin, A. & S. Peperkamp (2020). Phonetically natural rules benefit from a learning bias: a reexamination of vowel harmony versus disharmony. Phonology 37, 65-90.

Melnik, G. & S. Peperkamp (2021). High-variability phonetic training enhances second language lexical processing: evidence from online training of French learners of English. Bilingualism: Language and Cognition, 24, 497-506.

Lavechin, M., de Seyssel, M., Gautheron, L., Dupoux, E., & Cristia, A. (2021). Reverse-engineering language acquisition with child-centered long-form recordings. Annual Review of Linguistics 8, 389-407.

Cristia, A., (2022). A systematic review suggests marked differences in the prevalence of infant-directed vocalization across groups of populations. Developmental Science e13265.

While almost all children master at least one language effortlessly over the course of a few years, adults’ substantial efforts to learn a second language typically lead to less-than-perfect mastery. It seems reasonable to postulate that infants and adults have the same overall architecture for their language system. What, then, can account for the obvious differences in the final outcome of the learning process?. As language acquisition unfolds, learners apply certain processing mechanisms onto their language input, in ways that are constrained by their linguistic and cognitive biases. Logically, differences between children and adults may stem from all three of these aspects of language acquisition, i.e. 1) the input, 2) the learning mechanisms, and 3) the linguistic and cognitive biases. In each of these domains, we will compare adults and young children, in order to identify differences that may impact the end-result of the language acquisition process. In some instances, specific areas of divergence can be identified from the previous literature (e.g. the acquisition of phonetic categories); in others we offer novel avenues of research to test potential differences between infants and adults, with specific hypotheses for why they may or may not diverge (e.g. comparing the burstiness of child- vs adult-directed speech).
To achieve our goals, we will rely on an interdisciplinary approach in which we combine corpora studies to describe the input, computational modeling to study the learning mechanisms and effects of learning biases in perfectly controlled situations, and psychological laboratory experiments on both young children and adults to assess whether the putative learning mechanisms and biases we attribute to learners are actually in operation. We will study the acquisition of multiple linguistic levels, and whenever possible, their interactions, ranging from the sound system to morphosyntax, through the lexicon.
Overall, this project examines language acquisition, both in young children acquiring their first language, and in adults acquiring a second language, and will inform fundamental research on language learnability and acquisition. In addition, the project has a potential longer-term societal impact, in two directions: a deeper understanding of first language acquisition for early diagnosis and remediation, and improved methods for second or foreign language teaching to adults.