LingLunch : Mikhail Kissine

Jeudi 12 Novembre 2020, 12:00 to 13:00
Philip Miller

Mikhail Kissine (ULB)
Why is autism important for linguistic theory?

Autism Spectrum Disorder is characterised by early-onset and lifelong socio-pragmatic deficits; furthermore, in around 60% of autistic children language development is significantly delayed and around 25% of individuals on the spectrum remain non-verbal. In that sense, data from autism underscore the importance of inter-subjective communication for the emergence of language. However,  while early-onset socio-pragmatic deficits likely cause language delays, a very different explanation may be needed to understand how verbal autistics eventually do acquire language. Early engagement in inter-subjective communicative interaction is clearly decisive for (typical) language acquisition, but there is also evidence that at least some autistic individuals learn language in a non-interactive way. Non-interactional language learning thus represents a crucial test case for the nativism/constructionism debate in linguistics. Many linguistic theories include usage-based components that are couched in terms of domain-general, statistical mechanisms, independent of inter-subjective communication, or, for that matter, of language. However, all these theories require some kind of upstream constraint on such distributional learning: nativist models posit an innate universal grammar, but those who reject nativism usually ground usage-based linguistic learning in cooperative communication. Assuming that the core linguistic faculty is intact and that there is sufficient linguistic input, nativism clearly allows for language learning in autism to unfold in a non-interactive way. By contrast, in order to accommodate data from autism, constructionist models have to show that the weight of statistical, associative learning processes in language learning is sufficient to dispense with the idea of any kind of innate mechanism specific to language. In the case of autistic language learning, such a deflationary position would require that, without any socio-communicative bootstrap and without any innate specifically linguistic predisposition, language acquisition be modeled as a fully unsupervised statistical learning, in which most of the lowest-level parsing of the acoustic stream occurs without any kind of top-down guidance.