Event Date: Wednesday, 14 September, 2022
Location: 11:00-12:00, Sala Riunioni, Pal. Venera (II piano)
Speaker: Prof. Philippe Blache (CNRS & Aix-Marseille Université)
Title: Compositionality, unification and the good-enough perspective
Abstract: Many different works address the question of knowing whether language processing is compositional or not. As Dowty underlines, asking this question comes to “express skepticism about a yes answer” (Dowty, 2006). He also explains that because of language creativity, 1/ there is necessarily a finitely characterizable procedure for accessing to the meaning, 2/ because sentences can only be enumerated via derivations in the grammar then the procedure for interpreting them must be determined by their syntactic structures as generated by this grammar. If we can only agree with the first part of the argument, the second is questionable. First, usage-based theories as well as works on spoken language syntax have shown the invalidity of a strict derivational view of grammaticality (in other words, it is not always possible to build a complete syntactic structure). Second, multimodality studies show that the information can be spread over different domains other than syntax. Finally, explaining how to interpret a linguistic message must go beyond the limits of the sentence, and take into account the Principle of Contextuality. As a consequence, it is necessary to propose a more general architecture explaining both the prototypical situation (syntactically driven) and the one where the meaning is not transparent, the syntactic structure is not complete, etc.
We propose in this talk to approach the question of comprehension (and then the access to the meaning) trying to bring together the compositional (strictly syntactic-based) and non-compositional (direct access to the meaning) into a unique framework. This architecture is based on three basic mechanisms: prediction, delay and unification. It relies on the hypothesis that by default, we understand a linguistic message in a shallow and “good-enough” manner. In this framework, the meaning representation is frame-based. Building the meaning consists in instantiating incrementally frame slots by unification, each slot value (possibly atomic or complex) being built in a direct or compositional manner. During this talk, this architecture will be put in perspective with some neurolinguistic models.
Philippe Blache is Director of Research at the CNRS and a member of the Laboratoire Parole et Langage (CNRS Aix-Marseille University). His works focus on how language is processed by humans and machines. He is more precisely interested in studying the question of how humans process language in a natural context, typically a conversation. He has developed a new linguistic theory making it possible to formalize mutlimodality as well as the specificities of natural interaction. This theory is called Property Grammars. Since recently, he is using these approach to study the neural basis of language processing architecture.