CIDL25 Workshop: Do LLMs exhibit natural-language processing cognitive abilities?

We invite submissions to a workshop to be held within the 25th International Conference of the Department of Linguistics of the Faculty of Letters.

CIDL25 Workshop: Do LLMs exhibit natural-language processing cognitive abilities?

Relevant topics: Cognitive Science; Computational Linguistics; General Linguistics; Semantics; Syntax

Questions we aim to explore within the ambit of these topics:

• Do LLMs have metalinguistic abilities (do they have the ability to generate analyses of language data/theoretical linguistic abilities of language samples so as to identify whether a sentence is syntactically ambiguous? How well do they handle linguistic recursion tasks? Can they identify what type of recursion a sentence contains (adjectival, possessive, PP…)? Can they draw a syntactic tree for it and add more
layers of recursion?

• Do LLMs meet plausible criteria associated with linguistic metasemantic theories or mental metasemantic theories (cf. https://link.springer.com/article/10.1007/s11229-024-04723-8)?

• What would it mean for a Large Language Model to understand or acquire a language? What would it mean for them to meaningfully use a language or words? Is understanding language or grasping the meaning of words based on an innate structure? Does that structure need to be biological? Do LLMs need sensory grounding for language understanding?

• Are LLMs outputs based on linguistic representations? Do LLMs have internal representations? What would that mean? What kind of representations would that mean? If they do, then do they acquire them? Can connectionism account for it?

• Can LLMs actually tell us something about Universal Grammar? Are there alternative ways for acquiring language? Can LLMs learn new languages via identification of grammar rules within data patterns?

Please see the associated call for abstracts!