Construction Grammar (CxG) is a paradigm from cognitive linguistics that emphasises the connection between syntax and semantics. Rather than rules that operate on lexical items, it posits constructions as the central building blocks of language, i.e., linguistic units of different granularity that combine syntax and semantics. When combined with novel probing methodology, this theory enables us to ask interesting questions about the performance, and holes therein, of large language models. In this talk, we will briefly introduce Construction Grammar for an NLP audience. We then report on the methodology and results of several empirical studies focusing on language models’ behaviour on specific constructions, exemplifying how Construction Grammar can be used to shine a light on properties and current challenges of language models. We also outline how larger, more generalisable conclusions could be drawn, by utilising existing resources from linguistics, combined with further support from the NLP community.
Leonie Weissweiler is a PhD student in Computational Linguistics at LMU Munich, working with Hinrich Schütze, as well as with Lori Levin and David Mortensen at CMU. Her research interests are applying cognitively plausible theories of linguistics to NLP, as well as unsupervised crosslingual computational morphosyntax.