Home | Publications | Wei24

Computational Approaches to Construction Grammar and Morphology

MCML Authors

Leonie Weissweiler

* Former Member

→ Group Hinrich Schütze
Computational Linguistics

Abstract

This thesis investigates how well modern pretrained language models (PLMs) capture human linguistic capabilities, using novel methods from Linguistics to probe rare and less compositional phenomena. It evaluates PLMs’ understanding of constructions and finds that while they learn syntactic structure fairly well, they struggle with non-compositional meaning. The work also shows that PLMs generalize morphology in human-like ways but still make systematic errors and rely on different mechanisms. Finally, it explores how NLP can support Linguistics by developing human-in-the-loop and hybrid annotation pipelines for construction grammar and by using multilingual corpora to induce morphological structure for low-resource languages. (Shortened).

phdthesis Wei24