Modeling German Verb Argument Structures: LSTMs vs. Humans

Charlotte Rochereau (1, 2) , Benoît Sagot (3) , Emmanuel Dupoux (1, 2)

1 LSCP - Laboratoire de sciences cognitives et psycholinguistique
2 CoML - Apprentissage machine et développement cognitif
3 ALMAnaCH - Automatic Language Modelling and ANAlysis & Computational Humanities

LSTMs have proven very successful at language modeling. However, it remains unclear to what extent they are able to capture complex morphosyntactic structures. In this paper, we examine whether LSTMs are sensitive to verb argument structures. We introduce a German grammaticality dataset in which ungrammatical sentences are constructed by manipulating case assignments (eg substituting nominative by accusative or dative). We find that LSTMs are better than chance in detecting incorrect argument structures and slightly worse than humans tested on the same dataset. Surprisingly, LSTMs are contaminated by heuristics not found in humans like a preference toward nominative noun phrases. In other respects they show human-similar results like biases for particular orders of case assignments.

Domaines: Informatique et langage [cs.CL]

Charlotte Rochereau, Benoît Sagot, Emmanuel Dupoux. Modeling German Verb Argument Structures: LSTMs vs. Humans. 2019. ⟨hal-02417640⟩