EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Do Language Models Know Who Did What to Whom?

Joseph M Denning iD, Xiaohan Hannah Guo iD, Bryor Snefjella, Idan A Blank iD

Open Mind · 07.07.2026 · 10 · 923-950

DOI: 10.1162/opmi.a.365 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Open Mind, online erschienen am 07.07.2026. Verfasst von Joseph M Denning und 3 weiteren Autor:innen. Beteiligte Einrichtung: Department of Psychology, University of California, Los Angeles, Los Angeles, CA, USA. Die Arbeit nennt 158 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Language models (LMs) are commonly criticized for not “understanding” language. However, many critiques focus on cognitive abilities that, in humans, are distinct from language processing. Here, we instead study a kind of understanding tightly linked to language: inferring “who did what to whom” (thematic roles) in a sentence. Does the central training objective of LMs—word prediction—result in sentence representations that capture thematic roles? In two experiments, we characterized sentence representations in four LMs that have been proposed as models of human language processing. The overall representational similarity of sentence pairs did not reflect whether they had the same agent/patient assignments or opposite agent/patient assignments. Furthermore, we found limited evidence that thematic role information was available in any subspace of hidden activations. However, some attention heads robustly captured thematic roles, independently of syntax. Therefore, LMs can extract thematic roles but this information influences their representations weakly.

Schlagwörter

Stichwörter der Autor:innen

comprehensionlanguage modelsrepresentational similaritythematic roles

Autor:innen und Institutionen

  • Department of Psychology, University of California, Los Angeles, Los Angeles, CA, USA.
  • Department of Psychology, University of Chicago, Chicago, IL, USA.

ORCID-Profile: Joseph M Denning, Xiaohan Hannah Guo, Idan A Blank

Zitieren

APA

Denning, J. M., Guo, X. H., Snefjella, B., & Blank, I. A. (2026). Do Language Models Know Who Did What to Whom?. Open Mind, 10, 923-950. https://doi.org/10.1162/opmi.a.365

Vancouver

Denning JM, Guo XH, Snefjella B, Blank IA. Do Language Models Know Who Did What to Whom?. Open Mind. 2026;10:923-950. doi:10.1162/opmi.a.365

BibTeX
@article{Denning2026,
  author = {Denning, Joseph M and Guo, Xiaohan Hannah and Snefjella, Bryor and Blank, Idan A},
  title = {Do Language Models Know Who Did What to Whom?},
  journal = {Open Mind},
  year = {2026},
  volume = {10},
  pages = {923--950},
  doi = {10.1162/opmi.a.365}
}
Kostenlos starten →