Event Knowledge in Large Language Models: The Gap Between the Impossible and the Unlikely
Cognitive Science · 2023-11 · 47 (11)
DOI: 10.1111/cogs.13386 ↗At a glance
Journal article in Cognitive Science, published online on 2023-11-27. Written by Carina Kauf and 7 co-authors. Institution: Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology. Major MEDLINE topics: Language, Semantics. Cited 45 times so far (Crossref); 172 references. The full text is freely available (open access, CC BY 4.0). Funded by: European Commission.
Abstract (original language)
Word co‐occurrence patterns in language corpora contain a surprising amount of conceptual knowledge. Large language models (LLMs), trained to predict words in context, leverage these patterns to achieve impressive performance on diverse semantic tasks requiring world knowledge. An important but understudied question about LLMs’ semantic abilities is whether they acquire generalized knowledge of common events. Here, we test whether five pretrained LLMs (from 2018's BERT to 2023's MPT) assign a higher likelihood to plausible descriptions of agent−patient interactions than to minimally different implausible versions of the same event. Using three curated sets of minimal sentence pairs (total n = 1215), we found that pretrained LLMs possess substantial event knowledge, outperforming other distributional language models. In particular, they almost always assign a higher likelihood to possible versus impossible events ( The teacher bought the laptop vs. The laptop bought the teacher ). However, LLMs show less consistent preferences for likely versus unlikely events ( The nanny tutored the boy vs. The boy tutored the nanny ). In follow‐up analyses, we show that (i) LLM scores are driven by both plausibility and surface‐level sentence features, (ii) LLM scores generalize well across syntactic variants (active vs. passive constructions) but less well across semantic variants (synonymous sentences), (iii) some LLM errors mirror human judgment ambiguity, and (iv) sentence plausibility serves as an organizing dimension in internal LLM representations. Overall, our results show that important aspects of event knowledge naturally emerge from distributional linguistic patterns, but also highlight a gap between representations of possible/impossible and likely/unlikely events.
Subjects
MeSH terms (MEDLINE)
Author keywords
Authors and institutions
- Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology.
- McGovern Institute for Brain Research, Massachusetts Institute of Technology.
- Computer Science and Artificial Intelligence Lab, Massachusetts Institute of Technology.
- Department of Modern Languages, Literatures and Cultures, University of Bologna.
- Department of Chinese and Bilingual Studies, Hong Kong Polytechnic University.
- Department of Mathematics, University of Washington.
Funding
- European Commission
Cite
APA
Kauf, C., Ivanova, A. A., Rambelli, G., Chersoni, E., She, J. S., Chowdhury, Z., Fedorenko, E., & Lenci, A. (2023). Event Knowledge in Large Language Models: The Gap Between the Impossible and the Unlikely. Cognitive Science, 47(11). https://doi.org/10.1111/cogs.13386
Vancouver
Kauf C, Ivanova AA, Rambelli G, Chersoni E, She JS, Chowdhury Z, et al. Event Knowledge in Large Language Models: The Gap Between the Impossible and the Unlikely. Cognitive Science. 2023;47(11). doi:10.1111/cogs.13386
BibTeX
@article{Kauf2023,
author = {Kauf, Carina and Ivanova, Anna A and Rambelli, Giulia and Chersoni, Emmanuele and She, Jingyuan Selena and Chowdhury, Zawad and Fedorenko, Evelina and Lenci, Alessandro},
title = {Event Knowledge in Large Language Models: The Gap Between the Impossible and the Unlikely},
journal = {Cognitive Science},
year = {2023},
volume = {47},
number = {11},
doi = {10.1111/cogs.13386}
}
EUVIMED