Language Models Are Poor Learners of Directional Inference
Li, Tianyi; Hosseini, Mohammad Javad; Weber, Sabine; u. a. (2022): Language Models Are Poor Learners of Directional Inference, in: Yoav Goldberg, Zornitsa Kozareva, Yue Zhang, u. a. (Hrsg.), Findings of the Association for Computational Linguistics: EMNLP 2022, Association for Computational Linguistics, S. 903–921, doi: 10.18653/v1/2022.findings-emnlp.64.
Details
- Author
- Li, TianyiHosseini, Mohammad JavadSteedman, Mark
- Title of the compilation
- Findings of the Association for Computational Linguistics: EMNLP 2022
- Editors
- Goldberg, YoavKozareva, ZornitsaZhang, Yue
- Conference
- EMNLP 2022, Dezember 2022, Abu Dhabi
- Publisher Information
- Association for Computational Linguistics
- Year of publication
- 2022
- Language
- English
- DOI
- Abstract
We examine LMs’ competence of directional predicate entailments by supervised fine-tuning with prompts. Our analysis shows that contrary to their apparent success on standard NLI, LMs show limited ability to learn such directional inference; moreover, existing datasets fail to test directionality, and/or are infested by artefacts that can be learnt as proxy for entailments, yielding over-optimistic results. In response, we present BoOQA (Boolean Open QA), a robust multi-lingual evaluation benchmark for directional predicate entailments, extrinsic to existing training sets. On BoOQA, we establish baselines and show evidence of existing LM-prompting models being incompetent directional entailment learners, in contrast to entailment graphs, however limited by sparsity.
- Keywords
- Language Models
- Peer Reviewed
- yes
- International Distribution
- yes
- Open Access Journal
- yes
- Type
- Conferenceobject
- URI
- Activation date
- 2024-08-12