Strong hallucinations from negation and how to fix them

Swarnadeep Bhar, Nicholas Asher


Abstract
Despite great performance on many tasks, language models (LMs) still struggle with reasoning, sometimes providing responses that cannot possibly be true because they stem from logical incoherence. We call such responses strong hallucinations and prove that they follow from an LM’s computation of its internal representations for logical operators and outputs from those representations. Focusing on negation, we provide a novel solution in which negation is treated not as another element of a latent representation, but as an operation over an LM’s latent representations that constrains how they may evolve. We show that our approach improves model performance in cloze prompting and natural language inference tasks with negation without requiring training on sparse negative data.
Anthology ID:
2024.findings-acl.752
Volume:
Findings of the Association for Computational Linguistics: ACL 2024
Month:
August
Year:
2024
Address:
Bangkok, Thailand
Editors:
Lun-Wei Ku, Andre Martins, Vivek Srikumar
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
12670–12687
Language:
URL:
https://aclanthology.org/2024.findings-acl.752
DOI:
10.18653/v1/2024.findings-acl.752
Bibkey:
Cite (ACL):
Swarnadeep Bhar and Nicholas Asher. 2024. Strong hallucinations from negation and how to fix them. In Findings of the Association for Computational Linguistics: ACL 2024, pages 12670–12687, Bangkok, Thailand. Association for Computational Linguistics.
Cite (Informal):
Strong hallucinations from negation and how to fix them (Bhar & Asher, Findings 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.findings-acl.752.pdf