Abstract
In this work, we show the pre-trained language models return distinguishable generation probability and uncertainty distribution to unfaithfully hallucinated texts, regardless of their size and structure. By examining 24 models on 6 data sets, we find out that 88-98% of cases return statistically significantly distinguishable generation probability and uncertainty distributions. Using this general phenomenon, we showcase a hallucination-reducing training algorithm. Our algorithm outperforms other baselines by achieving higher faithfulness metrics while maintaining sound general text quality measures.- Anthology ID:
- 2024.findings-emnlp.738
- Volume:
- Findings of the Association for Computational Linguistics: EMNLP 2024
- Month:
- November
- Year:
- 2024
- Address:
- Miami, Florida, USA
- Editors:
- Yaser Al-Onaizan, Mohit Bansal, Yun-Nung Chen
- Venue:
- Findings
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 12630–12639
- Language:
- URL:
- https://preview.aclanthology.org/add_missing_videos/2024.findings-emnlp.738/
- DOI:
- 10.18653/v1/2024.findings-emnlp.738
- Cite (ACL):
- Taehun Cha and Donghun Lee. 2024. Pre-trained Language Models Return Distinguishable Probability Distributions to Unfaithfully Hallucinated Texts. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 12630–12639, Miami, Florida, USA. Association for Computational Linguistics.
- Cite (Informal):
- Pre-trained Language Models Return Distinguishable Probability Distributions to Unfaithfully Hallucinated Texts (Cha & Lee, Findings 2024)
- PDF:
- https://preview.aclanthology.org/add_missing_videos/2024.findings-emnlp.738.pdf