Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2409.16658 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913517362216960 |
|---|---|
| author | Cha, Taehun Lee, Donghun |
| author_facet | Cha, Taehun Lee, Donghun |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2409_16658 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Pre-trained Language Models Return Distinguishable Probability Distributions to Unfaithfully Hallucinated Texts Cha, Taehun Lee, Donghun Computation and Language 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. |
| title | Pre-trained Language Models Return Distinguishable Probability Distributions to Unfaithfully Hallucinated Texts |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2409.16658 |