Summary of Mitigating Hallucination in Abstractive Summarization with Domain-conditional Mutual Information, by Kyubyung Chae et al.
Mitigating Hallucination in Abstractive Summarization with Domain-Conditional Mutual Information
by Kyubyung Chae, Jaepill Choi, Yohan Jo, Taesup Kim
First submitted to arxiv on: 15 Apr 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes a novel decoding strategy to alleviate the issue of hallucination in abstractive summarization, where models generate plausible but absent text. The strategy, domain-conditional pointwise mutual information (DCPMI), adjusts generation probabilities based on token marginal probabilities within the source text’s domain. By using this approach, the model demonstrates improved faithfulness and source relevance on the XSUM dataset. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper aims to solve a problem in summarization models that sometimes make up details not present in the original text. The researchers came up with a new way to generate text that is more accurate and relevant to the topic. They tested their method on a specific dataset called XSUM and found it works better than before. |
Keywords
» Artificial intelligence » Hallucination » Summarization » Token
