Oct 24, 2022 · In this paper, we identify a simple criterion under which models are significantly more likely to assign more probability to hallucinated content during ...
Mutual Information Alleviates Hallucinations in Abstractive ... mutual information to increase the faithfulness of summaries in abstractive summarization.
A decoding strategy is proposed that switches to optimizing for pointwise mutual information of the source and target token—rather than purely the ...
People also ask
What are hallucinations in abstractive summarization?
What are factual hallucinations?
It has been used as a score to select the most informative or surprising sentences in extractive summarization (Padmakumar and He, 2021) or as an alternative ...
Mutual Information Alleviates Hallucinations in Abstractive Summarization. Authors: Liam van der Poel, Ryan Cotterell, Clara Meister. To install fairseq on ...
Despite recent advances in abstractive summarization, current summarization systems still suffer from content hallucinations where models generate text that is ...
Apr 15, 2024 · We introduce a decoding strategy based on domain-conditional pointwise mutual information. This strategy adjusts the generation probability of each token.
2022. pdf bib abs. Mutual Information Alleviates Hallucinations in Abstractive Summarization · Liam van der Poel | Ryan Cotterell | Clara Meister · Proceedings ...
24 References · Mutual Information Alleviates Hallucinations in Abstractive Summarization · Detecting Hallucinated Content in Conditional Neural Sequence ...
Dec 8, 2022 · Mutual Information Alleviates Hallucinations in Abstractive Summarization ... Salience Allocation as Guidance for Abstractive Summarization.