Hallucinated but Factual! Inspecting the Factuality of Hallucinations in Abstractive Summarization

Cao, Meng; Dong, Yue; Cheung, Jackie Kit

doi:10.18653/v1/2022.acl-long.236

Cited by 39 publications

(41 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Second, it is important to explore generative capabilities with qualitative metrics (Figure 2 in the appendix illustrates retrieved text and answers generated by rag-end2end and rag-original). This could be aligned with research areas like measuring factual consistency (Kryściński et al, 2019;Cao et al, 2022) and hallucinations (Cao et al, 2022;Shuster et al, 2021;Nie et al, 2019) of generative language models. Future work could explore whether updating the retriever and document embeddings during the training phase could improve factual consistency and reduce hallucinations in final generations.…”

Section: Discussionmentioning

confidence: 66%

Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering

Siriwardhana

Weerasekera

Wen

et al. 2023

Transactions of the Association for Computational Linguistics

View full text Add to dashboard Cite

Retrieval Augment Generation (RAG) is a recent advancement in Open-Domain Question Answering (ODQA). RAG has only been trained and explored with a Wikipedia-based external knowledge base and is not optimized for use in other specialized domains such as healthcare and news. In this paper, we evaluate the impact of joint training of the retriever and generator components of RAG for the task of domain adaptation in ODQA. We propose RAG-end2end, an extension to RAG that can adapt to a domain-specific knowledge base by updating all components of the external knowledge base during training. In addition, we introduce an auxiliary training signal to inject more domain-specific knowledge. This auxiliary signal forces RAG-end2end to reconstruct a given sentence by accessing the relevant information from the external knowledge base. Our novel contribution is that, unlike RAG, RAG-end2end does joint training of the retriever and generator for the end QA task and domain adaptation. We evaluate our approach with datasets from three domains: COVID-19, News, and Conversations, and achieve significant performance improvements compared to the original RAG model. Our work has been open-sourced through the HuggingFace Transformers library, attesting to our work’s credibility and technical consistency.

show abstract

Section: Discussionmentioning

confidence: 66%

Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering

Siriwardhana

Weerasekera

Wen

et al. 2023

Transactions of the Association for Computational Linguistics

View full text Add to dashboard Cite

show abstract

“…These three types of hallucinations are well-documented in literature studying generative model hallucinations [18,32,41,67]. We add to this previous literature by showing how such hallucinations occur in this reading context.…”

Section: E Factuality In Generated Summariesmentioning

confidence: 56%

“…The extent and kind of hallucinations in our summaries can tell us what risk such hallucinations pose and how much effort an expert must invest to make the summaries publishable. For example, if the majority of hallucinations are new but correct information (a common type of hallucination [18]), then they pose less of a risk and require less expert knowledge to fix than if the hallucinations instead reverse the direction of a found effect (another type of hallucination [32]). We generated summaries with no restriction on hallucinated content.…”

Section: E Factuality In Generated Summariesmentioning

confidence: 99%

Know Your Audience: The benefits and pitfalls of generating plain language summaries beyond the "general" audience

August,

Lo,

Smith

et al. 2024

Proceedings of the CHI Conference on Human Factors in Computing Systems

View full text Add to dashboard Cite

Figure 1: Simpler summaries were often the best reading experience for participants with little to no background in a scientific topic. However, readers with high topic familiarity, even those considered part of the general public (i.e., not a researcher), ignored more information in low complexity summaries while still reporting these simple summaries as equally engaging as high complexity ones. Our results provide guidance on generating plain language summaries for a wider range of general audiences.

show abstract

“…• Summaries suffer from hallucinations, i.e., information leaked to the output from the outside of source text. However, Cao et al, 2022 find that much hallucinated content is mainly consistent with world knowledge.…”

Section: Challenges Of Abstractive Summarizationmentioning

confidence: 75%

Abstractive Summarization for the Ukrainian Language: Multi-Task Learning with Hromadske.ua News Dataset

Galeshchuk

2023

Proceedings of the Second Ukrainian Natural Language Processing Workshop (UNLP)

View full text Add to dashboard Cite

Despite recent NLP developments, abstractive summarization remains a challenging task, especially in the case of low-resource languages like Ukrainian. The paper aims at improving the quality of summaries produced by mT5 for news in Ukrainian by fine-tuning the model with a mixture of summarization and text similarity tasks using summary-article and titlearticle training pairs, respectively. The proposed training set-up with small, base, and large mT5 models produce higher quality résumé. Besides, we present a new Ukrainian dataset for the abstractive summarization task that consists of circa 36.5K articles collected from Hromadske.ua until June 2021.

show abstract

Hallucinated but Factual! Inspecting the Factuality of Hallucinations in Abstractive Summarization

Cited by 39 publications

References 0 publications

Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering

Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering

Know Your Audience: The benefits and pitfalls of generating plain language summaries beyond the "general" audience

Abstractive Summarization for the Ukrainian Language: Multi-Task Learning with Hromadske.ua News Dataset

Contact Info

Product

Resources

About