On Faithfulness and Factuality in Abstractive Summarization

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1906–1919 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 1906 On Faithfulness and Factuality in Abstractive Summarization Joshua Maynez∗ Shashi Narayan∗ Bernd Bohnet Ryan McDonald Google Research {joshuahm,shashinarayan,bohnetbd,ryanmcd}@google.com Abstract It is well known that the standard likelihood training and approximate decoding objectives in neural text generation models lead to less human-like responses for open-ended tasks such as language modeling and story gener- ation. In this paper we have analyzed limi- tations of these models for abstractive docu- ment summarization and found that these mod- els are highly prone to hallucinate content that is unfaithful to the input document. We con- ducted a large scale human evaluation of sev- eral neural abstractive summarization systems to better understand the types of hallucinations they produce. Our human annotators found substantial amounts of hallucinated content in all model generated summaries. However, our analysis does show that pretrained models are better summarizers not only in terms of raw metrics, i.e., ROUGE, but also in generating faithful and factual summaries as evaluated by humans. Furthermore, we show that textual en- tailment measures better correlate with faith- fulness than standard metrics, potentially lead- ing the way to automatic evaluation metrics as well as training and decoding criteria.1 1 Introduction Current state of the art conditional text generation models accomplish a high level of fluency and co- herence, mostly thanks to advances in sequence- to-sequence architectures with attention and copy (Sutskever et al., 2014; Bahdanau et al., 2015; Gu et al., 2016), fully attention-based Transformer ar- chitectures (Vaswani et al., 2017; Dai et al., 2019) and more recently pretrained language modeling for natural language understanding (Devlin et al., 2019; Radford et al., 2018; Yang et al., 2019; Liu et al., 2019). There has been a growing interest in ∗ The first two authors contributed equally. 1Our human annotated summaries for faithfulness and fac- tuality will be released at https://github.com/google-research- datasets/xsum hallucination annotations. understanding how maximum likelihood training and approximate beam-search decoding in these models lead to less human-like text in open-ended text generation such as language modeling and story generation (Holtzman et al., 2020; Welleck et al., 2020; See et al., 2019). In this paper we investigate how these models are prone to gener- ate hallucinated text in conditional text generation, specifically, extreme abstractive document summa- rization (Narayan et al., 2018a). Document summarization — the task of produc- ing a shorter version of a document while preserv- ing its information content (Mani, 2001; Nenkova and McKeown, 2011) — requires models to gener- ate text that is not only human-like but also faith- ful and/or factual given the document. The exam- ple in Figure 1 illustrates that the faithfulness and factuality are yet to be conquered by conditional text generators. The article describes an event of “Conservative MP Zac Smith winning the pri- mary for 2016 London mayoral election”, but sum- maries often forge entities (e.g., “Nigel Goldsmith” or “Zac Goldwin”) or information (e.g., …

Content truncated to 500 words