Understanding Neural Abstractive Summarization Models via Uncertainty

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 6275–6281, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 6275 Understanding Neural Abstractive Summarization Models via Uncertainty Jiacheng Xu Shrey Desai Greg Durrett Department of Computer Science The University of Texas at Austin {jcxu,gdurrett}@cs.utexas.edu shreydesai@utexas.edu Abstract An advantage of seq2seq abstractive summa- rization models is that they generate text in a free-form manner, but this flexibility makes it difficult to interpret model behavior. In this work, we analyze summarization decoders in both blackbox and whitebox ways by studying on the entropy, or uncertainty, of the model’s token-level predictions. For two strong pre- trained models, PEGASUS (Zhang et al., 2020) and BART (Lewis et al., 2020) on two summarization datasets, we find a strong cor- relation between low prediction entropy and where the model copies tokens rather than gen- erating novel text. The decoder’s uncertainty also connects to factors like sentence position and syntactic distance between adjacent pairs of tokens, giving a sense of what factors make a context particularly selective for the model’s next output token. Finally, we study the rela- tionship of decoder uncertainty and attention behavior to understand how attention gives rise to these observed effects in the model. We show that uncertainty is a useful perspective for analyzing summarization and text genera- tion models more broadly.1 1 Introduction Recent progress in abstractive summarization has been fueled by the advent of large-scale Transform- ers pre-trained on autoregressive language mod- eling objectives (Hoang et al., 2019; Khandelwal et al., 2019; Lewis et al., 2020; Zhang et al., 2020). Despite their strong performance on automatic met- rics like ROUGE (Lin, 2004), abstractive models are not as straightforward and interpretable as their extractive counterparts. Free-form generation in these models also leads to serious downstream er- rors, such as factual inconsistencies with the input document (Cao et al., 2018; Kry´sci´nski et al., 2020; 1Code is available at https://github.com/ jiacheng-xu/text-sum-uncertainty Wang et al., 2020; Durmus et al., 2020; Goyal and Durrett, 2020). Although the interpretability of NLU models has been extensively studied (Ribeiro et al., 2016; Ghaeini et al., 2018; Jain and Wallace, 2019; Desai and Durrett, 2020), summarization models specifically have not received similar atten- tion, with analysis efforts often focused on datasets and evaluation (Kryscinski et al., 2019). In this work, we focus on interpreting and under- standing abstractive summarization models through the lens of decoder uncertainty, or the entropy of decisions during generation. While uncertainty in generation has been studied from the perspective of data (Ott et al., 2018), sampling (Fan et al., 2018; Holtzman et al., 2019), and training (Correia et al., 2019; Kang and Hashimoto, 2020), it is underuti- lized as a technique for analysis and inspection of generation systems. We study two prominent summarization models, PEGASUS (Zhang et al., 2020) and BART (Lewis et al., 2020), fine-tuned on two English summarization datasets, CNN/Daily Mail (Hermann et al., 2015) and XSum (Narayan et al., 2018), to understand model behavior in each setting. First, by comparing …

Content truncated to 500 words