Human Guided Exploitation of Interpretable Attention Patterns in Summarization and Topic Segmentation

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 10189 - 10204 December 7-11, 2022 ©2022 Association for Computational Linguistics Human Guided Exploitation of Interpretable Attention Patterns in Summarization and Topic Segmentation Raymond Li†, Wen Xiao†, Linzi Xing†, Lanjun Wang‡∗, Gabriel Murray§, Giuseppe Carenini† †University of British Columbia, Vancouver, BC, Canada ‡Tianjin University, Tianjin, China §University of Fraser Valley, Abbotsford, BC, Canada {raymondl, xiaowen3, lzxing, carenini}@cs.ubc.ca wanglanjun@tju.edu.cn gabriel.murray@ufv.ca Abstract The multi-head self-attention mechanism of the transformer model has been thoroughly in- vestigated recently. In one vein of study, re- searchers are interested in understanding why and how transformers work. In another vein, researchers propose new attention augmenta- tion methods to make transformers more accu- rate, efficient and interpretable. In this paper, we combine these two lines of research in a human-in-the-loop pipeline to first discover im- portant task-specific attention patterns. Then those patterns are injected, not only to smaller models, but also to the original model. The benefits of our pipeline and discovered pat- terns are demonstrated in two case studies with extractive summarization and topic segmenta- tion. After discovering interpretable patterns in BERT-based models fine-tuned for the two downstream tasks, experiments indicate that when we inject the patterns into attention heads, the models show considerable improvements in accuracy and efficiency. 1 Introduction With transformer-based models (Vaswani et al., 2017) dominating the leaderboard for many key NLP tasks such as summarization (Liu and Lapata, 2019), topic segmentation (Lukasik et al., 2020), and sentiment analysis (Adhikari et al., 2019), their core multi-head self-attention mechanism has also been thoroughly investigated. In particular, to ex- plain why and how transformers work, researchers have analyzed the learnt self-attention matrices of trained transformer models (e.g., Raganato and Tiedemann (2018); Kovaleva et al. (2019)), with Vig and Belinkov (2019) for instance, exploring the attention patterns in BERT (Devlin et al., 2019) and GPT-2 (Radford et al., 2019), as well as analyzing their alignment with syntax. Meanwhile, a parallel line of research has ex- plored injecting predefined patterns into attention ∗Corresponding author. matrices of transformers in an attempt to reduce the run-time complexity of self-attention while main- taining competitive accuracy. This can be done by either replacing the attention weights with a fixed matrix (Raganato et al., 2020; Tay et al., 2021; Xiao et al., 2020); or alternatively by guiding the attention weights through more flexible masking strategies (Mihaylov and Frank, 2019; Child et al., 2019; Guo et al., 2019; Li et al., 2019; Beltagy et al., 2020; Zaheer et al., 2020; Bai et al., 2021). In this work, we propose and test a novel human- in-the-loop pipeline that to the best of our knowl- edge is the first attempt to combine research on analyzing self-attention with work on injecting pat- terns into attention matrices. To start, human users visually explore the attention matrices of transform- ers to identify task-specific patterns that could be formalized as a predicate. After quantitatively eval- uating the patterns on the validation set, they can be injected into …

Content truncated to 500 words