Incomplete Utterance Rewriting as Semantic Segmentation

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 2846–2857, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 2846 Incomplete Utterance Rewriting as Semantic Segmentation Qian Liu†∗, Bei Chen§, Jian-Guang Lou§, Bin Zhou†, Dongmei Zhang§ †School of Computer Science and Engineering, Beihang University, China §Microsoft Research, Beijing, China †{qian.liu, zhoubin}@buaa.edu.cn; §{beichen, jlou, dongmeiz}@microsoft.com Abstract Recent years the task of incomplete utterance rewriting has raised a large attention. Previ- ous works usually shape it as a machine trans- lation task and employ sequence to sequence based architecture with copy mechanism. In this paper, we present a novel and extensive approach, which formulates it as a semantic segmentation task. Instead of generating from scratch, such a formulation introduces edit op- erations and shapes the problem as prediction of a word-level edit matrix. Benefiting from being able to capture both local and global information, our approach achieves state-of- the-art performance on several public datasets. Furthermore, our approach is four times faster than the standard approach in inference. 1 Introduction A dramatic progress has been achieved in single- turn dialogue modeling such as open-domain re- sponse generation (Shang et al., 2015), question answering (Rajpurkar et al., 2016), etc. By con- trast, multi-turn dialogue modeling is still in its infancy, as users tend to use incomplete utter- ances which usually omit or refer back to enti- ties or concepts appeared in the dialogue context, namely ellipsis and coreference. According to previous studies, ellipsis and coreference exist in more than 70% of the utterances (Su et al., 2019), for which a dialogue system must be equipped with the ability of understanding them. To tackle the problem, early works include learning a hier- archical representation (Serban et al., 2017; Zhang et al., 2018) and concatenating the dialogue ut- terances selectively (Yan et al., 2016). Recently, researchers focus on a more explicit and explain- able solution: the task of Incomplete Utterance Rewriting (IUR, also known as context rewriting) (Kumar and Joshi, 2016; Su et al., 2019; Liu et al., ∗Work done during an internship at Microsoft Research. Turn Utterance (Translation) x1 (A) 北京今天天气如何 How is the weather in Beijing today x2 (B) 北京今天是阴天 Beijing is cloudy today x3 (A) 为什么总是这样 Why is always this x∗ 3 北京为什么总是阴天 Why is Beijing always cloudy Table 1: An example dialogue between user A and B, including the context utterances (x1, x2), the incom- plete utterance (x3) and the rewritten utterance (x∗ 3). 2019a; Pan et al., 2019; Elgohary et al., 2019; Zhou et al., 2019). IUR aims to rewrite an incom- plete utterance into an utterance which is seman- tically equivalent but self-contained to be under- stood without context. As shown in Table 1, the incomplete utterance x3 not only omits the subject “北京”(Beijing), but also refers to the semantic of “阴天”(cloudy) via “这样”(this). By explicitly recovering the hidden semantics behind x3 into x∗ 3, IUR makes the downstream dialogue model- ing more precise. To deal with IUR, a natural idea is to trans- fer models from …

Content truncated to 500 words