Effective Inter-Clause Modeling for End-to-End Emotion-Cause Pair Extraction

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 302–312 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 302 Unsupervised Paraphrasing by Simulated Annealing Xianggen Liu1 Lili Mou2 Fandong Meng3 Hao Zhou4 Jie Zhou3 Sen Song1 1Laboratory for Brain and Intelligence and Department of Biomedical Engineering, Tsinghua University 2Department of Computing Science, University of Alberta; Alberta Machine Intelligent Institute (Amii) 3Pattern Recognition Center, WeChat AI, Tencent Inc, 4ByteDance AI Lab liuxg16@mails.tsinghua.edu.cn, doublepower.mou@gmail.com {fandongmeng,withtomzhou}@tencent.com zhouhao.nlp@bytedance.com, songsen@tsinghua.edu.cn Abstract We propose UPSA, a novel approach that accomplishes Unsupervised Paraphrasing by Simulated Annealing. We model paraphrase generation as an optimization problem and pro- pose a sophisticated objective function, involv- ing semantic similarity, expression diversity, and language fluency of paraphrases. UPSA searches the sentence space towards this objec- tive by performing a sequence of local edits. We evaluate our approach on various datasets, namely, Quora, Wikianswers, MSCOCO, and Twitter. Extensive results show that UPSA achieves the state-of-the-art performance com- pared with previous unsupervised methods in terms of both automatic and human evalua- tions. Further, our approach outperforms most existing domain-adapted supervised models, showing the generalizability of UPSA.1 1 Introduction Paraphrasing aims to restate one sentence as an- other with the same meaning, but different word- ings. It constitutes a corner stone in many NLP tasks, such as question answering (Mckeown, 1983), information retrieval (Knight and Marcu, 2000), and dialogue systems (Shah et al., 2018). However, automatically generating accurate and different-appearing paraphrases is a still challeng- ing research problem, due to the complexity of natural language. Conventional approaches (Prakash et al., 2016; Gupta et al., 2018) model the paraphrase genera- tion as a supervised encoding-decoding problem, inspired by machine translation systems. Usually, such models require massive parallel samples for training. In machine translation, for example, the WMT 2014 English-German dataset contains 4.5M sentence pairs (Neidert et al., 2014). 1Code and data available at: https://github.com/ Liuxg16/UPSA Replace What would you do when you have the power to be invisible ? What would you do when you have the power to become invisible ? What would you do if given the power to become invisible ? What would you do if given you have the power to become invisible ? Sentence space Score of the generated paraphrase Insert Delete What would you do if given you the power to become invisible ? 1 2 3 4 5 1 …5 Editing steps What would you do if given you have the power to become invisible ? Figure 1: UPSA generates a paraphrase by a series of editing operations (i.e., insertion, replacement, and deletion). At each step, UPSA proposes a candidate modification of the sentence, which is accepted or re- jected according to a certain acceptance rate (only ac- cepted modifications are shown). Although sentences are discrete, we make an analogue in the continuous real x-axis where the distance of two sentences is roughly given by the number of edits. However, the training corpora for paraphrasing are usually small. The widely-used Quora dataset2 only …

Content truncated to 500 words