Not Just Classification: Recognizing Implicit Discourse Relation on Joint Modeling of Classification and Generation

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2418–2431 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 2418 Not Just Classification: Recognizing Implicit Discourse Relation on Joint Modeling of Classification and Generation Feng Jiang, Yaxin Fan, Xiaomin Chu, Peifeng Li∗, Qiaoming Zhu Soochow University {fjiang,yxfansupery}@stu.suda.edu.cn, {xmchu,pfli,qmzhu}@suda.edu.cn Abstract Implicit discourse relation recognition (IDRR) is a critical task in discourse analysis. Pre- vious studies only regard it as a classifica- tion task and lack an in-depth understanding of the semantics of different relations. There- fore, we first view IDRR as a generation task and further propose a method joint modeling of the classification and generation. Specif- ically, we propose a joint model, CG-T5, to recognize the relation label and generate the target sentence containing the meaning of re- lations simultaneously. Furthermore, we de- sign three target sentence forms, including the question form, for the generation model to in- corporate prior knowledge. To address the is- sue that large discourse units are hardly em- bedded into the target sentence, we also pro- pose a target sentence construction mechanism that automatically extracts core sentences from those large discourse units. Experimental re- sults both on Chinese MCDTB and English PDTB datasets show that our model CG-T5 achieves the best performance against several state-of-the-art systems. 1 Introduction Discourse relation describes the logical connection between two discourse units (e.g., clauses, sen- tences, or paragraphs). As an essential discourse analysis task, discourse relation recognition is to recover what rhetorical relation exists between dis- course units (DUs). Due to the absence of explicit connectives, implicit discourse relation recogni- tion (IDRR) is still a challenging task and research hotspot. Moreover, IDRR is beneficial to many downstream natural language processing (NLP) ap- plications, such as machine translation (Webber et al., 2017), text generation (Bosselut et al., 2018), and text summarization (Xu et al., 2020). With the success of representation learning in discourse analysis, most existing methods of IDRR ∗Corresponding author focus on three aspects: enhancing discourse units representation (Ji and Eisenstein, 2015; Qin et al., 2016; Liu and Li, 2016), enhancing semantic in- teraction (Guo et al., 2018; Ruan et al., 2020; Guo et al., 2020), and joint learning with other tasks (Bai and Zhao, 2018; Nguyen et al., 2019; He et al., 2020). They all regard IDRR as a classification task and lack a deeper understanding of the rela- tion semantics; even recent work (Nguyen et al., 2019; He et al., 2020) with labeling embedding cannot directly introduce the prior knowledge of the discourse relation semantics into their models. Ningbo Free Trade Zone ... has achieved DU1 fruitful results after three years of construction. ... the development level is among the best... ... the Ningbo Free Trade Zone had completed DU2 a total of US$812 million in import and ... At the same time, the bonded zone has ... Relation Elaboration Target Sentence DU2 is a detailed description of DU1. Table 1: The example of implicit discourse relation be- tween two DUs where DU1 and …

Content truncated to 500 words