Facilitating Contrastive Learning of Discourse Relational Senses by Exploiting the Hierarchy of Sense Relations

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 10704 - 10716 December 7-11, 2022 ©2022 Association for Computational Linguistics Facilitating Contrastive Learning of Discourse Relational Senses by Exploiting the Hierarchy of Sense Relations Wanqiu Long†, and Bonnie Webber† † University of Edinburgh, Edinburgh, UK Wanqiu.long@ed.ac.uk, Webber.Bonnie@ed.ac.uk Abstract Implicit discourse relation recognition is a chal- lenging task that involves identifying the sense or senses that hold between two adjacent spans of text, in the absense of an explicit connec- tive between them. In both PDTB-2 (Prasad et al., 2008) and PDTB-3 (Webber et al., 2019), discourse relational senses are organized into a three-level hierarchy ranging from four broad top-level senses, to more specific senses below them. Most previous work on implicitf dis- course relation recognition have used the sense hierarchy simply to indicate what sense labels were available. Here we do more — incorpo- rating the sense hierarchy into the recognition process itself and using it to select the negative examples used in contrastive learning. With no additional effort, the approach achieves state- of-the-art performance on the task. Our code is released in https://github.com/wanqiulong 0923/Contrastive_IDRR. 1 Introduction Discourse relations are an important aspect of textual coherence. In some cases, a speaker or writer signals the sense or senses that hold between clauses and/or sentences in a text using an explicit connective. Recognizing the sense or senses that hold can be more difficult, in the absense of an explicit connective. Automatically identifying the sense or senses that hold between sentences and/or clauses can be useful for downstream NLP tasks such as text summarization (Cohan et al., 2018), machine trans- lation (Meyer et al., 2015) and event relation ex- traction (Tang et al., 2021). Recent studies on im- plicit discourse relation recognition have shown great success. Especially, pre-trained neural lan- guage models (Peters et al., 2018; Devlin et al., 2019; Liu et al., 2019) have been used and dramat- ically improved the performances of models (Shi and Demberg, 2019b; Liu et al., 2020; Kishimoto et al., 2020). The senses available for labelling Level2 Level1 Temporal Asynchronous Level3 Precedence Succession Root Comparison Contigency Expansion Synchronous Concession Contrast Cause ... Purpose Cause+Belief ... Reason Result Equivalence Manner Reason+Belief Result+Belief Figure 1: The PDTB-3 Sense Hierarchy discourse relations in the PDTB-2 (and later in the PDTB-3) are arranged in a three-level hierarchy, with the most general senses at the top and more specific senses further down. In the PDTB-3, anno- tators could only choose senses at terminal nodes in the hierarchy – level-2 senses for symmetric re- lations such as EXPANSION.EQUIVALENCE and TEMPORAL.SYNCHRONOUS, and level-3 senses for asymmetric relations, with the direction of the relation encoded in its sense label such as SUBSTITUTION.ARG1-AS-SUBST (where the text labelled ARG1 substitutes for the denied text labelled ARG2) and SUBSTITUTION.ARG2-AS- SUBST (where the text labelled ARG2 substitutes for the denied text labelled ARG1). Early work on recognizing the implicit relations only used the hi- erarchy to choose a target for recognition (e.g., the senses …

Content truncated to 500 words