A New Representation for Span-based CCG Parsing

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10579–10584 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 10579 A New Representation for Span-based CCG Parsing Yoshihide Kato and Shigeki Matsubara Information & Communications, Nagoya University Furo-cho, Chikusa-ku, Nagoya, 464-8601 Japan yoshihide@icts.nagoya-u.ac.jp Abstract This paper proposes a new representation for CCG derivations. CCG derivations are rep- resented as trees whose nodes are labeled with categories strictly restricted by CCG rule schemata. This characteristic is not suitable for span-based parsing models because they predict node labels independently. In other words, span-based models may generate in- valid CCG derivations that violate the rule schemata. Our proposed representation de- composes CCG derivations into several inde- pendent pieces and prevents the span-based parsing models from violating the schemata. Our experimental result shows that an off-the- shelf span-based parser with our representa- tion is comparable with previous CCG parsers. 1 Introduction Combinatory Categorial Grammar (CCG) (Steed- man, 2000) is a mildly context-sensitive grammar formalism. Several neural CCG parsing methods have been proposed so far (Lewis and Steedman, 2014; Xu et al., 2015; Lewis et al., 2016; Vaswani et al., 2016; Lee et al., 2016; Xu, 2016; Yoshikawa et al., 2017; Stanojevi´c and Steedman, 2019, 2020; Bhargava and Penn, 2020; Tian et al., 2020; Prange et al., 2021; Liu et al., 2021). Currently, neural span-based models (Cross and Huang, 2016; Stern et al., 2017; Gaddy et al., 2018; Kitaev and Klein, 2018) have been successful in the field of con- stituency parsing. However, we cannot directly apply this technique to CCG parsing. Span-based models assume that each node label in parse trees can be predicted independently, while, in CCG, each node label (category) is strictly restricted by CCG rule schemata. The independence assumption of span-based models implies that the models are not guaranteed to generate valid CCG derivations. To solve this problem, we propose a method of representing CCG derivations in a way suitable for span-based parsing models. Our proposed repre- X/Y Y |1Z1 · · · |dZd ⇒ X|1Z1 · · · |dZd (>d) Y |1Z1 · · · |dZd X\Y ⇒ X|1Z1 · · · |dZd (<d) Figure 1: CCG rule schemata. sentation decomposes CCG derivations into sev- eral independent pieces and can prevent the span- based parsing models from violating the CCG rule schemata. Furthermore, as a by-product of our representation, the parsing models can assign out- of-vocabulary (OOV) categories, which have not appeared in training data. This characteristic has been attracting attention in CCG parsing research (Bhargava and Penn, 2020; Prange et al., 2021; Liu et al., 2021). Our experimental result shows that an off-the-shelf span-based parser with our representa- tion is comparable with previous CCG parsers and can generate correct OOV categories. 2 CCG and Span-based Parsing This section gives an overview of Combinatory Categorial Grammar (CCG) (Steedman, 2000) and explains why we cannot directly apply the span- based approach to CCG parsing. 2.1 Combinatory Categorial Grammar CCG represents syntactic information by basic cat- egories (e.g., S, …

Content truncated to 500 words