Parsing into Variable-in-situ Logico-Semantic Graphs

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6772–6782 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 6772 Parsing into Variable-in-situ Logico-Semantic Graphs Yufei Chen1 and Weiwei Sun12 1Wangxuan Institute of Computer Technology 1The MOE Key Laboratory of Computational Linguistics 2Center for Chinese Linguistics Peking University {yufei.chen,ws}@pku.edu.cn Abstract We propose variable-in-situ logico-semantic graphs to bridge the gap between semantic graph and logical form parsing. The new type of graph-based meaning representation allows us to include analysis for scope-related phenomena, such as quantification, negation and modality, in a way that is consistent with the state-of-the-art underspecification ap- proach. Moreover, the well-formedness of such a graph is clear, since model-theoretic in- terpretation is available. We demonstrate the effectiveness of this new perspective by de- veloping a new state-of-the-art semantic parser for Minimal Recursion Semantics. At the core of this parser is a novel neural graph rewriting system which combines the strengths of Hy- peredge Replacement Grammar, a knowledge- intensive model, and Graph Neural Networks, a data-intensive model. Our parser achieves an accuracy of 92.39% in terms of ELEMENTARY DEPENDENCY MATCH, which is a 2.88 point improvement over the best data-driven model in the literature. The output of our parser is highly coherent: at least 91% graphs are valid, in that they allow at least one sound scope- resolved logical form. 1 Introduction Graphs have recently become popular as a strat- egy for encoding sentence-level semantics, and related data-driven parsing techniques have been making rapid progress. The primary component of popular semantic graphs, e.g. Elementary Depen- dency Structure (EDS; Oepen and Lønning, 2006) and Abstract Meaning Representation (AMR; Ba- narescu et al., 2013), is the predicate–argument structure, with the predicate being a concept that takes a number of arguments. Though expressive for many applications, this predicative core does not fully match the need for logical forms that used to stand in the central area of semantic parsing. Partly due to the lack of model-theoretic seman- tics, it is rather difficult to add scope information related to quantification, negation and modality to a graph. Partly due to the lack of logical deduction engines, it is rather difficult to directly perform au- tomated reasoning over graphs. This paper proposes to express logical forms with variable-in-situ graphs for the ongoing ad- vances in graph-centric formalisms, algorithms and neural architectures. This leads us to a novel neural graph rewriting system that combines the strengths of Hyperedge Replacement Grammar (HRG; Drewes et al., 1997) and Graph Neural Net- works (Song et al., 2018a). On the one hand, it can be viewed as an improved graph embedding model that explicitly explores recursive structures that are defined by an HRG. On the other hand, it can be viewed as an enhanced graph grammar with which all nodes involved in derivations of graphs are assigned vector-based distributed encodings. Based on our neural graph rewriting system, we develop a new parser for Minimal Recursion Se- mantics (MRS; Copestake et al., 2005). By means of the …

Content truncated to 500 words