Discourse Representation Parsing for Sentences and Documents

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 6248–6262 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 6248 Discourse Representation Parsing for Sentences and Documents Jiangming Liu Shay B. Cohen Mirella Lapata Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh 10 Crichton Street, Edinburgh EH8 9AB Jiangming.Liu@ed.ac.uk, {scohen,mlap}@inf.ed.ac.uk Abstract We introduce a novel semantic parsing task based on Discourse Representation Theory (DRT; Kamp and Reyle 1993). Our model operates over Discourse Representation Tree Structures which we formally define for sen- tences and documents. We present a general framework for parsing discourse structures of arbitrary length and granularity. We achieve this with a neural model equipped with a su- pervised hierarchical attention mechanism and a linguistically-motivated copy strategy. Ex- perimental results on sentence- and document- level benchmarks show that our model outper- forms competitive baselines by a wide margin. 1 Introduction Semantic parsing is the task of mapping natural language to machine interpretable meaning repre- sentations. Various models have been proposed over the years to learn semantic parsers from lin- guistic expressions paired with logical forms, SQL queries, or source code (Kate et al., 2005; Liang et al., 2011; Zettlemoyer and Collins, 2005; Ba- narescu et al., 2013; Wong and Mooney, 2007; Kwiatkowski et al., 2011; Zhao and Huang, 2015). The successful application of encoder-decoder models (Sutskever et al., 2014; Bahdanau et al., 2015) to a variety of NLP tasks has prompted the reformulation of semantic parsing as a sequence- to-sequence learning problem (Dong and Lap- ata, 2016; Jia and Liang, 2016; Koˇcisk`y et al., 2016), although most recent efforts focus on ar- chitectures which make use of the syntax of meaning representations, e.g., by developing tree or graph-structured decoders (Dong and Lapata, 2016; Cheng et al., 2017; Yin and Neubig, 2017; Alvarez-Melis and Jaakkola, 2017; Rabinovich et al., 2017; Buys and Blunsom, 2017). In this work we focus on parsing formal mean- ing representations in the style of Discourse Rep- resentation Theory (DRT; Kamp and Reyle 1993). a. k1 : x1 e1 t1 max(x1) fall(e1) Agent(e1, x1) now(t1) e1 ≤ t1 k2 : ⋄ : x2 e2 john(x2) push(e2) Patient(e2, x2) male(x1) e2 ≤ e1 because(k1, k2) b. SDRS max(x1) k1 k2 DRS fall(e1) Agent(e1,x1) now(t1) temp_before(e1, t1) DRS POS DRS john(x2) push(e2) Patient(e2, x1) male(x1) temp_before(e2, e1) because(k1, k2) Max fell . John might push him . Figure 1: Meaning representation for the discourse “Max fell. John might push him.” in box-like format (top) and as a tree (bottom). Red lines indicate termi- nals corresponding to words and green lines indicate non-terminals corresponding to sentences. ⋄ and POS are modality operators for possibility. DRT is a popular theory of meaning representa- tion (Kamp, 1981; Kamp and Reyle, 1993; Asher, 1993; Asher and Lascarides, 2003) designed to ac- count for a variety of linguistic phenomena, in- cluding the interpretation of pronouns and tempo- ral expressions within and across sentences. The basic meaning-carrying units in DRT are …

Content truncated to 500 words