Exploring Neural Methods for Parsing Discourse Representation Structures

• DOI: 10.1162/tacl_a_00241
• Publication Date: 2019-03-01T20:33:06Z
• Authors: Rik van Noord; Lasha Abzianidze; Antonio Toral; Johan Bos
• Keywords: Representation; Sequence (biology)

Neural methods have had several recent successes in semantic parsing, though they yet to face the challenge of producing meaning representations based on formal semantics. We present a sequence-to-sequence neural parser that is able produce Discourse Representation Structures (DRSs) for English sentences with high accuracy, outperforming traditional DRS parsers. To facilitate learning output, we represent DRSs as sequence flat clauses and introduce method verify produced are well-formed interpretable. compare models using characters words input see (somewhat surprisingly) former performs better than latter. show eliminating variable names from output De Bruijn indices increases performance. Adding silver training data boosts performance even further.