Compositional Semantic Parsing across Graphbanks

Abstract

Most semantic parsers that map sentences to graph-based meaning representations are handdesigned for specific graphbanks. We present a compositional neural semantic parser which achieves, for the first time, competitive accuracies across a diverse range of graphbanks. Incorporating BERT embeddings and multi-task learning improves the accuracy further, setting new states of the art on DM, PAS, PSD, AMR 2015 and EDS.

Saarland at MRP 2019: Compositional parsing across all graphbanks

“…Our parser achieved the highest accuracy on PSD and did very well on DM and EDS. It did much worse on AMR than we expected based on earlier results (Lindemann et al, 2019). Table 2 shows a more detailed evaluation of the system on the development sets.…”

“…In earlier work, we showed how to accurately predict AM dependency trees for AMR using a neural dependency parser and supertagger (Groschwitz et al, 2018). We extended this parser from AMR to the DM, PAS, PSD, and EDS graphbanks and obtained state-of-the-art results across all of these graphbanks (Lindemann et al, 2019); we will call this system the ACL-19 parser throughout this paper. Earlier semantic parsers were only available for one or two families of closely related graphbanks; our system was the first to parse accurately across a range of different graphbanks.…”

Peking at MRP 2019: Factorization- and Composition-Based Parsing for Elementary Dependency Structures

“…On the computational side, such rules must be governed by a well-defined grammar formalism. In particular, to manipulate graph construction in a principled way, Hyperedge Replacement Grammar (HRG; Drewes et al, 1997) and AM Algebra (Groschwitz et al, 2017) have been applied to build semantic parsers for various graph banks (Chen et al, 2018b;Groschwitz et al, 2018;Lindemann et al, 2019).…”

Graph-Based Meaning Representations: Design and Processing

“…• Composition-based methods (Callmeier, 2000;Bos et al, 2004;Artzi et al, 2015;Groschwitz et al, 2018;Lindemann et al, 2019;Chen et al, 2018);…”