Reducing Discontinuous to Continuous Parsing with Pointer Network Reordering

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10570–10578 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 10570 Reducing Discontinuous to Continuous Parsing with Pointer Network Reordering Daniel Fernández-González and Carlos Gómez-Rodríguez Universidade da Coruña, CITIC FASTPARSE Lab, LyS Group Depto. de Ciencias de la Computación y Tecnologías de la Información Campus de Elviña, s/n, 15071 A Coruña, Spain d.fgonzalez@udc.es, carlos.gomez@udc.es Abstract Discontinuous constituent parsers have always lagged behind continuous approaches in terms of accuracy and speed, as the presence of con- stituents with discontinuous yield introduces extra complexity to the task. However, a dis- continuous tree can be converted into a contin- uous variant by reordering tokens. Based on that, we propose to reduce discontinuous pars- ing to a continuous problem, which can then be directly solved by any off-the-shelf contin- uous parser. To that end, we develop a Pointer Network capable of accurately generating the continuous token arrangement for a given in- put sentence and define a bijective function to recover the original order. Experiments on the main benchmarks with two continuous parsers prove that our approach is on par in accuracy with purely discontinuous state-of-the-art algo- rithms, but considerably faster. 1 Introduction Discontinuous phrase-structure trees (with cross- ing branches like the one in Figure 1(a)) are crucial for fully representing the wide range of syntac- tic phenomena present in human languages such as long-distance extractions, dislocations or cross- serial dependencies, among others. Although continuous approaches ignore these linguistic phenomena by, for instance, removing them from the original treebank (a common prac- tice in the Penn Treebank (Marcus et al., 1993)), there exist different algorithms that can handle dis- continuous parsing. Currently, we can highlight (1) those based in Linear Context-Free Rewriting Sys- tems (LCFRS) (Vijay-Shanker et al., 1987), which allow exact CKY-style parsing of discontinuous structures at a high computational cost (Gebhardt, 2020; Ruprecht and Mörbitz, 2021); (2) a variant of the former that, while still making use of LCFRS formalisms, increases parsing speed by implement- ing a span-based scoring algorithm (Stern et al., 2017) and not explicitly defining a set of rules (Stanojevi´c and Steedman, 2020; Corro, 2020); (3) transition-based parsers that deal with discontinu- ities by adding a specific transition in charge of changing token order (Versley, 2014; Maier, 2015; Maier and Lichte, 2016; Stanojevi´c and Alhama, 2017; Coavoux and Crabbé, 2017) or by designing new data structures that allow interactions between already-created non-adjacent subtrees (Coavoux et al., 2019; Coavoux and Cohen, 2019); and, fi- nally, (4) several approaches that reduce discon- tinuous constituent parsing to a simpler problem, converting it, for instance, into a non-projective dependency parsing task (Fernández-González and Martins, 2015; Fernández-González and Gómez- Rodríguez, 2020a) or into a sequence labelling problem (Vilares and Gómez-Rodríguez, 2020). In (4), we can also include the solutions proposed by Boyd (2007) and Versley (2016), which transform discontinuous treebanks into continuous variants where discontinuous constituents are encoded by creating additional constituent nodes and extend- ing the original non-terminal label set (following a …

Content truncated to 500 words