Multi-Task Semantic Dependency Parsing with Policy Gradient for Learning Easy-First Strategies

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2420–2430 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 2420 Multi-Task Semantic Dependency Parsing with Policy Gradient for Learning Easy-First Strategies Shuhei Kurita Center for Advanced Intelligence Project RIKEN Tokyo, Japan shuhei.kurita@riken.jp Anders Søgaard Department of Computer Science University of Copenhagen Copenhagen, Denmark soegaard@di.ku.dk Abstract In Semantic Dependency Parsing (SDP), se- mantic relations form directed acyclic graphs, rather than trees. We propose a new iterative predicate selection (IPS) algorithm for SDP. Our IPS algorithm combines the graph-based and transition-based parsing approaches in or- der to handle multiple semantic head words. We train the IPS model using a combination of multi-task learning and task-specific pol- icy gradient training. Trained this way, IPS achieves a new state of the art on the SemEval 2015 Task 18 datasets. Furthermore, we ob- serve that policy gradient training learns an easy-first strategy. 1 Introduction Dependency parsers assign syntactic structures to sentences in the form of trees. Semantic depen- dency parsing (SDP), first introduced in the Se- mEval 2014 shared task (Oepen et al., 2014), in contrast, is the task of assigning semantic struc- tures in the form of directed acyclic graphs to sen- tences. SDP graphs consist of binary semantic re- lations, connecting semantic predicates and their arguments. A notable feature of SDP is that words can be the semantic arguments of multiple predi- cates. For example, in the English sentence: “The man went back and spoke to the desk clerk” – the word “man” is the subject of the two predicates “went back” and “spoke”. SDP formalisms typi- cally express this by two directed arcs, from the two predicates to the argument. This yields a di- rected acyclic graph that expresses various rela- tions among words. However, the fact that SDP structures are directed acyclic graphs means that we cannot apply standard dependency parsing al- gorithms to SDP. Standard dependency parsing algorithms are of- ten said to come in two flavors: transition-based The man went back and spoke to the desk clerk. a) DM The man went back and spoke to the desk clerk. b) PAS The man went back and spoke to the desk clerk. c) PSD ROOT ROOT ROOT ROOT BV ARG1 ARG1 AND_C LOC ARG1 ARG2 BV COMPOUND DET_ARG1 VERB_ARG1 VERB_ARG1 ADJ_ARG1 COORD_ARG1 COORD_ARG2 PREP_ARG1 PREP_ARG2 DET_ARG1 NOUN_ARG1 ACT DIR3 CONJ.MEMBER RSTR ADDR CONJ.MEMBER Figure 1: Semantic dependency parsing arcs of DM, PAS and PSD formalisms. parsers score transitions between states, and grad- ually build up dependency graphs on the side. Graph-based parsers, in contrast, score all candi- date edges directly and apply tree decoding algo- rithms for the resulting score table. The two types of parsing algorithms have different advantages (McDonald and Nivre, 2007), with transition- based parsers often having more problems with er- ror propagation and, as a result, with long-distance dependencies. This paper presents a compromise between transition-based and graph-based parsing, called iterative predicate selection (IPS) – inspired by head …

Content truncated to 500 words