AMR Parsing with Latent Structural Information

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4306–4319 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 4306 AMR Parsing with Latent Structural Information Qiji Zhou1†, Yue Zhang2,3, Donghong Ji1∗, Hao Tang1 1Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, China {qiji.zhou,dhji,tanghaopro}@whu.edu.cn 2School of Engineering, Westlake University 3Institute of Advanced Technology, Westlake Institute for Advanced Study yue.zhang@wias.org.cn Abstract Abstract Meaning Representations (AMRs) capture sentence-level semantics structural representations to broad-coverage natural sen- tences. We investigate parsing AMR with ex- plicit dependency structures and interpretable latent structures. We generate the latent soft structure without additional annotations, and fuse both dependency and latent structure via an extended graph neural networks. The fused structural information helps our experi- ments results to achieve the best reported re- sults on both AMR 2.0 (77.5% Smatch F1 on LDC2017T10) and AMR 1.0 (71.8% Smatch F1 on LDC2014T12). 1 Introduction Abstract Meaning Representations (AMRs) (Ba- narescu et al., 2013) model sentence level seman- tics as rooted, directed, acyclic graphs. Nodes in the graph are concepts which represent the events, objects and features of the input sentence, and edges between nodes represent semantic relations. AMR introduces re-entrance relation to depict the node reuse in the graphs. It has been adopted in downstream NLP tasks, including text summariza- tion (Liu et al., 2015; Dohare and Karnick, 2017), question answering (Mitra and Baral, 2016) and machine translation (Jones et al., 2012; Song et al., 2019). AMR parsing aims to transform natural language sentences into AMR semantic graphs. Similar to constituent parsing and dependency parsing (Nivre, 2008; Dozat and Manning, 2017), AMR parsers mainly employ two parsing techniques: transition- based parsing (Wang et al., 2016; Damonte et al., 2017; Wang and Xue, 2017; Liu et al., 2018; Guo and Lu, 2018) use a sequence of transition actions †Part of work was done when the author was visiting West- lake University *Corresponding author. The boy came and left and boy come-01 leave-11 det nsubj cc conj :op1 :op2 :ARG1 :ARG0 Figure 1: An example of AMR graph and its corre- sponding syntactic dependencies for the sentence “The boy came and left”. The dashed lines denote the con- nected relations in the syntactic dependencies but not appear in the AMR graph. to incrementally construct the graph, while graph- based parsing (Flanigan et al., 2014; Lyu and Titov, 2018; Zhang et al., 2019a; Cai and Lam, 2019) divides the task into concept identification and rela- tion extraction stages and then generate a full AMR graph with decoding algorithms such as greedy and maximum spanning tree (MST). Additionally, re- inforcement learning (Naseem et al., 2019) and sequence-to-sequence (Konstas et al., 2017) have been exploited in AMR parsing as well. Previous works (Wang et al., 2016; Artzi et al., 2015) shows that structural information can bring benefit to AMR parsing. Illustrated by Figure 1, for example syntactic dependencies can convey the main predicate-argument structure. However, dependency structural information …

Content truncated to 500 words