Robust Subgraph Generation Improves Abstract Meaning Representation Parsing

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 982–991, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Robust Subgraph Generation Improves Abstract Meaning Representation Parsing Keenon Werling Stanford University keenon@stanford.edu Gabor Angeli Stanford University angeli@stanford.edu Christopher D. Manning Stanford University manning@stanford.edu Abstract The Abstract Meaning Representation (AMR) is a representation for open- domain rich semantics, with potential use in fields like event extraction and machine translation. Node generation, typically done using a simple dictionary lookup, is currently an important limiting factor in AMR parsing. We propose a small set of actions that derive AMR subgraphs by transformations on spans of text, which allows for more robust learning of this stage. Our set of construction actions generalize better than the previous ap- proach, and can be learned with a sim- ple classifier. We improve on the previ- ous state-of-the-art result for AMR pars- ing, boosting end-to-end performance by 3 F1 on both the LDC2013E117 and LDC2014T12 datasets. 1 Introduction The Abstract Meaning Representation (AMR) (Banarescu et al., 2013) is a rich, graph-based lan- guage for expressing semantics over a broad do- main. The formalism is backed by a large data- labeling effort, and it holds promise for enabling a new breed of natural language applications rang- ing from semantically aware MT to rich broad- domain QA over text-based knowledge bases. Fig- ure 1 shows an example AMR for “he gleefully ran to his dog Rover,” and we give a brief introduction to AMR in Section 2. This paper focuses on AMR parsing, the task of mapping a natural language sentence into an AMR graph. We follow previous work (Flanigan et al., 2014) in dividing AMR parsing into two steps. The first step is concept identification, which generates AMR nodes from text, and which we’ll refer to as NER++ (Section 3.1). The second step is relation Figure 1: The AMR graph for He gleefully ran to his dog Rover. We show that improving the gen- eration of low level subgraphs (e.g., Rover gener- ating name :op1 −−→ “Rover”) significantly improves end-to-end performance. identification, which adds arcs to link these nodes into a fully connected AMR graph, which we’ll call SRL++ (Section 3.2). We observe that SRL++ is not the hard part of AMR parsing; rather, much of the difficulty in AMR is generating high accuracy concept sub- graphs from the NER++ component. For example, when the existing AMR parser JAMR (Flanigan et al., 2014) is given a gold NER++ output, and must only perform SRL++ over given subgraphs it scores 80 F1 – nearly the inter-annotator agree- ment of 83 F1, and far higher than its end to end accuracy of 59 F1. SRL++ within AMR is relatively easy given a perfect NER++ output, because so much pressure is put on the output of NER++ to carry meaningful information. For example, there’s a strong type- check feature for the existence and type of any arc …

Content truncated to 500 words