World Knowledge for Abstract Meaning Representation Parsing

World Knowledge for Abstract Meaning Representation Parsing Charles Welch1, Jonathan K. Kummerfeld1, Song Feng2, Rada Mihalcea1 University of Michigan1, IBM Research2 {cfwelch, jkummerf, mihalcea}@umich.edu, sfeng@us.ibm.com Abstract In this paper we explore the role played by world knowledge in semantic parsing. We look at the types of errors that currently exist in a state-of-the-art Abstract Meaning Representation (AMR) parser, and explore the problem of how to integrate world knowledge to reduce these errors. We look at three types of knowledge from (1) WordNet hypernyms and super senses, (2) Wikipedia entity links, and (3) retraining a named entity recognizer to identify concepts in AMR. The retrained entity recognizer is not perfect and cannot recognize all concepts in AMR and we examine the limitations of the named entity features using a set of oracles. The oracles show how performance increases if it can recognize different subsets of AMR concepts. These results show improvement on multiple fine-grained metrics, including a 6% increase in named entity F-score, and provide insight into the potential of world knowledge for future work in Abstract Meaning Representation parsing. Keywords: semantic parsing, Abstract Meaning Representation, world knowledge, named entity recognition 1. Introduction Abstract Meaning Representation (AMR), introduced by Banarescu et al. (2012), aims to capture the semantic meaning of sentences using directed acyclic graphs where nodes are labeled with concepts and edges are labeled with relations. An example is shown in Figure 1. A number of recent studies use AMR graphs for downstream tasks (Pan et al., 2015; Liu et al., 2015; Sachan and Xing, 2016; Burns et al., 2016) and growing amounts of annotated data enable the development of statistical parsing algorithms and standardized evaluations. Several parsers have been created that generate an AMR graph given a sentence (Flanigan et al., 2014; Wang et al., 2015b; Damonte et al., 2016), but even the most recent results suggest that there is still significant room to improve the performance for this challenging task. In this paper, we explore the role of world knowledge for the task of semantic parsing with AMR. The paper makes three main contributions. First, we examine the effect of different types of world knowledge for semantic parsing with AMR for the first time.1 Second, we examine the up- per bound on world knowledge using gold annotations, and provide new insights into the potential of world knowledge in computational approaches to AMR parsing. Finally, we show that we can improve the parsing score over a state-of- the-art parser, with improvement on multiple fine-grained evaluation metrics, including a 6% increase in named en- tity F-score. 2. Background There are several semantic parsers built for AMR anno- tations (Flanigan et al., 2014; Zhou et al., 2016; Wang et al., 2015a; Damonte et al., 2016; Barzdins and Gosko, 2016; Misra and Artzi, 2016) using data released through the LDC (LDC2014T12, LDC2015E86, LDC2016E25, 1The code modifications and features are available at https: //github.com/cfwelch/amr_world_knowledge. Figure 1: Example AMR graph for the sentence “China is expanding its military power to attempt to join the …

Content truncated to 500 words