ICL-HD at SemEval-2016 Task 8: Meaning Representation Parsing - Augmenting AMR Parsing with a Preposition Semantic Role Labeling Neural Network

Proceedings of SemEval-2016, pages 1160–1166, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics ICL-HD at SemEval-2016 Task 8: Meaning Representation Parsing - Augmenting AMR Parsing with a Preposition Semantic Role Labeling Neural Network Lauritz Brandt and David Grimm and Mengfei Zhou and Yannick Versley Institute for Computational Linguistics Ruprecht-Karls-Universit¨at Heidelberg {brandt, grimm, zhou, versley}@cl.uni-heidelberg.de Abstract We describe our submission system to the SemEval-2016 Task 8 on Abstract Meaning Representation (AMR) Parsing. We attempt to improve AMR parsing by exploiting prepo- sition semantic role labeling information re- trieved from a multi-layer feed-forward neu- ral network. Prepositional semantics is in- cluded as features into the transition-based AMR parsing system CAMR (Wang, Xue, and S. Pradhan 2015a). The inclusion of the fea- tures modifies the behavior of CAMR when creating meaning representations triggered by prepositional semantics. Despite the useful- ness of preposition semantic role labeling in- formation for AMR parsing, it does not have an impact to the parsing F-score of CAMR, but reduces the parsing recall by 1%. 1 Introduction Progress in Natural Language Processing has led to a multitude of well-motivated tasks that each rep- resent part of a sentence’s meaning but result in a meaning description spread over separate, uncon- nected descriptions. These separate levels of se- mantic annotation, like co-reference or named en- tities, and the lack of simple human-readable cor- pora where whole sentence meanings are encoded led to the Abstract Meaning Representation (AMR) formalism (Banarescu et al. 2013). AMR structures capture sentence meanings with rooted, directed and labeled graphs where sentences with the same mean- ing receive the same AMR. These graphs are en- coded in a bracketed format and can be visually rep- resented in a human-understandable way (see Figure 1). AMR structures are organized with nodes repre- senting concepts and the semantic relationships that hold between these concepts1. Hence, AMRs can be useful for every NLP component that relies on or exploits semantic meaning resources. Particular ap- plication areas are, among others, entity linking (Pan et al. 2015), event detection (Li et al. 2015) and ma- chine translation. An example for a AMR graph is given in Figure 1: there is a concept RECOMMEND- 01 which is the root of the graph and there is a con- cept OFFER-01 that stands in semantic relationship to RECOMMEND-01 with the edge ARG1. We augment the existing AMR parser CAMR (Wang, Xue, and S. Pradhan 2015a) with a prepo- sition semantic role labeling (prepSRL) neural net- work with the intention to improve the AMR graph creation accuracy. Prepositions in conjunction with their arguments make a crucial contribution to the meaning of sentences and are therefore a very intu- itive supplement to AMR parsing. For example, see how in Figure 2 the meaning of the preposition in is involved in the creation of the AMR edge :LOCA- TION. in semantically expresses the agency’s spatial location and therefore triggers the identically named AMR edge :LOCATION. Prepositional semantics is a knowledge resource that has not yet been exploited …

Content truncated to 500 words