The Meaning Factory at SemEval-2016 Task 8: Producing AMRs with Boxer

Proceedings of SemEval-2016, pages 1179–1184, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics The Meaning Factory at SemEval-2016 Task 8: Producing AMRs with Boxer Johannes Bjerva CLCG University of Groningen j.bjerva@rug.nl Johan Bos CLCG University of Groningen johan.bos@rug.nl Hessel Haagsma CLCG University of Groningen hessel.haagsma@rug.nl Abstract We participated in the shared task on mean- ing representation parsing (Task 8 at SemEval- 2016) with the aim of investigating whether we could use Boxer, an existing open-domain semantic parser, for this task. However, the meaning representations produced by Boxer, Discourse Representation Structures, are con- siderably different from Abstract Meaning Representations, AMRs, the target meaning representations of the shared task. Our hybrid conversion method (involving lexical adapta- tion as well as post-processing of the out- put) failed to produce state-of-the-art results. Nonetheless, F-scores of 53% on development and 47% on test data (50% unofficially) were obtained. 1 Introduction With the currently increasing interest in semantic parsing, and the diversity of the meaning representa- tions being used, an important challenge is to adapt existing semantic parsers for different semantic rep- resentations. Shared Task 8 of the SemEval-2016 campaign for semantic evaluation is an interesting venue for this, where a system is given an English sentence and has to produce an Abstract Meaning Representation (AMR) for it. We participated in this shared task with a system rooted in formal semantics based on Discourse Rep- resentation Theory (DRT). In particular, we were in- terested in finding out whether the representations from DRT (Kamp, 1984; Kamp and Reyle, 1993), Discourse Representation Structures (DRSs), could be easily converted into AMRs. In this paper we outline our method, which is based on the seman- tic parser Boxer (Bos, 2008; Bos, 2015), and then present and discuss our results. 2 Background Before we outline our method, we will say a little about the open-domain semantic parser that we used in this shared task. We also give an overview of the differences between the meaning representations produced by Boxer and those that are required for the shared task. To get a first taste of these differ- ences, compare the analysis of ‘All equipment will be completely manufactured’ carried out by Boxer (Figure 1) and that of the gold-standard AMR (Fig- ure 2). |___________________________________ ||...................................|| _____________ _______________ | | |x1 | |e1 s1 | | | |.............| |...............| | | |equipment(x1)|>|manufacture(e1)| | | |_____________| |Manner(e1,s1) | | | |Theme(e1,x1) | | | |complete(s1) | | | |_______________| | |___________________________________| Figure 1: DRS, as produced by Boxer. (m / manufacture-01 :ARG1 (e2 / equipment :mod (a2 / all)) :ARG1-of (c / complete-02)) Figure 2: Gold-standard AMR. 1179 2.1 Boxer The semantic parser that we employed is Boxer (Bos, 2008; Bos, 2015). It is the last component in the C&C tools pipeline (Curran et al., 2007), comprising a tokenizer (Evang et al., 2013), POS- tagger, lemmatizer (Minnen et al., 2001), and a ro- bust parser for CCG, Combinatory Categorial Gram- mar (Steedman, 2001). Overall, this parsing frame- work shows many points …

Content truncated to 500 words