A Structured Syntax-Semantics Interface for English-AMR Alignment

Proceedings of NAACL-HLT 2018, pages 1169–1180 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics A Structured Syntax-Semantics Interface for English-AMR Alignment Ida Szubert Adam Lopez School of Informatics University of Edinburgh Edinburgh, Scotland, UK {k.i.szubert@sms, alopez@inf}.ed.ac.uk Nathan Schneider Linguistics and Computer Science Georgetown University Washington, DC, USA nathan.schneider@georgetown.edu Abstract Abstract Meaning Representation (AMR) an- notations are often assumed to closely mir- ror dependency syntax, but AMR explicitly does not require this, and the assumption has never been tested. To test it, we devise an expressive framework to align AMR graphs to dependency graphs, which we use to anno- tate 200 AMRs. Our annotation explains how 97% of AMR edges are evoked by words or syntax. Previously existing AMR alignment frameworks did not allow for mapping AMR onto syntax, and as a consequence they ex- plained at most 23%. While we find that there are indeed many cases where AMR annota- tions closely mirror syntax, there are also per- vasive differences. We use our annotations to test a baseline AMR-to-syntax aligner, find- ing that this task is more difficult than AMR- to-string alignment; and to pinpoint errors in an AMR parser. We make our data and code freely available for further research on AMR parsing and generation, and the relationship of AMR to syntax. 1 Introduction Abstract Meaning Representation (AMR; Ba- narescu et al., 2013) is a popular framework for an- notating whole sentence meaning. An AMR anno- tation is a directed, usually acyclic graph in which nodes represent entities and events, and edges rep- resent relations between them, as on the right in figure 1.1 AMR annotations include no explicit mapping between elements of an AMR and the correspond- ing elements of the sentence that evoke them, and this presents a challenge to developers of machine learning systems that parse sentences to AMR or generate sentences from AMR, since they must 1For clarity of presentation, we have constructed the sen- tences and AMRs shown in figures—except for figure 3, which is a simplified version of a sentence in the corpus. lies lie l cat cat c My i i sun sun s the Figure 1: “My cat lies in the sun.” An alignment be- tween the dependency parse (left) and AMR (right). Nodes participating in lexical alignments are marked with boxes, but the links between them are not dis- played. Structural alignments are colour-coded and linked by dotted lines. Sense numbers for concepts that are PropBank frames are omitted for brevity. first infer this mapping in the training data (e.g. Flanigan et al., 2014; Wang et al., 2015; Artzi et al., 2015; Flanigan et al., 2016; Pourdamghani et al., 2016; Misra and Artzi, 2016; Damonte et al., 2017; Peng et al., 2017, inter alia).2 This AMR alignment problem was first for- malized by Flanigan et al. (2014), who mapped AMR nodes or connected subgraphs to words or sequences of words under the assumption of a one- to-one mapping—we call this JAMR alignment. Pourdamghani et al. …

Content truncated to 500 words