Cross-Lingual Abstract Meaning Representation Parsing

Proceedings of NAACL-HLT 2018, pages 1146–1155 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Cross-lingual Abstract Meaning Representation Parsing Marco Damonte Shay B. Cohen School of Informatics, University of Edinburgh 10 Crichton Street, Edinburgh EH8 9AB, UK m.damonte@sms.ed.ac.uk scohen@inf.ed.ac.uk Abstract Abstract Meaning Representation (AMR) re- search has mostly focused on English. We show that it is possible to use AMR annota- tions for English as a semantic representation for sentences written in other languages. We exploit an AMR parser for English and paral- lel corpora to learn AMR parsers for Italian, Spanish, German and Chinese. Qualitative analysis show that the new parsers overcome structural differences between the languages. We further propose a method to evaluate the parsers that does not require gold standard data in the target languages. This method highly correlates with the gold standard evaluation, obtaining a (Pearson) correlation of 0.95. 1 Introduction Abstract Meaning Representation (AMR) parsing is the process of converting natural language sen- tences into their corresponding AMR representa- tions (Banarescu et al., 2013). An AMR is a graph with nodes representing the concepts of the sen- tence and edges representing the semantic rela- tions between them. Most available AMR datasets large enough to train statistical models consist of pairs of English sentences and AMR graphs. The cross-lingual properties of AMR across languages has been the subject of preliminary dis- cussions. The AMR guidelines state that AMR is not an interlingua (Banarescu et al., 2013) and Bojar (2014) categorizes different kinds of diver- gences in the annotation between English AMRs and Czech AMRs. Xue et al. (2014) show that structurally aligning English AMRs with Czech and Chinese AMRs is not always possible but that refined annotation guidelines suffice to re- solve some of these cases. We extend this line of research by exploring whether divergences among languages can be overcome, i.e., we investigate This is the sovereignty of each country sovereignty country this each Questa `e la sovranit`a di ogni paese :poss :domain :mod Figure 1: AMR alignments for a English sentence and its Italian translation. whether it is possible to maintain the AMR an- notated for English as a semantic representation for sentences written in other languages, as in Fig- ure 1. We implement AMR parsers for Italian, Span- ish, German and Chinese using annotation pro- jection, where existing annotations are projected from a source language (English) to a target lan- guage through a parallel corpus (e.g., Yarowsky et al., 2001; Hwa et al., 2005; Pad´o and Lapata, 2009; Evang and Bos, 2016). By evaluating the parsers and manually analyzing their output, we show that the parsers are able to recover the AMR structures even when there exist structural differ- ences between the languages, i.e., although AMR is not an interlingua it can act as one. This method also provides a quick way to prototype multilin- gual AMR parsers, assuming that Part-of-speech (POS) taggers, Named Entity Recognition (NER) taggers and dependency parsers are available for the target languages. We …

Content truncated to 500 words