Cross-Lingual Abstract Meaning Representation Parsing

Abstract

Meaning Representation (AMR) research has mostly focused on English. We show that it is possible to use AMR annotations for English as a semantic representation for sentences written in other languages. We exploit an AMR parser for English and parallel 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.

Towards Universal Semantic Representation

“…Using techniques similar to those used to transfer SRL and AMR from one language to another (Akbik et al, 2015;Damonte and Cohen, 2018), it is possible to transfer labeling schemes for the additional fewatures and structures discussed in this paper from one language to another. The cross-lingual transfer may also help to discover better feature sets from data.…”

Proceedings of the 2nd Workshop on Multilingual Surface Realisation (MSR 2019)

“…Among the efforts to build or adapt semantic representations for non-English languages, it is possible to cite Abstract Meaning Representation (AMR) as an example. Although AMR was not born as an interlingua, several works have tried to use it in that way to annotate sentences in other languages like Chinese and Czech (Xue et al, 2014), Italian, Spanish, and German (Damonte and Cohen, 2018) and Brazilian Portuguese (Anchiêta and Pardo, 2018). Other works have tried to adapt the English AMR guidelines to Spanish and Brazilian Portuguese with some success (Migueles-Abraira et al, 2018;Sobrevilla Cabezudo and Pardo, 2019).…”

“…6 The current AMR-annotated corpus for English contains 39,260 sentences. Some efforts have been performed to build a corpus for Non-English languages leveraging the alignments and the parallel corpora that exist and trying to consider AMR an interlingua (Xue et al, 2014;Damonte and Cohen, 2018;Anchiêta and Pardo, 2018). Other works tried to adapt the AMR guidelines to other languages (Migueles-Abraira et al, 2018;Sobrevilla Cabezudo and Pardo, 2019).…”

Cross-lingual

“…Recent years have seen much interest in crosslingual learning, that is, learning tagging and parsing models for languages without training data for that language, instead relying on training data or existing systems for another language, and on parallel data to transfer knowledge from one language to the other. This is either done by automatically projecting source-language annotations from the source text to the target text (Yarowsky et al, 2001;Hwa et al, 2005;Tiedemann, 2014;Rasooli and Collins, 2015;Damonte and Cohen, 2018), sharing parameters between models for different languages (Zeman and Resnik, 2008;Ganchev et al, 2009;McDonald et al, 2011;Naseem et al, 2012;Täckström et al, 2013;de Lhoneux et al, 2018), or automatically translating the text from the source language to the target language and synchronously projecting the annotations . Our work is an application of the first approach to CCG, which as a grammar formalism provides a more systematic framework for the study of syntax and for compositional interpretation than dependency parsers.…”