Global Methods for Cross-lingual Semantic Role and Predicate Labelling

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1279–1290, Dublin, Ireland, August 23-29 2014. Global Methods for Cross-lingual Semantic Role and Predicate Labelling Lonneke van der Plas Institute for NLP Pfaffenwaldring 5B 70569 Stuttgart, Germany vdplasme@ims.uni-stuttgart.de Marianna Apidianaki LIMSI-CNRS Rue John von Neumann 91405 Orsay Cedex, France marianna@limsi.fr Chenhua Chen Institute for NLP Pfaffenwaldring 5B 70569 Stuttgart, Germany cch.chenhua@googlemail.com Abstract We address the problem of transferring semantic annotations to new languages using parallel corpora. Previous work has transferred these annotations on a token-to-token basis, an approach that is sensitive to alignment errors and translation shifts. We present a global approach to transfer that aggregates information across the whole parallel corpus and leads to more robust labellers. We build two global models, one for predicate labelling and one for role labelling, each tailored to the task at hand. We show that the combination of direct and global methods outperforms previous results. 1 Introduction With the proliferation of the Internet in non-English speaking countries, the need for multilingual processing becomes more and more pressing. Various efforts have focused on developing language- independent NLP tools and extending to other languages tools that had been exclusive to English. Fur- thermore, several annotation efforts have been devoted to developing resources for different languages, needed for supervised learning (Hajiˇc et al., 2009). However, there is still a large number of languages for which corpora with semantic annotations do not exist. Since manual annotation is a costly and time- consuming approach to resource development, cross-lingual annotation transfer offers an alternative. Semantic parsing or semantic role labelling (SRL) is the task of automatically labelling predicates and arguments with predicate-argument structure. This level of analysis provides a more stable semantic representation across syntactically different sentences. The example sentences (1a) and (1b) illustrate how the semantic annotation remains stable across the locative alternation of the verb load. (1) a. [AGENT Jessica] [REL-LOAD.01 loaded] [THEME boxes] [DESTINATION into the wagon]. b. [AGENT Jessica] [REL-LOAD.01 loaded] [DESTINATION the wagon] [THEME with boxes]. Also in the cross-lingual setting, the predicate-argument structure of a sentence is considered to be more stable than its syntactic form as the English sentence in (2a) and its French translation in (2b) show: (2) a. [EXPERIENCER Mary] [REL-LIKE.01 liked] [CONTENT the idea]. (English) b. [CONTENT L’id´ee] a [REL-LIKE.01 plu] [EXPERIENCER `a Marie]. (French) This is why several pieces of work have transferred semantic annotations from a source language, for which semantic annotations exist, to a target language using parallel corpora (Pad´o, 2007; Basili et al., 2009; Annesi and Basili, 2010). These transfer methods rely on the assumption of semantic equivalence of the original and the translated sentences, but also on correct and complete alignments between words This work is licensed under a Creative Commons Attribution 4.0 International Licence. Page numbers and proceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ 1279 or constituents in those sentences. We will refer to these traditional methods as direct transfer because the semantic annotations are transferred …

Content truncated to 500 words