Cross-lingual Word Sense Disambiguation for Predicate Labelling of French

[O-S1.4] 46 21ème Traitement Automatique des Langues Naturelles, Marseille, 2014 Cross-lingual Word Sense Disambiguation for Predicate Labelling of French Lonneke van der Plas1 Marianna Apidianaki2 (1) IMS, Pfaffenwaldring 5B, 70569 Stuttgart, Germany (2) LIMSI-CNRS, Rue John von Neumann, Campus Universitaire d’Orsay Bât 508, 91405 Orsay Cedex, France Lonneke.vanderPlas@ims.uni-stuttgart.de, Marianna.Apidianaki@limsi.fr Résumé. Nous abordons la question du transfert d’annotations sémantiques, et plus spécifiquement d’étiquettes sur les prédicats, d’une langue à l’autre sur la base de corpus parallèles. Des travaux antérieurs ont transféré ces annotations directement au niveau des tokens, conduisant à un faible rappel. Nous présentons une approche globale de transfert qui agrège des informations repérées dans l’ensemble du corpus parallèle. Nous montrons que la performance de la méthode globale est supérieure aux résultats antérieurs en termes de rappel sans trop affecter la précision. Abstract. We address the problem of transferring semantic annotations, more specifically predicate labellings, from one language to another using parallel corpora. Previous work has transferred these annotations directly at the token level, leading to low recall. We present a global approach to annotation transfer that aggregates information across the whole parallel corpus. We show that this global method outperforms previous results in terms of recall without sacrificing precision too much. Mots-clés : transfert inter-langue, annotation sémantique automatique, prédicats, désambiguïsation lexicale, cor- pus parallèles. Keywords: cross-lingual transfer, automatic semantic annotation, predicates, Word Sense Disambiguation, pa- rallel corpora. 1 Introduction There has recently been a large interest in multilingual natural language processing. Several annotation efforts have been devoted to developing resources for different languages, needed for supervised learning (Hajiˇc et al., 2009). However, there is a large number of languages that still lack linguistically annotated resources. For example for French, one of the most important European languages, there exists no corpus with predicate-argument annotations. Predicate-argument annotations are the representation of the grammatically relevant aspects of a sentence meaning. This level of analysis provides a means of expressing a relation between syntactically different sentences, such as the sentence with the transitive verb in (1a) and the one with the intransitive verb in (1b). The semantic label theme expresses the fact that the object in (1a) has the same conceptual relation with the verb as the subject in (1b). (1) a. [AGENT Mary] [REL-STOP.01 stopped] [THEME the car]. b. [THEME The car] [REL-STOP.01 stopped]. Semantic parsing or semantic role labelling refers to the task of automatically labelling predicates and arguments with predicate-argument structure. This task can be divided in two parts. One part refers to the labelling of the predicates with a predicate sense, and the other to the labelling of the arguments with semantic roles. Semantic parsers and the results of cross-lingual annotation transfer are therefore evaluated on both tasks : predicate labelling and role labelling. We focus on predicate labelling in this paper. Cross-linguistically, the predicate-argument structure of a sentence is considered to be more stable than its syntactic form. The English sentence in (2a) can be considered as equivalent to the French sentence in (2b), despite the fact that …

Content truncated to 500 words