Multilingual Supervision of Semantic Annotation

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1007–1017, Osaka, Japan, December 11-17 2016. Multilingual Supervision of Semantic Annotation Peter Exner Marcus Klang Lund University Department of Computer Science Lund, Sweden {Peter.Exner, Marcus.Klang, Pierre.Nugues}@cs.lth.se Pierre Nugues Abstract In this paper, we investigate the annotation projection of semantic units in a practical setting. Previous approaches have focused on using parallel corpora for semantic transfer. We evaluate an alternative approach using loosely parallel corpora that does not require the corpora to be exact translations of each other. We developed a method that transfers semantic annotations from one language to another using sentences aligned by entities, and we extended it to include alignments by entity-like linguistic units. We conducted our experiments on a large scale using the English, Swedish, and French language editions of Wikipedia. Our results show that the annotation projection using entities in combination with loosely parallel corpora provides a viable approach to extending previous attempts. In addition, it allows the generation of proposition banks upon which semantic parsers can be trained. 1 Introduction Data-driven approaches using natural language processing tackle increasingly complex tasks with ever growing scales and in more varied domains. Semantic role labeling is a type of shallow semantic pars- ing that is becoming an increasingly important component in information extraction (Christensen et al., 2010), question answering (Shen and Lapata, 2007), and text summarization (Khan et al., 2015). The development of semantic resources such as FrameNet (Baker et al., 1998) and PropBank (Palmer et al., 2005) made the training of models for semantic role labelers using supervised techniques possible. However, as a consequence of the considerable manual efforts needed to build proposition banks, they exist only for a few languages. An alternative approach to using supervision is to transfer knowledge between resources, a form of distant or related supervision. Methods for directly projecting semantic labels from a resource-rich language to a resource-scarce one were introduced in Pad´o (2007). In this paper, we describe a method for aligning and projecting semantic annotation in loosely parallel corpora by using entities and entity-like linguistic units. Our goal is to generate multilingual PropBanks for resource-scarce languages. We used multiple language editions of Wikipedia: An English edition annotated up to a semantic level using the PropBank semantic roles, and syntactically annotated editions of Swedish and French Wikipedias. By aligning Wikipedias by entities, we constructed loosely parallel corpora and we used them to generate PropBanks in Swedish and French. We provide an evaluation of the quality of the generated PropBanks, together with an evaluation on two external FrameNets. 2 Previous Work As an alternative to using supervised efforts for relation extraction, distant supervision can be employed to transfer relational knowledge representations from one resource to another. Distant supervision for relation extraction was introduced by Craven and Kumlien (1999) in the context of biomedical informa- tion extraction. Mintz et al. (2009) describe a method of using an external knowledge base as an indirect way of annotating text. Hoffmann et …

Content truncated to 500 words