SAGAN: A Machine Translation Approach for Cross-Lingual Textual Entailment

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 721–726, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics SAGAN: A Machine Translation Approach for Cross-Lingual Textual Entailment Julio Castillo1,2 and Marina Cardenas2 1UNC-FaMAF, Argentina 2UTN-FRC, Argentina {jotacastillo, ing.marinacardenas}@gmail.com Abstract This paper describes our participation in the task denominated Cross-Lingual Textual En- tailment (CLTE) for content synchronization. We represent an approach to CLTE using machine translation to tackle the problem of multilinguality. Our system resides on ma- chine learning and in the use of WordNet as semantic source knowledge. Results are very promising always achieving results above mean score. 1 Introduction This paper describes the participation of Sagan, a TE and CLTE system, in the new task of Cross Lingual Textual Entailment for Content Synchro- nization. The objective of the Recognizing Textual En- tailment (RTE) task (Dagan et al., 2006) is deter- mining whether the meaning of a text fragment that we call hypothesis H can be inferred from another text fragment T. In this manner, we say that T en- tails H, if a person reading T would infer that H is most likely true. Thus, this definition assumes common human understanding of language and common background knowledge. In that context, Cross-Lingual Textual Entail- ment addresses textual entailment recognition in the challenging application scenario of content synchronization. Thus, CLTE constitutes a gener- alization of Textual Entailment task (also Mono- lingual Textual Entailment) , but envisioning a larger number of application areas in NLP, includ- ing question answering, information retrieval, in- formation extraction, and document summariza- tion, across different languages. Content synchronization could be used to keep consistence among documents written in different languages. For example, a CLTE system can be used in Wikipedia articles to inform lectors which information is absent or inconsistent in comparison to other page in a different language. This new task has to face more additional issues than monolingual TE. Among them, we emphasize the ambiguity, polysemy, and coverage of the re- sources. Another additional problem is the necessi- ty for semantic inference across languages, and the limited availability of multilingual knowledge resources. The CLTE for content synchronization specifi- cally consist on determining the entailment rela- tionship between two text fragment T1 and T2 which are assumed belong a related topic. Four alternatives are possible in this relation- ship: - Bidirectional : It is a semantic equivalence be- tween T1 and T2. - Forward : It is an unidirectional entailment from T1 to T2. - Backward: It is an unidirectional entailment from T2 to T1. - No Entailment: It means that there is no en- tailment between T1 and T2. The paper is organized as follows: Section 2 de- scribes the relevant work done on cross-lingual textual entailment and related tasks, Section 3 de- scribes the architecture of the system, then Section 4 shows experiments and results; and finally Sec- 721 tion 5 summarize some conclusions and future work. 2 Related work In this section we briefly describe two tasks that are closely …

Content truncated to 500 words