Celi: EDITS and Generic Text Pair Classification

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 592–597, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics Celi: EDITS and Generic Text Pair Classification Milen Kouylekov Celi S.R.L. via San Quintino 31 Torino,10121, Italy kouylekov@celi.it Luca Dini Celi S.R.L. via San Quintino 31 Torino,10121, Italy dini@celi.it Alessio Bosca Celi S.R.L. via San Quintino 31 Torino,10121, Italy alessio.bosca@celi.it Marco Trevisan Celi S.R.L. via San Quintino 31 Torino, Italy trevisan@celi.it Abstract This paper presents CELI’s participation in the SemEval The Joint Student Response Anal- ysis and 8th Recognizing Textual Entailment Challenge (Task7) and Cross-lingual Textual Entailment for Content Synchronization task (Task 8). 1 Introduction Recognizing an existing relation between two text fragments received a significant interest as NLP task in the recent years. A lot of the approaches were focused in the filed of Textual Entailment(TE). TE has been proposed as as a comprehensive frame- work for applied semantics (Dagan and Glickman, 2004), where the need for an explicit mapping be- tween linguistic objects can be, at least partially, bypassed through the definition of semantic infer- ences at the textual level. In the TE framework, a text (T) is said to entail the hypothesis (H) if the meaning of H can be derived from the meaning of T. Initially defined as binary relation between texts (YES/NO there is an entailment or there is not) the TE evolved in the third RTE3 (Giampiccolo et al., 2007) challenge into a set of three relations between texts: ENTAILMENT, CONTRADICTION and UNKNOWN. These relations are interpreted as fol- lows: • ENTAILMENT - The T entails the H. • CONTRADICTION - The H contradicts the T • UNKNOWN - There is no semantic connection between T and H. With more and more applications available for recognizing textual entailment the researches fo- cused their efforts in finding practical applications for the developed systems. Thus the Cross-Lingual Textual Entailment task (CLTE) was created using textual entailment (TE) to define cross-lingual con- tent synchronization scenario proposed in (Mehdad et. al., 2011), (Negri et. al., 2011) (Negri et. al., 2012). The task is defined by the organizers as fol- lows: Given a pair of topically related text fragments (T1 and T2) in different languages, the CLTE task consists of automatically annotating it with one of the following entailment judgments: • Bidirectional: the two fragments entail each other (semantic equivalence) • Forward: unidirectional entailment from T1 to T2 • Backward: unidirectional entailment from T2 to T1 • No Entailment: there is no entailment between T1 and T2 The textual entailment competition also evolved. In this year SEMEVAL The Joint Student Response Analysis and 8th Recognizing Textual Entailment Challenge - JRSA-RTE8 (Task7) the textual entail- ment was defined in three subtasks: 5-way task , where the system is required to clas- sify the student answer according to one of the fol- lowing judgments: • Correct, if the student answer is a complete and correct paraphrase of the reference answer; …

Content truncated to 500 words