DirRelCond3: Detecting Textual Entailment Across Languages With Conditions On Directional Text Relatedness Scores

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 710–714, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics DirRelCond3: Detecting Textual Entailment Across Languages With Conditions On Directional Text Relatedness Scores Alp´ar Perini ‘Babes¸-B´olyai’ University Cluj-Napoca, Romania palpar at gmail.com Abstract There are relatively few entailment heuristics that exploit the directional nature of the entail- ment relation. Cross-Lingual Text Entailment (CLTE), besides introducing the extra dimen- sion of cross-linguality, also requires to de- termine the exact direction of the entailment relation, to provide content synchronization (Negri et al., 2012). Our system uses sim- ple dictionary lookup combined with heuris- tic conditions to determine the possible di- rections of entailment between the two texts written in different languages. The key mem- bers of the conditions were derived from (Cor- ley and Mihalcea, 2005) formula initially for text similarity, while the entailment condition used as a starting point was that from (Tatar et al., 2009). We show the results obtained by our implementation of this simple and fast approach at the CLTE task from the SemEval- 2012 challenge. 1 Introduction Recognizing textual entailment (TE) is a key task for many natural language processing (NLP) prob- lems. It consists in determining if an entailment re- lation exists between two texts: the text (T) and the hypothesis (H). The notation T → H says that the meaning of H can be inferred from T, in order words, H does not introduce any novel information with re- spect to T. Even though RTE challenges lead to many ap- proaches for finding textual entailment, fewer au- thors exploited the directional character of the en- tailment relation. Due to the fact that the entailment relation, unlike the equivalence relation, is not sym- metric, if T → H, it is less likely that the reverse H → T can also hold (Tatar et al., 2009). The novel Cross-Lingual Text Entailment (CLTE) approach increases the complexity of the traditional TE task in two way, both of which have been only partially researched and have promise for great po- tential (Negri et al., 2012): • the two texts are no longer written in the same language (cross-linguality); • the entailment needs to be queried in both di- rections (content synchronization). Mehdad et al. (2010) presented initial research di- rections and experiments for the cross-lingual con- text and explored possible application scenarios. 2 Theoretical Background The semantic similarity formula from (Corley and Mihalcea, 2005) defines the similarity of a pair of documents differently depending on with respect to which text it is computed. The formula involves only the set of open-class words (nouns, verbs, adjectives and adverbs) from each text. Based on this text-to-text similarity metric, Tatar et al. (2009) have derived a textual entailment recog- nition system. The paper demonstrated that in the case when T → H holds, the following relation will take place: sim(T, H)H > sim(T, H)T (1) however, the opposite of this statement is not always true, nevertheless it is likely. In (Tatar …

Content truncated to 500 words