# Different Approaches for Modeling Textual Entailment

2160 Words9 Pages

Different approaches for modeling textual entailment have been suggested in the literature. These approaches ranging from shallow approaches like measuring lexical overlap to syntactic parsing and the WordNet relations’ utilization (Bar-Haim et al., 2006b). Most of these approaches apply lexical matching. Some of the approaches represent the text snippets (T and H) as dependency (or syntactic) tree before apply an actual task. Other systems, for solving the T-H entailment problems use the semantic relation such as logical inference. Lately, there has been an activity with respect to more structured meaning representations, abstracting away from the semantically irrelevant surface.
A bag of words (BoW) method is presented by (Glickman, Dagan, & Koppel, 2005), which depend on lexical entailment. This system is comparatively simple, because it does not depend on syntactic or other deeper analysis. Glickman et al. assume that the content word meanings in a H= {u1, …, um} can be assigned truth values. Furthermore, if all lexical components of H are true then H is assumed to be true. As estimating the probability of entailment, they presume that the truth probability of a term in H is independent of the truth of the other terms in H. This corresponds to expect that each word in H will not be entailed from the cumulative context of T as a whole, but from a specific word in T. The authors apply unsupervised empirical estimation of the lexical entailment probabilities depend on word co-occurrence frequencies from the web to find the T-H pairs similarity. Finally, an entailment hold is decided if the estimated entailment probability exceeding a threshold, which empirically tuned, or not.
Adams’s system (Adams, 2006), which based primarily...
... middle of paper ...
...ment (CLTE) as a semantic relation between two different language text portions is investigated by (Mehdad, Negri, Federico, & Trento, 2010). Also, the definition of textual entailment is adapted by define CLTE as follows: “a relation between two natural language portions in different languages, namely a text T (e.g. in English), and a hypothesis H (e.g. in French), that holds if a human after reading T would infer that H is most likely true, or otherwise stated, the meaning of H can be entailed (inferred) from T”. So, two main directions for CLTE can be seen. First, simply bring CLTE back to the monolingual case by translating H into the language of T, or vice versa. Second, try to embed cross-lingual processing techniques inside the textual entailment recognition process. So, the CLTE can be a core technology for several cross-lingual NLP applications and tasks.