Semi Cognitive approach to RTE 6 - Using FrameNet for Semantic Clustering

We present a new system, for recognizing textual entailment, known as Sangyan 1 , which tackles the inherently complex syntactic and semantic ambiguities involved in RTE Task by making use of multiple techniques of text processing. Sangyan employs Syntactic Dependency Tree Match for recognizing syntactic similarity with the aid of an anaphora resolution system, an in-house developed NER system etc. Since syntactic matching techniques cannot map the semantically same but syntactically different constructs, Sangyan utilizes certain heuristics that attempt to bring the different dependency tags together under the umbrella of equivalent semantics. To handle semantic variability, instead of a typical rule based approach, Sangyan uses “FrameNet Frames” and Shalmanesar semantic parser for „Semantic Clustering‟ of words into groups. To improve coverage, we‟ve added many new frames over and above the ones provided by the FrameNet after considerable experimentation over the past RTE data and the RTE6 Development Sets.