An Extensible Probabilistic Transformation-based Approach to the Third Recognizing Textual Entailment Challenge

Stefan Harmeling

Abstract: We introduce a system for textual entailment that is based on a probabilistic model of entailment. The model is defined using some calculus of transformations on dependency trees, which is characterized by the fact that derivations in that calculus preserve the truth only with a certain probability. We also describe a possible set of transformations (and with it implicitly a calculus) that was successfully applied to the RTE3 challenge data. However, our system can be improved in many ways and we see it as the starting point for a promising new approach to textual entailment.