Encoding Tree Pair-Based Graphs in Learning Algorithms: The Textual Entailment Recognition Case

In this paper, we provide a statistical machine learning representation of textual entailment via syntactic graphs constituted by tree pairs. We show that the natural way of representing the syntactic relations between text and hypothesis consists in the huge feature space of all possible syntactic tree fragment pairs, which can only be managed using kernel methods. Experiments with Support Vector Machines and our new kernels for paired trees show the validity of our interpretation.