Sentence comparison using robust minimal recursion semantics and an ontology

Abstract

We design and test a sentence comparison method using the framework of Robust Minimal Recursion Semantics which allows us to utilise the deep parse information produced by Jacy, a Japanese HPSG based parser and the lexical information available in our ontology. Our method was used for both paraphrase detection and also for answer sentence selection for question answering. In both tasks, results showed an improvement over Bag-of-Words, as well as providing extra information useful to the applications.