Fuzzy semantic measurement for synonymy and its application in an automatic question-answering system

We present a novel methodology for the representation of sentences by fuzzy semantics, which is applied to the measurement of synonymy. The novelty of this methodology lies in a new way of dealing with the semantics of words and their functions in a sentence. Through the concept of "information mass", a fuzzy semantic construct, a multidimensional information mass structure of a sentence is realized. The synonymy between sentences is then measured in terms of sentential information mass. We show how to measure the semantic closeness between sentences in order to cluster questions in an FAQ database and how to match a user's question to the closest database record. Experiment is done with a database of FAQ concerning intellectual property (patents and copyrights).