Semantic interpretation of superlative expressions via structured knowledge bases

Abstract

This paper addresses a novel task of se-mantically analyzing the comparative con-structions inherent in attributive superla-tive expressions against structured knowl-edge bases (KBs). The task can be de-fined in two-fold: first, selecting the com-parison dimension against a KB, on which the involved items are compared; and sec-ond, determining the ranking order, in which the items are ranked (ascending or descending). We exploit Wikipedia and Freebase to collect training data in an un-supervised manner, where a neural net-work model is then learnt to select, from Freebase predicates, the most appropriate comparison dimension for a given superla-tive expression, and further determine its ranking order heuristically. Experimen-tal results show that it is possible to learn from coarsely obtained training data to semantically characterize the comparative constructions involved in attributive su-perlative expressions.