A computational account of comparative implicatures for a spoken dialogue agent

Abstract

Comparative constructions are common in dialogue, especially in negotiative dialogue where a choice must be made between different options, and options must be evaluated using multiple metrics. Comparatives explicitly assert a relationship between two elements along a scale, but they may also implicate positions on the scale especially if constraints on the possible values are present. Dialogue systems must often understand more from a comparative than the explicit assertion in order to understand why the comparative was uttered. In this paper we examine the pragmatic meaning of comparative constructions from a computational perspective.