Abstract

In trying to control various aspects concerning utterance production in multi-party human–computer dialogue, argumentative considerations play an important part, particularly in choosing appropriate lexical units so that we fine-tune the degree of persuasion that each utterance has. A preliminary step in this endeavor is the ability to place an ordering relation between semantic forms (that are due to be realized as utterances, by the machine), concerning their persuasion strength, with respect to certain (explicit or implicit) conclusions. Thus, in this article, we propose a mechanism for assessing utterances, in terms of their argumentative force. The framework designed conflates insights from Asher and Lascarides’ SDRT (“Segmented Discourse Representation Theory”), and from Anscombre and Ducrot’s AT (“Argumentation Theory”). These mechanisms are included in a language generation component of a multi-party dialogue system for book reservation applications (i.e., a “virtual librarian”), and thus evaluated via typical human–machine conversations