Using argumentation in text generation

Text generation is a field of artificial intelligence aiming at producing computer models of natural language production. This paper discusses the use of the theory of 'Argumentation in Language' in the field of Text Generation. Most text generators follow the same sequence of steps to produce a coherent paragraph: content determination - selecting of information to include in the text from an underlying computer database; content organization - structuring of the propositions to be included in the text according to an appropriate rhetorical plan; finally lexical choice and syntactic realization - phrasing of the information using appropriate lexical items and syntactically correct constructions. The paper provides examples of the use of argumentative features to influence the generator's decisions throughout the generation process. The introduction of argumentative features allows the designer of a text generator to account for linguistic decisions that were not addressed in previous work, such as the selection of judgment determiners (many, few), scalar adjectives (interesting, difficult), connotative verbs (require, necessitate, enjoy, struggle) and connectives (but, therefore). Argumentative features are also used to organize the paragraph in a coherent manner. They are finally used to select appropriate information from a database to satisfy a given persuasive goal. The paper shows how the same descriptive device - the topoi introduced by Anscombre and Ducrot - can serve as the unifying representation level during the whole generation process and build a bridge between the conceptual and linguistic components of the generator.