Discourse connectives: Theoretical models and empirical validations in humans and computers

Abstract

Discourse connectives are procedural markers of textual cohesion that have long been an object of study in the Geneva school of pragmatics. In this chapter, we argue that Jacques Moeschler’s descriptions of causal connectives have contributed to provide theoretical insights on the nature of their procedural meaning, which have been recently shown to be compatible with models of human cognition from processing and acquisition studies across several languages. We review these studies in Sects. 2 and 3 respectively. In many of his contributions, Jacques Moeschler has also strived to find precise and testable features of connectives, with a potential for empirical validations in computer applications. In Sect. 4, we describe recent attempts to label automatically some of the meanings of connectives, using parallel corpora as training data, and show that this procedure improves their translation by automatic systems.