Bayesian Approaches to Preference-based Answer Generation in Dialogue (BAPAGiD)

The key innovative aspect of this project is the application of game theoretic methods to the generation of pragmatically optimal answers in a dialogue system. It concentrates on the use of numerical expressions, gradable adjectives and biscuit conditionals (relevance conditionals) in a sales scenario. Each of these linguistic means trigger implicatures that show a sensitivity to interlocutors' (possibly opposing) preferences. These implicatures consist of soft expectations that call for a Bayesian probabilistic approach to modelling. Recent years saw a growing interest in data oriented, probabilistic models in pragmatics. The implementation in a dialog system should help to further refine these models, and to test the sustainability of their theoretical assumptions.



Funding period


Principal Investigator

Dr. Anton Benz