TALK is an EC FP6 project, involving the universities of Cambridge, Edinburgh, Gothenburg, Saarbrucken, and Seville, and non-academic partners Bosch, BMW, DFKI, and Linguamatics.
The TALK project focusses on basic research for adaptive multimodal and multilingual human-computer dialogue systems. As a starting point, we share an approach to dialogue management which has been sucessful in a series of projects and applications: the "Information State Update" approach. This methodology will be extended to cover multimodal and multilingual dialogues. We also aim to improve the robustness of dialogue systems by using techniques from machine learning, and to improve the reconfigurability of interfaces by using domain and task representations. The TALK project will build showcases for in-car and in-home information and control, which will demonstrate advances made in basic research.
The main focus of the project at Edinburgh is on Reinforcement Learning of dialogue strategies. Other Edinburgh research involves improving speech recognition via use of contextual information, and programming of devices using dialogue.