Toolkit for next generation spoken dialogue systems
The emerging human language technologies are increasingly becoming a reality in our lives. In particular, the rapidly evolving dialogue systems that are widely used in call centre applications may provide considerable profits to businesses. While most current dialogue systems involve information seeking dialogues, the real challenge lies on developing command and control, negotiative and tutorial dialogues. To answer this need, the SIRIDUS project focused on specification, interaction and reconfiguration in dialogue-understanding systems. The project provided innovative research and two demonstrator software packages, one of which is the TrindiKit version 3.0. This is an integrated toolset for dialogue researchers that represents the novel Information State Update view. The basic concept behind this is that the dialogue system allows information to be internally stored in information states. According to observed dialogue moves, dialogue move engines allow the user to update the information states by opting for suitable moves. With this approach the architecture provides increased potentialities to explore various combinations of dialogue move engines and information states for implementing flexible spoken dialogue systems. Apart from the general architecture, there are specified formats for defining information states, update rules, dialogue moves, and associated algorithms. By defining the internal structure of the information state, update rules, moves and algorithms the user can build a dialogue move engine. Moreover, the toolkit includes an extensible library of tools for speech recognition and synthesis. TrindiKit version 3 is available at the project website: http://www.ling.gu.se/projekt/siridus/(opens in new window)
