Spoken Dialog Management using Dialog Examples



    Abstract

    We are developing a novel generic dialog modeling methodology for multi-domain spoken dialog systems. A generic dialog modeling for the efficient production of various spoken dialog systems is gradually required in order to rapidly dvelop its dialog manager. In this perspective, we developed an Example-based Dialog Modeling (EBDM) technique using dialog examples. EBDM is a simple and effective data-driven method for developing and fast deploying of various spoken dialog systems. We also have developed an example-based error recovery strategies to improve the robustness of EBDM framework. We expect our methodologies to be an alternative data-driven approach comparable to stochastic approaches for more practical deployment of spoken dialog systems.


  • Example-based Dialog Modeling
    • We implemented the EBDM methodology as a generic dialog model by which we can handle different dialog genres and domains.
    EBDM framework
    • Query Generation: The dialog manager makes SQL statement using discourse history and spoken language understanding (SLU) results.
    • Example Search: The dialog manager tries to search semantically close dialog examples in DEDB given the current dialog state. If no example is retrieved, some state variables can be ignored by relaxing particular variables according to the level of importance given the dialog genre and domain.
    • Example Selection: The dialog manager selects the best example to maximize the utterance similarity measure based on lexico-semantic similarity and discourse history similarity.

  • Example-based Error Reocvery Strategy
    • One of the most critical problems is the propagation of errors. We developed the error recovery strategies using dialog examples to handle ASR/SLU errors. The dialog manager detects that something is wrong in the user's utterance and takes immediate turns to address the problem using some help messages.
    • Error recovery strategies can be defined as three types of help messages.
      • UtterHelp: The system gives an utterance template of what the user could say at this point when no example is retrieved.
      • InfoHelp: The system recommends some contents which can be retrieved using the current dialog frame.
      • UsageHelp: The system provides longer help messages to easily use the spoken dialog system.
    • UtterHelp Example

  • Demo Video
    • You can download and play the demo video of POSSDS-HRI (POStech Spoken Dialog System for Human-Robot Interaction on Building Guidance domain) from here
    • You can download and play the demo video of POSUMDM (POStech Unified Multi-domain Dialog Manager) from here
    • You can download and play the demo video of POSSDS-EPG (POStech Spoken Dialog System for Electronic Program Guide Information Access) from here
    • You can download and play the demo video of POSSDS-EPG with inferencing from here
    • You can download and play the demo video of Chatbot from here
    • You can download and play the demo video of Embedded dialog manager from here








Research Page | Top | Home

 

San 31, Hyoja-Dong, Pohang, 790-784, Korea
Phone: +82 54 279 5581 | Fax: +82 54 279 2299