%@LANGUAGE="JAVASCRIPT" CODEPAGE="65001"%> NIST Speech Group Website
Information Technology Lab, Information Access Division NIST: National Institute of Standards and Technology

  • Speech Group Home

  • Benchmark Tests

  • Tools

  • Test Beds

  • Publications

  • Links

  • Contacts
    •

    Dr. Gregory Sanders

    Affiliation (hierarchically arranged)
    Speech Group (894.01), of the
    Information Access Division, (894), of the
    Information Technology Laboratory, of the
    National Institute of Standards and Technology (NIST), an agency of the
    U.S. Department of Commerce


    Position
    Computer Scientist, Speech Group


    Education
    B.A. Millikin University (Music)
    M.S. and Ph.D. Illinois Institute of Technology (Computer Science)


    Selected Publications
    Gregory A. Sanders, Sebastien Bronsart, Sherri Condon, and Craig Schlenoff, 2008. Odds of Successful Transfer of Low-level Concepts: A Key Metric for Bidirectional Speech-to-speech Machine Translation in DARPA's TRANSTAC Program. Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC-2008), Marrakesh, Morocco (May 28-30, 2008): European Language Resources Association (ELRA), pp. TBD.

    Gregory A. Sanders, and Audrey N. Le, 2004. Effects of Speech Recognition Accuracy on the Performance of DARPA Communicator Spoken Dialogue Systems. International Journal of Speech Technology, 7:293-309.

    Gregory A. Sanders, Audrey N. Le, and John S. Garofolo, 2002. Effects of Word Error Rate in the DARPA Communicator Data During 2000 and 2001. Proceedings of the Seventh International Conference on Spoken Language Processing (ICSLP-2002), Denver, Colorado (Sept. 16-20, 2002): International Speech Communication Association, pp. 277-280.

    Gregory A. Sanders and Jean Scholtz, 2000. Measurement and Evaluation of Embodied Conversational Agents. [Chapter 12 of] Cassell, J., Sullivan, J., Prevost, S., and Churchill, E., eds., 2000. Embodied Conversational Agents. MIT Press, 2000, ISBN 0-262-03278-3.


    Dissertation
    Sanders, G. A., 1995. Generation of Explanations and Multi-Turn Discourse Structures in Tutorial Dialogue, based on Transcript Analysis. Unpublished doctoral dissertation, Illinois Institute of Technology, Chicago, Illinois.

    Perhaps the main effect of my dissertation was its introduction of the term "Directed Line of Reasoning" (or DLR), which has since been adopted by other researchers. I was certainly not the first to describe the phenomenon, but I think I was the first to explain (in detail) how an intelligent tutoring system could produce them, and do so covering the appropriate material, at the appropriate time, in the appropriate form, for the appropriate purposes.

    In a tutoring dialogue, a "Directed Line of Reasoning" is a series of bite-sized leading questions from the tutor and answers from the student, through which the tutor intends to evoke correct cause-and-effect reasoning from the student, based on material the student plausibly already knows. DLRs thus appear when the student plausibly already knows all the steps. DLRs play various roles in tutoring sessions. They may serve as hints, where the tutor leads the student toward (or even to) something the student did not manage to produce (put together) without help. They may serve as a summary, allowing the tutor to verify that the student really knows all the steps, and reviewing the sequence of steps for the student (a useful tactic after a complicated stretch of tutoring the individual steps). When a student already "almost" knows the content, they often serve as an explanation, or even as a method of exposition. The key merit of a DLR as a tutoring tactic is that it requires the student to play a maximally active role, and, when done well, a DLR requires the student to produce essentially the entire cause-and-effect explanation independently. Gregory Hume has analyzed the uses of this tactic in great depth.


    Research Interests:
    • Evaluation of Machine Translation
    • Speech Understanding
    • Dialogue and discourse generation
      In studying dialogue, I'm particularly interested in
      • How one gets or holds the initiative (control) in a dialogue
          • In a dialogue, one participant has the initiative, with the other participant(s) responding. A typical computer system that participates in a dialogue cannot use some of the ways that a human participant gets and holds the initiative in a dialogue -- such as body language or interrupting with a word fragment. But some of the other ways that human participants get or hold the initiative in a dialogue work unchanged for a computer. How can a computer best go about getting or holding the initiative? From the other side, do people do this differently when they know they are talking to a computer? Further, how does the purpose of the dialogue affect these answers? For example, an intelligent tutoring system that actively controls the session has a different style of interaction than a system intended to provide information to a user who controls the session, and the differences in style result from the difference in purpose.
      • Conversational repair in dialogue
          • Because language is ambiguous, there is no such thing as a system so good at dialogue that errors will never arise. Not even in principle can this be achieved. So repair abilities are crucial to having complex extended dialogues. Computer systems need to have robust conversational abilities, able to deal with repairs and corrections by the user and they need to be able to generate repairs when appropriate. Unfortunately, complex repairs in dialogues with a computer occur most often in response to shortcomings of the computer's own language abilities, and a system that is incompetent at complex repairs can make a bad situation worse. For that reason, simple clarifications and repairs often work best, especially if there are strong conventions about what to say in response to a particular kind of clarification or repair. It's also important to look at ways of avoiding the need for repair. In practice, understanding the human participant's behavior can pay off, but the computer participant's own behavior should sometimes be to cut its losses and simplify the situation as much as possible, so as to maximize the chances of success. How can we achieve these goals? How can we measure success?
      • Evaluating dialogue, particularly natural language dialogue with a computer
    • Text manipulation (editing, formatting, typesetting)
    • The ways in which the meaning of language is rooted in physical experience.
    I did my Ph.D. research at IIT in the CircSim-Tutor project, generating the discourse and dialogue structure of tutorial dialogues.


     

     

    Page Created: August 9, 2007
    Last Updated: April 29, 2008

    Speech Group is part of IAD and ITL
    NIST is an agency of the U.S. Department of Commerce
         		Privacy Policy | Security Notices|
    Accessibility Statement | Disclaimer | FOIA