AIC Seminar Series
Summarizing Spoken Conversations: The Importance of Modeling Joint Activity
|John Niekrasz||Human Communication Research Centre, The University of Edinburgh||[Home Page]|
Notice: Job talk taking place at SRI San Diego. Room and WebEx session at Menlo Park, room TBA. Hosted by Bruce Harris.
Date: Friday September 10, 2010 at 14:00
Location: San Diego. Telecon at EJ228 (SRI E Building). Slides via WebEx from 1:45pm on, sound at 1-888-355-1249, 749045 (Directions)
Conversations are composed of episodes, each of which relate to a
dominant communicative activity of the participants, such as giving
instructions, telling a story, or making a group decision. Modeling
these activities is important because they are part of participants
commonsense understanding of what happens in a conversation. They
appear in natural summaries of conversations such as meeting minutes,
and participants talk about them within the conversation itself.
Dominant joint activities reflect the overall purpose of communication
and therefore represent an essential component of practical,
common-sense descriptions of conversations.
From this starting point, I will present my research on
*activity-oriented* summarization and information extraction for
workplace meetings. The ultimate goal of this research is to develop
methods for *abstraction* of conversations. My approach uses
exclusively unsupervised statistical methods and is fundamentally
based upon the exploitation of context for language understanding. I
hypothesize that activities are principally indicated in language by
expressions that establish a relationship between participants and the
subject matter, e.g., subjective language, reference to participants,
and deixis (context-dependent linguistic forms). In this talk, I will
present results testing this hypothesis on two important problems in
NLP: discourse segmentation and discourse segment labeling.
Segmentation results show that my approach is effective at different
levels and on different corpora, and competitive with state-of-the-art
lexical-semantic approaches. The segment labeling experiments
investigate algorithms that are the first to automatically label
segments according to activity type, e.g., presentation, discussion,
evaluation, in an unsupervised framework. I will also discuss some
points of departure for future work on intentionality and activity as
a contextual factor in automatic language understanding.
Please arrive at least 10 minutes early in order to sign in and be escorted to the conference room. SRI is located at 333 Ravenswood Avenue in Menlo Park. Visitors may park in the visitors lot in front of Building E, and should follow the instructions by the lobby phone to be escorted to the meeting room. Detailed directions to SRI, as well as maps, are available from the Visiting AIC web page.
©2013 SRI International 333 Ravenswood Avenue, Menlo Park, CA 94025-3493