Emerging Computational Linguistics

T.J. Hazen (http://people.csail.mit.edu/hazen//) from MIT presented a seminar about their current project for processing the audio portion of college lectures.  The end goal of the project is to help viewers of recorded lectures navigate and search through the lectures more easily. (http://groups.csail.mit.edu/sls/lectures/index.cgi).

The MIT Open Courseware project (http://ocw.mit.edu/index.html) contains the course content for many of the MIT classes.  Many of the classes have been video taped and those AV recordings are available on the OCW site.  Where  SLPP is useful is in trying to find specific items in the lecture tapes.  Without this tool, a student would likely have to view an entire taped session to find a specific section.  This tool will automatically create a summary of the spoken content of the video and also an outline that has links to the specific time where items are talked about.

The reason this project is likely to be viable where other attempts at automatic discourse transcription have not been is that in order to create the outline of the taped presentation, the accuracy of the words matched can be less than ideal.  When a viewer wants to know where a topic is discussed in the tape, they can search for keywords.  The goals of the project do not include the need to transcribe the video tapes, only to find where keywords have been spoken, and then provide a time stamp to that location within the video.This is a link to a paper on the project. http://people.csail.mit.edu/hazen//publications/HLT-2005-demo.pdf Dr. Hazen presented his topic to a meeting of the local IEEE at OGI in Beaverton, OR in January.