Lecture Search Engine

MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) researchers have developed a lecture search engine that could help students find a specific section of a video recording of a lecture. "Our goal is to develop a speech and language technology that will help educators provide structure to these video recordings, so it's easier for students to access the material," says James Glass, head of the CSAIL Spoken Language Systems Group and principal research scientist. More than 200 MIT lectures are currently available online. Most users are international students who access the lectures through MIT's OpenCourseWare OCW initiative. Searching through the lectures for specific topics is a difficult process, partially because there is no easy way to scan audio like there is with text files. The lecture search engine solves this problem by first creating lecture transcripts using speech recognition software. A major challenge in this process is that the lectures frequently contain technical terms that the software cannot recognize, so the researchers use textbooks, lecture notes, and abstracts to identify key terms and submit them to the computer. Once the transcript is finished, a language processing program divides the transcripts into sections by topic. Sections of text are compared to each other to determine the number of overlapping words between each section. Each word is weighted so the repetition of key words is more important than non-key words. Glass and MIT associate professor Regina Barzilay hope to add a lecture summarization feature to the language processing system, and to get users more involved in the project by adding a Wikipedia-like function that would allow users to correct errors in the lecture transcripts and to add notes.