interspeech2010_xie

interspeech2010_xie - Semi-Supervised Extractive Speech...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Semi-Supervised Extractive Speech Summarization via Co-Training Algorithm Shasha Xie 1 , Hui Lin 2 , Yang Liu 1 1 Department of Computer Science, The University of Texas at Dallas, U.S.A 2 Department of Electrical Engineering, University of Washington, U.S.A { shasha,yangl } @hlt.utdallas.edu, [email protected] Abstract Supervised methods for extractive speech summarization require a large training set. Summary annotation is often ex- pensive and time consuming. In this paper, we exploit semi- supervised approaches to leverage unlabeled data. In particu- lar, we investigate co-training algorithm for the task of extrac- tive meeting summarization. Compared with text summariza- tion, speech summarization task has its unique characteristic in that the features naturally split into two sets: textual fea- tures and prosodic/acoustic features. Such characteristic makes co-training an appropriate approach for semi-supervised speech summarization. Our experiments on ICSI meeting corpus show that by utilizing the unlabeled data, co-training algorithm sig- nificantly improves summarization performance when only a small amount of labeled data is available. Index Terms : extractive meeting summarization, co-training, semi-supervised learning 1. Introduction Automatic meeting summarization is a very useful technique that can help the users to browse a large amount of meeting recordings. In this paper, we investigate extractive summa- rization, in which the most representative segments from the original document are selected and concatenated together to form a final summary. This task can be formulated as a bi- nary classification problem and solved using supervised learn- ing approaches. Each training and testing instance (i.e., a sen- tence) is represented by a set of indicative features, and posi- tive or negative labels are used to indicate whether this sentence is in the summary or not. In previous work, various classifi- cation models have been investigated, such as hidden Markov model (HMM), conditional random field (CRF), maximum en- tropy classifier, and support vector machine (SVM) [1, 2, 3, 4]. Learning a summarization classifier requires a large amount of labeled data for training. Summary annotation is often dif- ficult, expensive, and time consuming. Annotation of meeting recordings is especially hard because the documents to be sum- marized are transcripts of natural meetings that have very spon- taneous style, contain many disfluencies, have multiple speak- ers, and are less coherent in content. It is very hard to read and understand the document, not to mention extracting the summary. On the contrary, meeting recordings and their tran- scripts are relatively much easier to collect. This situation cre- ates a good opportunity for semi-supervised learning that can use large amount of unlabeled data, together with the labeled data, to build better classifiers. This technique has been shown to be very promising in many speech and language processing...
View Full Document

This note was uploaded on 03/06/2012 for the course CIS 630 taught by Professor Cis630 during the Spring '08 term at UPenn.

Page1 / 4

interspeech2010_xie - Semi-Supervised Extractive Speech...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online