Please use this identifier to cite or link to this item: http://scholarbank.nus.edu.sg/handle/10635/78109
Title: Dynamic conditional random fields for joint sentence boundary and punctuation prediction
Authors: Wang, X.
Ng, H.T. 
Sim, K.C. 
Keywords: Dynamic conditional random fields
Punctuation
Sentence boundary detection
Issue Date: 2012
Source: Wang, X.,Ng, H.T.,Sim, K.C. (2012). Dynamic conditional random fields for joint sentence boundary and punctuation prediction. 13th Annual Conference of the International Speech Communication Association 2012, INTERSPEECH 2012 2 : 1382-1385. ScholarBank@NUS Repository.
Abstract: The use of dynamic conditional random fields (DCRF) has been shown to outperform linear-chain conditional random fields (LCRF) for punctuation prediction on conversational speech texts [1]. In this paper, we combine lexical, prosodic, and modified n-gram score features into the DCRF framework for a joint sentence boundary and punctuation prediction task on TDT3 English broadcast news. We show that the joint prediction method outperforms the conventional two-stage method using LCRF or maximum entropy model (MaxEnt). We show the importance of various features using DCRF, LCRF, Max-Ent, and hidden-event n-gram model (HEN) respectively. In addition, we address the practical issue of feature explosion by introducing lexical pruning, which reduces model size and improves the F1-measure. We adopt incremental local training to overcome memory size limitation without incurring significant performance penalty. Our results show that adding prosodic and n-gram score features gives about 20% relative error reduction in all cases. Overall, DCRF gives the best accuracy, followed by LCRF, MaxEnt, and HEN.
Source Title: 13th Annual Conference of the International Speech Communication Association 2012, INTERSPEECH 2012
URI: http://scholarbank.nus.edu.sg/handle/10635/78109
ISBN: 9781622767595
Appears in Collections:Staff Publications

Show full item record
Files in This Item:
There are no files associated with this item.

Page view(s)

20
checked on Feb 24, 2018

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.