Analysis and comparison of discriminative training objectives

Author

Li, Qi

Author_Institution

Li Creative Technol., Inc., NJ, USA

fYear

2003

fDate

26-29 Oct. 2003

Firstpage

545

Lastpage

548

Abstract

The minimum classification error (MCE) and maximum mutual information (MMI) objectives for discriminative training in automatic speech recognition and natural language processing are analyzed and compared theoretically. The results show that both objectives are related to posterior probability and error rates, and the MCE objective is more general and flexible than the MMI objective. The relations between the objectives and parameter optimization methods are also discussed. The results can help in understanding the discriminative objectives, in developing new objectives, and in discovering new training algorithms jointly with objectives.

Keywords

error statistics; maximum likelihood estimation; natural languages; pattern classification; probability; speech recognition; automatic speech recognition; discriminative training objective analysis; maximum mutual information; minimum classification error; natural language processing; parameter optimization; probability; Automatic speech recognition; Error analysis; Natural languages; Optimization methods; Robustness; Speaker recognition; Speech analysis; Speech processing; Speech recognition; Tiles;

fLanguage

English

Publisher

ieee

Conference_Titel

Natural Language Processing and Knowledge Engineering, 2003. Proceedings. 2003 International Conference on

Conference_Location

Beijing, China

Print_ISBN

0-7803-7902-0

Type

conf

DOI

10.1109/NLPKE.2003.1275964

Filename

1275964