Abstract | ||
---|---|---|
This paper proposes a novel method integrating multi-level linguistic knowledge for Chinese grapheme-to-phoneme(G2P) conversion. Pronunciation prediction of non-standard words(NSWs) and disambiguation of polyphonic characters are two important issues in Chinese grapheme-to-phoneme conversion. Considering effect of linguistic knowledge, multi-level linguistic cues, including word form, Part-of-Speech (POS), named entity, collocation and syntactic structure, are extracted under a unified syntactic parsing framework and integrated by maximum entropy approach to disambiguate polyphonic characters. Besides, the text normalization is incorporated in this framework to help predict pronunciation of non-standard words. Experiment results show that the proposed method can improve the performance from 95.64% to 99.23%. © 2013 Springer-Verlag Berlin Heidelberg. |
Year | DOI | Venue |
---|---|---|
2013 | 10.1007/978-3-642-42057-3-61 | IScIDE |
Keywords | Field | DocType |
chinese grapheme-to-phoneme conversion,multi-level linguistic knowledge,non-standard words,polyphonic characters,syntactic structure | Pronunciation,Computer science,Artificial intelligence,Natural language processing,Polyphony,Collocation,Syntactic structure,Grapheme,Named entity,Speech recognition,Principle of maximum entropy,Linguistics,Text normalization | Conference |
Volume | Issue | ISSN |
8261 LNCS | null | 16113349 |
Citations | PageRank | References |
0 | 0.34 | 2 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Yi Liu | 1 | 45 | 33.47 |
Xiaojun Chen | 2 | 14 | 6.41 |
Caixia Gong | 3 | 0 | 1.01 |
Xihong Wu | 4 | 279 | 53.02 |