Title :
Arabic ASR and MT Integration for GALE
Author :
Al-Onaizan, Yaser ; Mangu, Lidia
Author_Institution :
IBM T.J. Watson Res. Center, Yorktown Heights, NY
Abstract :
In this paper we describe our work in machine translation of Arabic speech into English. This work was done within the context of the GALE research program. We describe several integration techniques between our ASR and MT system. Our initial results suggest that tighter coupling between ASR and MT system improves the translation quality of speech input. We explore the effect of each integration technique on the overall system.
Keywords :
language translation; speech recognition; Arabic ASR-MT integration; Arabic speech; GALE; automatic speech recognition; machine translation; translation quality; Automatic speech recognition; Broadcasting; Cost function; Decoding; Europe; Gaussian processes; Natural languages; Speech recognition; Surface-mount technology; Vocabulary; Arabic Spoken Language Translation; GALE Translation System; Speech Translation; Statistical Machine Translation;
Conference_Titel :
Acoustics, Speech and Signal Processing, 2007. ICASSP 2007. IEEE International Conference on
Conference_Location :
Honolulu, HI
Print_ISBN :
1-4244-0727-3
Electronic_ISBN :
1520-6149
DOI :
10.1109/ICASSP.2007.367312
