Solutions to problems inherent in spoken-language translation: the ATR-MATRIX approach

MT Summit VII Sept. 1999 Solutions to Problems Inherent in Spoken-language Translation: The ATR-MATRIX Approach Eiichiro SUMITA, Setsuo YAMADA, Kazuhide YAMAMOTO, Michael PAUL, Hideki KASHIOKA, Kai ISHIKAWA and Satoshi SHIRAI ATR Interpreting Telecommunications Research Laboratories 2-2 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0288, Japan Abstract ATR has built a multi-language speech trans- lation system called ATR-MATRIX. It con- sists of a spoken-language translation sub- system, which is the focus of this paper, to- gether with a highly accurate speech recogni- tion subsystem and a high-definition speech synthesis subsystem. This paper gives a road map of solutions to the problems in- herent in spoken-language translation. Spoken- language translation systems need to tackle difficult problems such as ungrammatical- ity. contextual phenomena, speech recog- nition errors, and the high-speeds required for real-time use. We have made great strides towards solving these problems in recent years. Our approach mainly uses an example-based translation model called TDMT. We have added the use of extra- linguistic information, a decision tree learn- ing mechanism, and methods dealing with recognition errors. 1 Introduction ATR began its study of speech translation in the mid- eighties and has developed a multi-language speech translation system called ATR-MATRIX (ATR's Mul- tilingual Automatic Translation System for Informa- tion Exchange). The speech recognition subsystem of the ATR-MATRIX is highly accurate for sponta- neous speech. The translation subsystem exploits an example-based approach in order to handle spoken- language. The speech synthesis subsystem has succee- ded in high-definition synthesis using a corpus-based approach. This paper features the translation subsys- tem. Please refer to [Takezawa et al., 1999] for infor- mation on the speech recognition and synthesis sub- systems. Spoken-language translation faces problems differ- ent from those of written-language translation. The main requirements are 1) techniques for handling un- grammatical expressions, 2) means for processing con- textual expressions. 3) robust methods for speech recog- nition errors, and 4) real-time speed for smooth com- munication. The backbone of ATR's approach is the translation model called TDMT (Transfer-Driven Machine Trans- lation) [Furuse et al., 1995], which was developed with- in an example-based paradigm. Constituent Bound- ary parsing [Furuse and Iida, 1996] provides efficiency and robustness. We have also explored the processing of contextual phenomena and a method for dealing with recognition errors and have made much progress in these explorations. In the next section, we give a sketch of TDMT. Section 3 presents the contextual processing, section 4 describes the recognition error handling, section 5 explains evaluation measures and the latest perfor- mance. In section 6, we state our conclusions. 2 Sketch of TDMT In TDMT, translation is mainly performed by a trans- fer process that applies pieces of transfer knowledge of the language-pair to an input utterance. The transfer process is the same for each language pair. i.e.. Ja- panese-English, Japanese-Korean, Japanese-German and Japanese-Chinese, whereas morphological analy- sis and generation processes are provided for each lan- guage, i.e., Japanese. English, Korean, German and Chinese. Next, we briefly explain the transfer knowl- edge and transfer process. 2.1 Transfer Knowledge Transfer knowledge …

Content truncated to 500 words