Combining Linguistic and Statistical Methods for Bi-directional English Chinese Translation in the Flight Domain

Combining Linguistic and Statistical Methods for Bi-directional English Chinese Translation in the Flight Domain Stephanie Seneff, Chao Wang, and John Lee Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology 32 Vassar Street, Cambridge, MA 02139, USA {seneff,wangc,jsylee}@csail.mit.edu Abstract In this paper, we discuss techniques to combine an interlingua translation frame- work with phrase-based statistical meth- ods, for translation from Chinese into En- glish. Our goal is to achieve high-quality translation, suitable for use in language tutoring applications. We explore these ideas in the context of a flight domain, for which we have a large corpus of English queries, obtained from users interacting with a dialogue system. Our techniques exploit a pre-existing English-to-Chinese translation system to automatically pro- duce a synthetic bilingual corpus. Sev- eral experiments were conducted combin- ing linguistic and statistical methods, and manual evaluation was conducted for a set of 460 Chinese sentences. The best per- formance achieved an “adequate” or better analysis (3 or above rating) on nearly 94% of the 409 parsable subset. Using a Rover scheme to combine four systems resulted in an “adequate or better” rating for 88% of all the utterances. 1 Credits This research is supported by ITRI in Taiwan and by the Cambridge MIT Institute. The authors would like to thank Chihyu Chao, Danwen Chen, Mitch Peabody, and Han Shu for their help with data prepa- ration, particularly manual translation of English queries, and for rating translation outputs. 2 Introduction For over two decades, our group at MIT has been developing multilingual conversational systems as a natural user interface to on-line databases. Within the last few years, our interest has broadened to- wards the idea of configuring these dialogue sys- tems as a means for a student of a foreign language to practice and perfect conversational skills through spoken interaction with the system. In this language- learning mode, the student would be able to obtain translation assistance at any time over the course of a dialogue, by simply speaking a sentence in their native language with equivalent meaning. The sys- tem is then tasked with the challenging requirement to provide a fluent translation of their utterance. Two application domains where we have invested considerable previous effort, both towards multi- lingual dialogue interaction and towards translation assistance between English and Chinese, are the weather domain (Zue et al., 2000) and the flight do- main (Seneff and Polifroni, 2000). We have pre- viously reported on various strategies for achieving high quality and enhancing coverage and robustness for bidirectional speech translation in the weather domain (Wang and Seneff, 2006a; Lee and Seneff, 2005) and for translation from English to Chinese in the flight domain (Wang and Seneff, 2006b). This paper is focused on the specific (new) task of trans- lating from Chinese to English in the flight domain. We have found in general, as might be expected, that the flight domain is considerably more difficult than the weather domain, due to the much larger number of attributes that can be specified, as …

Content truncated to 500 words