Corpus-Based Statistics-Oriented (CBSO) Machine Translation Researches in Taiwan

A brief introduction to the MT research projects in Taiwan is given in this paper. Special attention is given to the more and more popular corpus-based statistics-oriented (CBSO) approaches in MT researches. In particular, the parameterized two-way training philosophy in designing the second generation BehaviorTran, which is the first and the largest operational system in this area, is introduced in this paper. 1. Overview of the Machine Translation Projects in Taiwan The first MT research project (the ArchTran MT system, aka the BehaviorTran system) in Taiwan was conducted in 1985 in National Tsing-Hua University [Hsu 86]. Later, MT related projects were subsequently started in many other academic institutions and private sectors, including the Institute of Computer Science and Information Engineering of National Taiwan University, Industry Technology Research Institute (ITRI), Matsushita Electric Institute of Technology (Taiwan), Wang Corporation (Taiwan), Institute of Information Engineering of National Chiao-Tung University, Institute of Information Science, National Tsing-Hua University, and so on. Related NLP researches are also conducted in other Institutions, such as the Institute of Information Science, Academia Sinica and the Telecommunication Technology Laboratory. Because of the blooming growth in the related researches, an annual conference, the ROCLING conference, was initiated in 1988; the ROCLING Society (ROC Computational Linguistics Society) was also founded as a regular organization for promoting various NLP researches, including machine translation. Many workshops were called in a non-regular basis since the foundation of the ROCLING Society. SIGMT for machine translation, and special MT workshops, in particular, were also held for people who are interested in this technical field of research. Since the BehaviorTran MT system is the first MT project in this area, and it is also the largest operational system in this area, its working experiences have significant influence on the other related researches. In fact, the adoption of corpus-based statistics-oriented approaches