What should we do next for MT system development

Machine translation (MT) research and development began at the end of 1950’s when not only natural language processing (NLP) technology but also linguistic theory was at a primitive level. Given the restricted memory sizes and computing power at that time, MT presented one of the most difficult and challenging research themes of the day. Thus. MT researchers and developers were forgiven when they complained that their poor translation results were due to shortages in memory and computing power. But now, we cannot say such things. About 40 years have passed since then and computing power and memory capacities have increased dramatically in that time. Many new NLP technologies and linguistic theories have been proposed, based on which MT systems have been developed. Consequently, the scale of MT has grown and many MT s y s t e m s a r e available at affordable prices. However, even though the domain of almost all curren t MT systems is highly restricted, more improvements are necessary, since their translation results are still unsatisfactory. As before, high quality MT remains a difficult and challenging research theme. In the 1990's, information networks have diffused throughout the world, enhancing the importance of MT systems. Through the Internet, we find ourselves surrounded by an enormous amount of documents written in many languages, such that we feel the urgent need for multilingual translation systems. MT technology is now considered a key technology in the field of Internet-based information retrieval. In this paper, after reviewing MT research history focusing on MT technology (not individual MT projects). I discuss what we should do in the next century.