Machine Translation Again?

Machine translation (MT) remains the paradigm task for natural language processing (NLP) since its inception in the 1950s. Unless NIP can succeed with the central task of machine translation, it cannot be considered successful as a field. We maintain that the most profitable approach to MT at the present time is an interlingual and modular one. MT is one the precious few computational tasks falling broadly within artificial intelligence (AI) that combine a fundamental intellectual research challenge with enormous proven need. To establish the latter, one only has to note that in Japan alone the current MT requirement is for 20 billion pages a year (a market of some $66 billion a year).