Survey of Current Speech Translation Research†

There are no commercial speech translation systems in operation to date, but a number of industrial and government projects are exploring their feasibility. The feasibility of speech translation depends largely on the scope of the application, and ranges from applications that are well within range, such as voice activated dictionaries, to those that will remain impossible for the foreseeable future (e.g., unrestricted simultaneous translation.) Current research therefore aims at milestones between these extremes, namely limited domain speech translation. Such systems restrict the user in what he/she can talk about, and hence constrain the otherwise daunting task of modelling the world of discourse. Nevertheless such systems could be of practical and commercial interest, as they could be used to provide language assistance in common yet critical situations, such as registration for conferences, booking hotels, airlines, car rentals and theatre tickets, ordering food, getting directions, scheduling meetings or in medical doctor-patient situations. If successful, it may also be possible to combine such domains to achieve translation in a class of domains (say, travel). [Waibel 96]