An English-persian Automatic Speech Translator: Recent Developments in Domain Portability and User Modeling

In this paper we describe the English-Persian speech to speech translation device, and provide insight on the lessons learned during the first phase of development of this system. We start by giving an overview of the underlying components of the device: the front end ASR, the machine translation system and the speech generation system. Challenges such as the sparseness of available spoken language data and solutions that have been employed to maximize the obtained benefits from using these limited resources are described. In addition, we present results on system evaluation, which identify that a large degree of the system errors can not be attributed to machine performance, but rather to user operation and process losses. The identification of this large margin of improvement led to the initiation of a new research field: user modeling in cross-lingual mediated communication, and we describe our first results in user type classification.