The Speechalator was developed in part as the next generation of automatic voice translation systems. The Phrasalator is a one-way device that can recognize a set of pre-defined phrases and play a recorded translation, [1]. This device can be ported easily to new languages, requiring only a hand translation of the phrases and a set of recorded sentences. However, such a system severely limits communication as the translation is one way, thus reducing one party's responses to simple pointing and perhaps yes and no.
The Babylon project addresses the issues of two-way communication where either party can use the device for conversation. A number of different groups throughout the US were asked to address specific aspects of the task, such as different languages, translation techniques and platform specifications. The Pittsburgh group was presented with three challenges. First, we were to work with Arabic, a language with which the group had little experience, to test our capabilities in moving to new languages quickly. Second, we were instructed to use an interlingua approach to translation, where the source language is translated into an intermediate form that is shared between all languages. This step streamlines expansion to new languages, and CMU has a long history in working with interlingua based translation systems. Third, we were constrained to one portable PDA-class device to host the entire two-way system: two recognizers, two translation engines, and two synthesizers.