Key points are not available for this paper at this time.
We present a new software architecture for NLP systems made of heterogeneous components, and demonstrate an architectural prototype we have built at ATR in the context of Speech Translation.
Boitet et al. (Sat,) studied this question.