This work was partially funded by a grant N66001-00-C-8007 under the DARPA Babylon program: ``Mobile Speech-to-Speech Translation for Military Field Application.'' The opinions expressed in this paper do not necessarily reflect those of DARPA.
