We use mostly n-gram back-off language models built with the CMU-Cambridge SLM toolkit. We also have an extensive set of local tools for pre-processing our corpora, LM interpolation, LM adaptation, evaluation, etc.
CISD Home Page | URCS Home Page