README for Models
=================

Currently, two sets of models (trained on tokenized lowercased data from 
WMT09) are provided:  Models trained on Europarl_v4 data, and models trained 
on GigaFrEn data.  The provided language models have been trained on 
Europarl_v4 and news-train08 monolingual data, and the decoder 
parameters tuned on news-dev2009 development data.  The parameters for 
training the translation models should be deducible from the corresponding
config files.

Matti Kaariainen 
Sun Jul 26 20:06:42 EEST 2009