Apart from bug fixes, new features are sentence-level extraction (saving
memory) and parallel translation (includes scripts working for Condor clusters that should easily be adaptable for other parallel architectures).
Also, the README underwent significant changes and now features a
step-by-step instruction using the Spanish-English Europarl as example. Check it out at:
http://www.cs.cmu.edu/~zollmann/samt/readme.html
Because people kept asking a lot about how to train our system on BTEC
data, we also included into the README the sequence of commands we used to train and tune our system for our Chinese-English submission of this year's IWSLT evaluation, starting from scratch with the original
IWSLT-supplied files. This should enable everyone to out-of-the-box score 38% BLEU on the 2007 test set.