Igreat

The main functionality of iGREAT is implemented by two programs:

giati -> a tool to train an SFST from a word-aligned parallel corpus

refx -> a tool to translate a test corpus by means of a trained SFST

Additionally, iGREAT includes two more tools for a log-linear approach to SMT:

arpa2tr -> a tool to turn ARPA language models into SFSTs

local -> several phrase-based local features are computed for a given

set of bilingual phrase pairs: phrase-based lexical

probabilities (direct and inverse lexical translation models)and a target-sentence length model.

simplex -> a tool to train the set of weights within a log-linear approach