2015-12-02 UTC
# 
dhalgren` kevinmarks_: yeah. and mostly phrase-based. Gets beaten in certain pairs despite its massive data advantage. Like en -> czech is consistently won by a czech system, incorporating lots of old-fashioned linguistic syntactic analysis (tectoML, transfer-based, using ideas of some prague school of syntax), a google-ish phrase-based system and a rules-based "fixer" of common mistakes. Chimera, they call the resulting monstrosity.