Maintenance release, fixed bugs when detecting existence of files on HDFS.

Download on Source Forge

Maintenance release, fixed bugs that may affect interaction with LoonyBin.

Download on Source Forge

Maintenance release, fixed bugs:

1. When –nocopy parameter is specified, –src and –tgt parameters cannot be omitted.
2. When –overwrite is not specified, source/target corpus will will not be combined properly.

Download

Maintenance release, fixed bugs:

1. Merging word alignment step may lost sentences
2. Empty sentences will crash the training procedure

Added filtering functionality so sentences that are too long or empty will be filtered out.

Please note, this version REQUIRES to work with MGIZA version 0.6.2.

Download

Maintenance release, fixed bugs:

1. Lexicon probability may be overestimated phrase pairs with (very) different alignments
2. Translation of Lexicon words –> Null may underflow for large corpus

Changed the feature orders in the final Moses-style phrase table so that the feature weights tuned on Moses can directly apply.

Download

Included integration of MGIZA to train word alignment model in parallel.

Download