The BerkeleyAligner is a word alignment software package that implements recent innovations in unsupervised word alignment.

News

9/28/09 As of release 2.1, we have split the Berkeley aligner into two downloads. The unsupervised aligner doesn't require a set of hand-labeled word alignments. The supervised aligner does, and it depends on the unsupervised aligner.

Recent changes and bug fixes

9/28 You can now run the unsupervised aligner without a hand-aligned test set; the evaluation phase will be skipped.

9/28 Loading trained models for evaluation only now works correctly (just give an empty training sequence)

9/28 Output can now be split into multiple alignment files corresponding to multiple input files (alignInputsSeparately option)

9/28 The test set does not need to be included in the training sets