Spacos is a system for finding word alignments in parallel corpora. It is based on Bayesian versions of the IBM alignment models, and uses Gibbs sampling for inference. In addition, Spacos is capable of jointly learning word alignments and transfering part-of-speech tags, which is useful when trying to align one well-resourced language with an under-resourced language that lacks basic natural language processing tools.

There is a new, considerably faster version of this tool:
efmaral (EFficient MARkov chain ALigner)

Downloads:
Main executable (spacos.tar.gz)
Part-of-speech tagger models (optional) (spacos-pos.tar.gz)
Source code (spacos-src.tar.gz)

Contact: Robert Östling