This project is about feature-rich unsupervised word alignment models. At this point we are working on 0th order HMMs, that is, IBM2-like models. We follow Berg-Kirkpatrick et al (2010) and reparameterise IBM2's categorical distributions using exponentiated linear functions (a log-linear parameterisation).
This code-base started as Guido Linder's final project towards his BSc.
Check our build instructions.
For a help message try:
$ python -m lola.aligner --help
Check some examples.
You can train different models with lola
, check our configuration file format.