IBM Model One, implemented for Spanish to English

This module is written for python2.7!

Dependencies not found in the standard lib include: NLTK and numpy

optional arguments: -h, --help show this help message and exit -m MODE, --mode MODE train or evaluate -t TARGET, --target TARGET the target language we want to translate to -s SOURCE, --source SOURCE the source language we're translating from --eval_source EVAL_SOURCE source language evaluation data --eval_target EVAL_TARGET target language evaluation data -i ITERATIONS, --iterations ITERATIONS number of iterations to run EM --sentences SENTENCES number of sentences to read from aligned data -e ETA, --eta ETA the value which EM's delta must fall below to be considered converged

typical usage would be something like:

python model_one.py -m evaluate -s corpus.es -t corpus.en --eval_source eval.es --eval_target eval.en --sentences 1000 -i 25

If one leaves out the -i flag, then EM will run until it converges. You can also adjust the value used to determine convergence with the -e flag.

As of right now, functionality to save and load translation tables has not yet been implemented.

If you want to evaluate how well your translation model works, in the test-resources folder, eval.en and eval.es can be useful. The first candidate sentence should be the one returned during an evaluation.