The data here is the usual PBC data, with the metadata information removed (and stored in *-x-bible-newworld2013.train files).

To run the experiments, use the following line:

source run_bible.sh ARGS

For ARGS, see the run_bible.sh script.

Data is formatted line-by-line. The way I have it, each line is a single sentence (split on punctuation). An example of such data (for pretraining) is in /mounts/data/proj/antmarakis/wikipedia/wikipedia_en_el_seq.txt where English and Greek data are concatenated. For the tokenization, separate files are needed for the two languages, again formatted line-by-line.

To run the experiments, use the following line:

source run_wikipedia.sh EN_SIZE OTHER_SIZE OTHER_LANG

After downloading the Public Bible Corpus, run the bash script run.sh (you will need to update the WORKDIR and BIBLEDIR variables) found in the static folder.