Interpretese vs. Translationese

Code and data for the paper "Interpretese vs. Translationese: The Uniqueness of Human Strategies in Simultaneous Interpretation" published in NAACL 2016.

Dataset

Please see README in dat.

Dependencies

You need to install Vowpal Wabbit for classification.

Run analysis scripts

The main script is src/count_statistics.py.

The following arguments are required to load the data:

--src ja --tgt en --mt_file dat/processed/monologue.mt.en.tok.tagged --si_file dat/processed/monologue.si.en.tok.tagged --src_file dat/processed/monologue.mt.ja.tok.all --mt_align_file dat/processed/monologue.mt.align --si_align dat/processed/monologue.si.align

You can then specify the analysis you want to perform. For example, to compare passive voice used in Interpretese and Translationese, run

PYTHONPATH=./lib python src/count_statistics.py --src ja --tgt en --mt_file dat/processed/monologue.mt.en.tok.tagged --si_file dat/processed/monologue.si.en.tok.tagged --src_file dat/processed/monologue.mt.ja.tok.all --mt_align_file dat/processed/monologue.mt.align --si_align dat/processed/monologue.si.align --passive

Use --help to see other types of analysis supported.

If you use the code or the dataset, please cite

@inproceedings{he2016interpretese,
    title={Interpretese vs. Translationese: The Uniqueness of Human Strategies in Simultaneous Interpretation},
    author={He He and Jorday Boyd-Graber and Hal {Daum\'e III}},
    booktitle={NAACL}
    year=2016
}