americasnlp2024

Checking out the submodule (shared task data)

Navigate to the root directory of this repository, then run

git submodule init

git submodule update

Installing yoyodyne

Please use the forked version at https://github.com/michaelpginn/yoyodyne.

1. You will need to use Python 3.10. You can use pyenv to manage versions.
2. Create a virtual environment (as some of the dependencies are nonstandard) using venv.
3. Navigate to the clone repo and run pip install .
4. Navigate back to the americasnlp2024 repo (ensuring you're still using the right environment and Python and run scripts as needed.

Training Scripts

Training scripts are in scripts/

• train.sh is used to train using the shared task data across all the model architectures supported by yoyodyne
• train_aug.sh is used to train using an additional augmented dataset
• conjoin_dataframes.py is a helper script that joins two dataframes. You can use it to combine multiple augmented datasets.
• copy_preds.py is a helper script used to produce the final predictions tsv file.

Augmenting Data

Any augmented data files should be added to data/augmented/{language}-{strategy}.tsv. The file should not include column headers and should follow the columns Source, Target, Change.