A Learning-Based Account of Phonological Tiers

Code for the 2024 Linguistic Inquiry paper A Learning-Based Account of Phonological Tiers.

@article{belth2024tiers,
  title={A Learning-Based Account of Phonological Tiers},
  author={Belth, Caleb},
  journal={Linguistic Inquiry},
  year={2024},
  publisher={MIT Press},
  url = {https://doi.org/10.1162/ling\_a\_00530},
}

Reproducing Results

The results are in the results/ directory. If you wish to reproduce them, please see the script exp.py in the experiments directory. The script takes two arguments:

--exp-name / -e, type=str (one of Turkish-CHILDES|Turkish-Morpho|Finnish|Latin)
--model / -m type=str, (one of D2L|GR|trigram|TSLIA|GG|LSTM)

exp-name specifies which dataset to produce the results for. model specifies which model to run. The results are saved in the directory results/{exp_name}/{model}.

Running on Your Own Data

D2L is implemented in the Python package algophon. I recommend that implementation for running the model on your own data—just pip install algophon!