This is the codebase for Neural Unsupervised Reconstruction of Protolanguage Word Forms (ACL 2023). A rough outline of the file contents:

• sampler.py runs baselines and re-implementations of models from previous work
• lstm_model.py runs our neural reconstruction method
• dp.py implements the forward-backward dynamic program
• proposal.py implements the proposal heuristic we use during Metropolis-Hastings
• data.py contains data preprocessing code
• lstm.sh, lstm2.sh, and baselines.sh contains commands used to run experiments and baselines