We implement the Viterbi algorithm in Python and C++, and a Recurrent Neural Network, to evaluate part of speech (POS) tagging.

Final project for LING 227.

The Python Viterbi implementation is in the viterbi_python directory. To specify the dataset, set the UD_BASE variable to the dataset's prefix (without "train.conllu", "dev.conllu", etc. as those are automatically appended). To run, use python viterbi.py.

The C++ Viterbi implementation is in the viterbi_cc directory. Specifics on terminal commands are found in usage.txt. The current executable, Viterbi, is compiled to run the Viterbi algorithm. To run the baseline approach, comment out line 90 in viterbi.cc and uncomment line 89, then run make in the terminal.

The Python RNN (Bidirectional LSTM) implementation is in the rnn directory. To run, upload to Google Colaboratory and run with train, dev, and test sets for all languages in a folder in Google Drive titled Colab Notebooks, then run all cells in notebook in order. Exact accuracy numbers will vary slightly due to randomness in NN initialization.

• English
• Chinese
• Korean
• German
• Spanish
• Indonesian
• Dutch

Kevin Chen, Nick Schoelkopf, Neal Ma