Char CNN LSTM Part-of-Speech Tagger based on the architecture described in the paper Character-Aware Neural Language Models. The main difference is usage of bidirectional LSTM. Some pieces of code are borrowed from TensorFlow implementation of lstm-char-cnn.
We train this network on OpenCorpora corpus for Russian language. To download the latest version of the corpus execute:
python download_data.py
Then you can train the network by executing the following command:
python train.py