Char CNN LSTM Part-of-Speech Tagger

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.

Running

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