Implementation of BiLSTM-CRF model with adversarial training.
Paper: Robust Multilingual Part-of-Speech Tagging via Adversarial Training (NAACL 2018).

• Python 2.7
• Theano 1.0
• Lasagne

• Penn Treebank - Wall Street Journal
• Universal Dependencies (UD) v1.2
• GloVe word embeddings
• Polyglot word embeddings

Configure and run multi_lingual_run_blstm-blstm-crf_pos.sh.

If you use this tool for your work, please consider citing:

@InProceedings{Yasunaga&al.18.naacl,
  author =  {Michihiro Yasunaga and Jungo Kasai and Dragomir R. Radev},
  title =   {Robust Multilingual Part-of-Speech Tagging via Adversarial Training},
  year =    {2018},  
  booktitle =   {Proceedings of NAACL},  
  publisher =   {Association for Computational Linguistics},
}

This tool uses the following open source component (big thank you to the developers). You can find its source code and license information below.

• LasagneNLP: https://github.com/XuezheMax/LasagneNLP (Apache License 2.0)