This is a PyTorch implement of 'Simple and Accurate Dependency Parsing Using Bidirectional LSTM Feature Representations'. and 'Arc-Hybrid Non-Projective Dependency Parsing with a Static-Dynamic Oracle'.
It is a transition-based dependency parser for both projective and non-projective trees, using BiLSTM networks and arc-hybrid systems. Although it is just a naive implement of these papers without much optimization, its performance is beyond expectation.
- Python (>= 3.6)
- PyTorch (>= 1.0)
The dataset I used for training is a treebank in Chinese. It is class-use only, so I can't upload it to github. It is preprocessed into 'json' format. Each word in the sentence has these attributes: 'id', 'word', 'pos', 'father', 'emb'.