BERT SOTA Baseline

Source code for paper Establishing Strong Baselines for the New Decade: Sequence Tagging, Syntactic and Semantic Parsing with BERT, to be published in The Thirty-Third International Flairs Conference proceedings.

Requirements

• Python>=3.6
• mxnet>=1.4.1
• pyhanlp (optional for preprocessing Chinese)

Datasets

See data. All preprocessing scripts are placed in test/preprocess.

How to Run

All experiment entrypoints are placed in test/script. One example is:

export PYTHONPATH=.:$PYTHONPATH
python3 test/script/ptb_bert_auto.py

References

If you use this repository in your research, please kindly cite our FLAIRS-33 paper:

@inproceedings{bertbaseline,
  title={Establishing Strong Baselines for the New Decade: Sequence Tagging, Syntactic and Semantic Parsing with BERT},
  author={He, Han and Choi, Jinho},
  booktitle={The Thirty-Third International Flairs Conference},
  year={2020}
}