This is the code repository for the ACL paper QuASE: Question-Answer Driven Sentence Encoding. If you use this code for your work, please cite

@inproceedings{HeNgRo20,
    author = {Hangfeng He and Qiang Ning and Dan Roth},
    title = {{QuASE: Question-Answer Driven Sentence Encoding}},
    booktitle = {Proc. of the Annual Meeting of the Association for Computational Linguistics (ACL)},
    year = {2020},
}

Use virtual environment tools (e.g miniconda) to install packages and run experiments
python>=3.6
install pytorch
pip install pytorch-pretrained-bert==0.6.2
pip install allennlp==0.9.0

Our pre-trained models can be found in the google drive.

Change the /path/to/data (/path/to/models) to your data (models) dir.

To reproduce our experiments based on the scripts:

sh scripts/run_script.sh
sh scripts/run_BERT_MRC.sh (an example)

To reproduce the experiments based on the allennlp (go into the allennlp-experiments dir):

allennlp train /path/to/model/configuration -s /path/to/serialization/dir --include-package models
allennlp train coref_bert.jsonnet -s coref-bert --include-package models (an example)

To reproduce the experiments based on the flair (go into the flair-experiments dir):

python ner_flair.py
python ner_flair_semanticbert.py

To train a s-QuASE/p-QuASE with your own QA data (SQuAD format):

sh scripts/run_SemanticBERT_SQuAD.sh
sh scripts/run_BERT_squad_51K.sh

SRL Evaluation:
The SRL metric implementation (SpanF1Measure) does not exactly track the output of the official PERL script (is typically 1-1.5 F1 below), and reported results used the official evaluation script (allennlp-experiments/run_srl_evaluation.sh).