AcTED: Automatic Acquisition of Typical Event Duration for Semi-supervised Temporal Commonsense QA

Data

We provide the generated pseudo-labeled QA data as specified in the paper under data_generation/pseudo_qa_data/. Each file contains different number of events.

Experiments

1. Generate the QA data for prediction with teacher model. Please refer to Section A in data_generation/README.md

2. Fine-tune the model with McTACO-duration and predict the QA data from the previous step. Please refer to Section A in question_answering/README.md

3. Generate the pseudo-labeled QA data. Please refer to Section B in data_generation/README.md

4. Fine-tune the model with pseudo-labeled QA data,. Please refer to Section B in question_answering/README.md

Citation

If you use this dataset, please cite our paper "Felix Virgo, Fei Cheng, Lis Kanashiro Pereira, Masayuki Asahara, Ichiro Kobayashi, Sadao Kurohashi. AcTED: Automatic Acquisition of Typical Event Duration for Semi-supervised Temporal Commonsense QA. arXiv preprint, (2024)"

Bibtex for citations:

@misc{virgo2024acted,
      title={AcTED: Automatic Acquisition of Typical Event Duration for Semi-supervised Temporal Commonsense QA}, 
      author={Felix Virgo and Fei Cheng and Lis Kanashiro Pereira and Masayuki Asahara and Ichiro Kobayashi and Sadao Kurohashi},
      year={2024},
      eprint={2403.18504},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}