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.
-
Generate the QA data for prediction with teacher model. Please refer to
Section A
indata_generation/README.md
-
Fine-tune the model with McTACO-duration and predict the QA data from the previous step. Please refer to
Section A
inquestion_answering/README.md
-
Generate the pseudo-labeled QA data. Please refer to
Section B
indata_generation/README.md
-
Fine-tune the model with pseudo-labeled QA data,. Please refer to
Section B
inquestion_answering/README.md
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}
}