This is the data and code repository for our ACL 2023 paper “Generic Temporal Reasoning with Differential Analysis and Explanation”.
TODAY is our crowdsourced dataset. The TODAY dataset and its overall framework are designed to evaluate systems’ ability to make temporal predictions with plausible reasons.
We include the dataset under data/
.
We provide our codebase to reproduce the experiment results reported in the paper. All models can be found on this page.
Download the entire directory ptntime-pretrained-model
and put it under model/ptntime-pretrained-model/
.
Download the entire directory best_model_checkpoint
and put it under model/best_model_checkpoint/
.
Work under code/
directory (This is very important as we refer all paths relative to this working directory below).
Install requirements by pip install -r requirement.txt
To run and train the best ptntime model, use sh train_ptntime.sh
To run and train the best T5 model, use sh train_t5.sh
To run the baseline without TODAY, use sh train_ptntime_wo_today.sh
To test the model, use sh inference.sh
GPT_today.ipynb
.We pre-trained the explanation sentence verifier. Download the entire directory ptntime_explanation_verifier
and put it under model/ptntime_explanation_verifier/
.
To train the general and additional sentence verifiers, use sh train_ptntime_verifier.sh
.
sh distill_ptntime.sh
.See the following paper:
@inproceedings{feng-etal-2023-generic,
title = "Generic Temporal Reasoning with Differential Analysis and Explanation",
author = "Feng, Yu and
Zhou, Ben and
Wang, Haoyu and
Jin, Helen and
Roth, Dan",
booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.acl-long.671",
pages = "12013--12029",
}