Introduction

This is the data and code repository for our ACL 2023 paper "Generic Temporal Reasoning with Differential Analysis and Explanation".

TODAY

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.

Dataset

We include the dataset under data/.

Models and Experiments

We provide our codebase to reproduce the experiment results reported in the paper. All models can be found on this page.

Pre-trained Models

• 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/.

Run Experiments

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

Generate LLM Supervision Signals

Generate GPT-3.5 Data

• Please follow the instructions and code in GPT_today.ipynb.

Train Verifiers

• 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.

Distill GPT-3.5 Data

• To distill the data, and further train with the distilled GPT-3.5 data, use sh distill_ptntime.sh.

Citation

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",
   }