| | --- |
| | license: mit |
| | dataset_info: |
| | features: |
| | - name: premise |
| | dtype: string |
| | - name: hypothesis |
| | dtype: string |
| | - name: relation |
| | dtype: string |
| | - name: id |
| | dtype: string |
| | splits: |
| | - name: train |
| | num_bytes: 791933574 |
| | num_examples: 411452 |
| | - name: dev |
| | num_bytes: 3140558 |
| | num_examples: 2246 |
| | - name: test |
| | num_bytes: 2415937 |
| | num_examples: 2246 |
| | download_size: 11038753 |
| | dataset_size: 797490069 |
| | --- |
| | https://github.com/causalNLP/corr2cause/ |
| |
|
| | The HF dataset provided by the author cannot be directly loaded. We use the NLI subset, which is the most general task. |
| |
|
| | ``` |
| | @article{jin2023can, |
| | title={Can Large Language Models Infer Causation from Correlation?}, |
| | author={Jin, Zhijing and Liu, Jiarui and Lyu, Zhiheng and Poff, Spencer and Sachan, Mrinmaya and Mihalcea, Rada and Diab, Mona and Sch{\"o}lkopf, Bernhard}, |
| | journal={arXiv preprint arXiv:2306.05836}, |
| | year={2023} |
| | } |
| | ``` |
