数据集:
cos_e
任务:
子任务:
open-domain-qa语言:
计算机处理:
monolingual大小:
10K<n<100K语言创建人:
crowdsourced批注创建人:
crowdsourced源数据集:
extended|commonsense_qa预印本库:
arxiv:1906.02361许可:
Common Sense Explanations (CoS-E) 允许训练语言模型自动生成解释,这些解释可以在训练和推理中使用在新颖的常识自动生成解释(CAGE)框架中。
"train" 的一个示例如下所示。
{
"abstractive_explanation": "this is open-ended",
"answer": "b",
"choices": ["a", "b", "c"],
"extractive_explanation": "this is selected train",
"id": "42",
"question": "question goes here."
}
v1.11 "train" 的一个示例如下所示。
{
"abstractive_explanation": "this is open-ended",
"answer": "b",
"choices": ["a", "b", "c"],
"extractive_explanation": "this is selected train",
"id": "42",
"question": "question goes here."
}
所有拆分的数据字段都相同。
v1.0| name | train | validation |
|---|---|---|
| v1.0 | 7610 | 950 |
| v1.11 | 9741 | 1221 |
未知。
@inproceedings{rajani2019explain,
title = "Explain Yourself! Leveraging Language models for Commonsense Reasoning",
author = "Rajani, Nazneen Fatema and
McCann, Bryan and
Xiong, Caiming and
Socher, Richard",
year="2019",
booktitle = "Proceedings of the 2019 Conference of the Association for Computational Linguistics (ACL2019)",
url ="https://arxiv.org/abs/1906.02361"
}
感谢 @lewtun , @thomwolf , @mariamabarham , @patrickvonplaten , @albertvillanova 和 @lhoestq 添加此数据集。