lmqg/qg_squad

---
license: cc-by-4.0
pretty_name: SQuAD for question generation
language: en
multilinguality: monolingual
size_categories: 10K<n<100K
source_datasets: squad
task_categories:
- text-generation
task_ids:
- language-modeling
tags:
- question-generation
---

# Dataset Card for "lmqg/qg_squad"

## Dataset Description
- **Repository:** [https://github.com/asahi417/lm-question-generation](https://github.com/asahi417/lm-question-generation)
- **Paper:** [https://arxiv.org/abs/2210.03992](https://arxiv.org/abs/2210.03992)
- **Point of Contact:** [Asahi Ushio](http://asahiushio.com/)

### Dataset Summary
This is a subset of [QG-Bench](https://github.com/asahi417/lm-question-generation/blob/master/QG_BENCH.md#datasets), a unified question generation benchmark proposed in
["Generative Language Models for Paragraph-Level Question Generation: A Unified Benchmark and Evaluation, EMNLP 2022 main conference"](https://arxiv.org/abs/2210.03992).
This is [SQuAD](https://rajpurkar.github.io/SQuAD-explorer/) dataset for question generation (QG) task. The split
of train/development/test set follows the ["Neural Question Generation"](https://arxiv.org/abs/1705.00106) work and is
compatible with the [leader board](https://paperswithcode.com/sota/question-generation-on-squad11).

### Supported Tasks and Leaderboards
* `question-generation`: The dataset is assumed to be used to train a model for question generation.
Success on this task is typically measured by achieving a high BLEU4/METEOR/ROUGE-L/BERTScore/MoverScore (see our paper for more in detail).
This task has an active leaderboard which can be found at [here](https://paperswithcode.com/sota/question-generation-on-squad11).

### Languages
English (en)

## Dataset Structure
An example of 'train' looks as follows.
```
{
"question": "What is heresy mainly at odds with?",
"paragraph": "Heresy is any provocative belief or theory that is strongly at variance with established beliefs or customs. A heretic is a proponent of such claims or beliefs. Heresy is distinct from both apostasy, which is the explicit renunciation of one's religion, principles or cause, and blasphemy, which is an impious utterance or action concerning God or sacred things.",
"answer": "established beliefs or customs",
"sentence": "Heresy is any provocative belief or theory that is strongly at variance with established beliefs or customs .",
"paragraph_sentence": "<hl> Heresy is any provocative belief or theory that is strongly at variance with established beliefs or customs . <hl> A heretic is a proponent of such claims or beliefs. Heresy is distinct from both apostasy, which is the explicit renunciation of one's religion, principles or cause, and blasphemy, which is an impious utterance or action concerning God or sacred things.",
"paragraph_answer": "Heresy is any provocative belief or theory that is strongly at variance with <hl> established beliefs or customs <hl>. A heretic is a proponent of such claims or beliefs. Heresy is distinct from both apostasy, which is the explicit renunciation of one's religion, principles or cause, and blasphemy, which is an impious utterance or action concerning God or sacred things.",
"sentence_answer": "Heresy is any provocative belief or theory that is strongly at variance with <hl> established beliefs or customs <hl> ."
}
```

The data fields are the same among all splits.

- `question`: a `string` feature.
- `paragraph`: a `string` feature.
- `answer`: a `string` feature.
- `sentence`: a `string` feature.
- `paragraph_answer`: a `string` feature, which is same as the paragraph but the answer is highlighted by a special token `<hl>`.
- `paragraph_sentence`: a `string` feature, which is same as the paragraph but a sentence containing the answer is highlighted by a special token `<hl>`.
- `sentence_answer`: a `string` feature, which is same as the sentence but the answer is highlighted by a special token `<hl>`.

Each of `paragraph_answer`, `paragraph_sentence`, and `sentence_answer` feature is assumed to be used to train a question generation model,
but with different information. The `paragraph_answer` and `sentence_answer` features are for answer-aware question generation and
`paragraph_sentence` feature is for sentence-aware question generation.

## Data Splits

|train|validation|test |
|----:|---------:|----:|
|75722| 10570|11877|

## Citation Information

```
@inproceedings{ushio-etal-2022-generative,
title = "{G}enerative {L}anguage {M}odels for {P}aragraph-{L}evel {Q}uestion {G}eneration",
author = "Ushio, Asahi and
Alva-Manchego, Fernando and
Camacho-Collados, Jose",
booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2022",
address = "Abu Dhabi, U.A.E.",
publisher = "Association for Computational Linguistics",
}
```

---
许可证: CC-BY-4.0
美观名称: 面向问题生成的SQuAD
语言: 英语（en）
多语言特性: 单语言
规模类别: 10K < n < 100K
源数据集: SQuAD（SQuAD）
任务类别:
- 文本生成
任务子类型:
- 语言建模
标签:
- 问题生成
---

# 数据集卡片："lmqg/qg_squad"

## 数据集说明
- **仓库地址**: [https://github.com/asahi417/lm-question-generation](https://github.com/asahi417/lm-question-generation)
- **论文链接**: [https://arxiv.org/abs/2210.03992](https://arxiv.org/abs/2210.03992)
- **联络人**: [Asahi Ushio](http://asahiushio.com/)

### 数据集概述
本数据集是[QG-Bench](https://github.com/asahi417/lm-question-generation/blob/master/QG_BENCH.md#datasets)的一个子集，QG-Bench是由论文《生成式语言模型（Generative Language Models）用于篇章级问题生成：统一基准与评估》（EMNLP 2022主会议）提出的统一化问题生成基准，论文完整链接为[https://arxiv.org/abs/2210.03992](https://arxiv.org/abs/2210.03992)。

本数据集是面向问题生成（Question Generation, QG）任务的[SQuAD（斯坦福问答数据集）](https://rajpurkar.github.io/SQuAD-explorer/)。其训练集、开发集与测试集的划分遵循《Neural Question Generation》（[https://arxiv.org/abs/1705.00106](https://arxiv.org/abs/1705.00106)）的设定，且可与[SQuAD 1.1排行榜](https://paperswithcode.com/sota/question-generation-on-squad11)兼容。

### 支持任务与排行榜
* `问题生成（question-generation）`: 本数据集用于训练问题生成模型。该任务的性能通常通过BLEU4、METEOR、ROUGE-L、BERTScore、MoverScore等指标的高分表现进行评估（详细说明请参阅本项目论文）。该任务设有活跃排行榜，可通过[此处](https://paperswithcode.com/sota/question-generation-on-squad11)访问。

### 语言
英语（en）

## 数据集结构
「训练集」的单条示例格式如下：
json
{
"question": "异端主要与什么相悖？",
"paragraph": "异端是指任何与既定信仰或习俗强烈相悖的激进信念或理论。异端者是此类主张或信念的支持者。异端与叛教（即明确放弃自身宗教、原则或事业）以及亵渎（即针对上帝或神圣事物的不敬言论或行为）均有所区别。",
"answer": "既定信仰或习俗",
"sentence": "异端是指任何与既定信仰或习俗强烈相悖的激进信念或理论。",
"paragraph_sentence": "<hl> 异端是指任何与既定信仰或习俗强烈相悖的激进信念或理论。 <hl> 异端者是此类主张或信念的支持者。异端与叛教（即明确放弃自身宗教、原则或事业）以及亵渎（即针对上帝或神圣事物的不敬言论或行为）均有所区别。",
"paragraph_answer": "异端是指任何与<hl>既定信仰或习俗<hl>强烈相悖的激进信念或理论。异端者是此类主张或信念的支持者。异端与叛教（即明确放弃自身宗教、原则或事业）以及亵渎（即针对上帝或神圣事物的不敬言论或行为）均有所区别。",
"sentence_answer": "异端是指任何与<hl>既定信仰或习俗<hl>强烈相悖的激进信念或理论。"
}

所有数据划分下的字段均保持一致，各字段说明如下：
- `question`: 字符串类型特征，代表生成的问题
- `paragraph`: 字符串类型特征，代表原始段落
- `answer`: 字符串类型特征，代表段落中的答案片段
- `sentence`: 字符串类型特征，代表包含答案的单句
- `paragraph_answer`: 字符串类型特征，与`paragraph`内容一致，但通过特殊标记`<hl>`高亮了答案所在位置
- `paragraph_sentence`: 字符串类型特征，与`paragraph`内容一致，但通过特殊标记`<hl>`高亮了包含答案的句子
- `sentence_answer`: 字符串类型特征，与`sentence`内容一致，但通过特殊标记`<hl>`高亮了答案所在位置

`paragraph_answer`、`paragraph_sentence`与`sentence_answer`三类特征均可用于训练问题生成模型，但输入信息存在差异：`paragraph_answer`与`sentence_answer`特征用于感知答案的问题生成，而`paragraph_sentence`特征用于感知句子的问题生成。

## 数据划分
| 训练集 | 验证集 | 测试集 |
|-------:|-------:|-------:|
| 75722 | 10570 | 11877 |

## 引用信息
bibtex
@inproceedings{ushio-etal-2022-generative,
title = "{生成式语言模型（Generative Language Models）用于篇章级问题生成：统一基准与评估}",
author = "Ushio, Asahi and Alva-Manchego, Fernando and Camacho-Collados, Jose",
booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2022",
address = "Abu Dhabi, U.A.E.",
publisher = "Association for Computational Linguistics",
}

原始信息汇总

数据集概述

基本信息

• 名称: SQuAD for question generation
• 许可证: cc-by-4.0
• 语言: 英语 (en)
• 多语言性: 单语种
• 大小: 10K<n<100K
• 来源数据集: squad
• 任务类别: 文本生成
• 任务ID: 语言建模
• 标签: 问题生成

数据集描述

• 摘要: 本数据集是QG-Bench的一部分，用于段落级问题生成任务。它是SQuAD数据集的子集，专门用于问题生成任务。训练/开发/测试集的划分遵循“Neural Question Generation”研究，并与leader board兼容。
• 支持的任务和leaderboards:
  • 任务: 问题生成
  • 评估指标: BLEU4/METEOR/ROUGE-L/BERTScore/MoverScore
  • Leaderboard: 链接

数据集结构

• 数据字段:
  • question: 字符串
  • paragraph: 字符串
  • answer: 字符串
  • sentence: 字符串
  • paragraph_answer: 字符串，答案部分用<hl>标记
  • paragraph_sentence: 字符串，包含答案的句子用<hl>标记
  • sentence_answer: 字符串，答案部分用<hl>标记
• 数据分割:
  • train: 75722
  • validation: 10570
  • test: 11877

引用信息

@inproceedings{ushio-etal-2022-generative, title = "{G}enerative {L}anguage {M}odels for {P}aragraph-{L}evel {Q}uestion {G}eneration", author = "Ushio, Asahi and Alva-Manchego, Fernando and Camacho-Collados, Jose", booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing", month = dec, year = "2022", address = "Abu Dhabi, U.A.E.", publisher = "Association for Computational Linguistics", }

搜集汇总

背景与挑战

背景概述

该数据集是用于问题生成任务的英语数据集，基于SQuAD构建，包含约98k条数据，分为训练、验证和测试集。数据以段落、问题、答案和句子等形式组织，支持答案感知和句子感知的问题生成模型训练，常用于自然语言处理中的文本生成研究。