AssameseQA
收藏资源简介:
Assamese问答数据集(AssameseQA)是一个专门为阿萨姆语设计的抽取式问答数据集,旨在支持多语言Transformer模型(如mBERT、IndicBERT、MuRIL、XLM-RoBERTa)的训练与评估,并推动低资源语言的NLP研究。该数据集包含阿萨姆语的上下文、问题及对应的答案,内容涵盖历史、地理、文化、教育、科学、常识、公司政策、客户支持和信息文章等多个领域。数据集规模在1,000到10,000个样本之间,以CSV格式组织,包含训练集、测试集和完整数据集文件。每个样本包含以下字段:阿萨姆语上下文、阿萨姆语问题、正确答案、答案在上下文中的起始字符位置以及样本是否经过人工验证的标识。数据集由两部分构成:约50%为作者手动创建并验证的原始QA对;另外约50%改编自AI4Bharat IndicQA数据集的阿萨姆语部分,并经过了清洗、标准化和格式统一处理。该数据集适用于抽取式问答、检索增强生成、阿萨姆语对话AI、教育NLP及低资源语言研究,但不适用于医疗诊断、法律咨询、财务决策或安全关键型应用。数据集采用CC BY 4.0许可证发布。
Assamese Question Answering Dataset (AssameseQA) is an extractive question answering dataset specifically designed for the Assamese language, aiming to support the training and evaluation of multilingual Transformer models such as mBERT, IndicBERT, MuRIL, and XLM-RoBERTa, and promote NLP research on low-resource languages. This dataset includes Assamese contexts, questions, and corresponding answers, covering multiple domains including history, geography, culture, education, science, common sense, corporate policies, customer support, and informational articles. The dataset has a size ranging from 1,000 to 10,000 samples, is organized in CSV format, and contains training set, test set, and full dataset files. Each sample encompasses the following fields: Assamese context, Assamese question, correct answer, the starting character position of the answer within the context, and a flag indicating whether the sample has undergone manual verification. The dataset is composed of two parts: approximately 50% are original QA pairs manually created and verified by the authors; the remaining approximately 50% are adapted from the Assamese section of the AI4Bharat IndicQA dataset, and have undergone cleaning, standardization, and format unification processing. This dataset is applicable to extractive question answering, retrieval-augmented generation, Assamese conversational AI, educational NLP, and low-resource language research, but is not suitable for medical diagnosis, legal consultation, financial decision-making, or safety-critical applications. The dataset is released under the CC BY 4.0 license.
Assamese Question Answering Dataset (AssameseQA)
基本信息
- 任务类型:抽取式问答(Extractive Question Answering)
- 语言:阿萨姆语(Assamese,语言代码
as) - 许可证:Creative Commons Attribution 4.0 International (CC BY 4.0)
- 数据集规模:1,000 < n < 10,000 条样本
- 数据格式:CSV 格式
数据集组成
- 约 50% 人工创建:由作者手动编写并验证的阿萨姆语上下文和问答对,涵盖客户支持、公司知识库、阿萨姆历史、教育、技术、科学、常识、日常生活等主题。
- 约 50% 改编自 AI4Bharat IndicQA:仅使用阿萨姆语部分,经过清洗、归一化、格式标准化后整合。
数据文件
仓库包含三个文件:
train.csvtest.csvcombined_dataset.csv
数据列说明
| 列名 | 说明 |
|---|---|
| Context | 阿萨姆语上下文或段落 |
| Question | 阿萨姆语问题 |
| Answer | 正确答案 |
| Starting Point | 答案在上下文中开始的字符位置 |
| Verified | 样本是否经过人工验证 |
示例
| Context | Question | Answer | Starting Point | Verified |
|---|---|---|---|---|
| অসমৰ ৰাজধানী দিছপুৰ। | অসমৰ ৰাজধানী কি? | দিছপুৰ | 10 | Yes |
适用场景
- 微调 mBERT、IndicBERT、MuRIL、XLM-RoBERTa
- 抽取式问答
- 检索增强生成(RAG)
- 阿萨姆语对话式 AI
- 低资源语言 NLP 研究
不适用场景
- 医疗诊断、法律咨询、财务决策、安全关键型应用
潜在局限
- 主题偏向教育和客户支持领域
- 方言多样性有限
- 缺少对话式阿萨姆语
- 高度技术领域样本较少
推荐使用方式
- 抽取式问答系统
- 检索系统
- 多语言 NLP 研究
- 阿萨姆语语言模型
- 学术研究
- 生产系统建议结合其他阿萨姆语语料库
引用信息
使用时请引用 AssameseQA 和 AI4Bharat IndicQA 两个数据集。




