google-research-datasets/xquad_r

Name: google-research-datasets/xquad_r
Creator: google-research-datasets
Published: 2024-01-04 17:11:57
License: 暂无描述

Hugging Face2024-01-04 更新2024-06-15 收录

下载链接：

https://hf-mirror.com/datasets/google-research-datasets/xquad_r

下载链接

链接失效反馈

官方服务：

资源简介：

--- annotations_creators: - expert-generated language_creators: - found language: - ar - de - el - en - es - hi - ru - th - tr - vi - zh license: - cc-by-sa-4.0 multilinguality: - multilingual size_categories: - 1K<n<10K source_datasets: - extended|squad - extended|xquad task_categories: - question-answering task_ids: - extractive-qa paperswithcode_id: xquad-r pretty_name: LAReQA config_names: - ar - de - el - en - es - hi - ru - th - tr - vi - zh dataset_info: - config_name: ar features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 1722775 num_examples: 1190 download_size: 263002 dataset_size: 1722775 - config_name: de features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 1283277 num_examples: 1190 download_size: 241957 dataset_size: 1283277 - config_name: el features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 2206666 num_examples: 1190 download_size: 324379 dataset_size: 2206666 - config_name: en features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 1116099 num_examples: 1190 download_size: 212372 dataset_size: 1116099 - config_name: es features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 1273475 num_examples: 1190 download_size: 236874 dataset_size: 1273475 - config_name: hi features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 2682951 num_examples: 1190 download_size: 322083 dataset_size: 2682951 - config_name: ru features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 2136966 num_examples: 1190 download_size: 321728 dataset_size: 2136966 - config_name: th features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 2854935 num_examples: 1190 download_size: 337307 dataset_size: 2854935 - config_name: tr features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 1210739 num_examples: 1190 download_size: 228364 dataset_size: 1210739 - config_name: vi features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 1477215 num_examples: 1190 download_size: 237644 dataset_size: 1477215 - config_name: zh features: - name: id dtype: string - name: context dtype: string - name: question dtype: string - name: answers sequence: - name: text dtype: string - name: answer_start dtype: int32 splits: - name: validation num_bytes: 984217 num_examples: 1190 download_size: 205768 dataset_size: 984217 configs: - config_name: ar data_files: - split: validation path: ar/validation-* - config_name: de data_files: - split: validation path: de/validation-* - config_name: el data_files: - split: validation path: el/validation-* - config_name: en data_files: - split: validation path: en/validation-* - config_name: es data_files: - split: validation path: es/validation-* - config_name: hi data_files: - split: validation path: hi/validation-* - config_name: ru data_files: - split: validation path: ru/validation-* - config_name: th data_files: - split: validation path: th/validation-* - config_name: tr data_files: - split: validation path: tr/validation-* - config_name: vi data_files: - split: validation path: vi/validation-* - config_name: zh data_files: - split: validation path: zh/validation-* --- # Dataset Card for [Dataset Name] ## Table of Contents - [Dataset Description](#dataset-description) - [Dataset Summary](#dataset-summary) - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards) - [Languages](#languages) - [Dataset Structure](#dataset-structure) - [Data Instances](#data-instances) - [Data Fields](#data-fields) - [Data Splits](#data-splits) - [Dataset Creation](#dataset-creation) - [Curation Rationale](#curation-rationale) - [Source Data](#source-data) - [Annotations](#annotations) - [Personal and Sensitive Information](#personal-and-sensitive-information) - [Considerations for Using the Data](#considerations-for-using-the-data) - [Social Impact of Dataset](#social-impact-of-dataset) - [Discussion of Biases](#discussion-of-biases) - [Other Known Limitations](#other-known-limitations) - [Additional Information](#additional-information) - [Dataset Curators](#dataset-curators) - [Licensing Information](#licensing-information) - [Citation Information](#citation-information) - [Contributions](#contributions) ## Dataset Description - **Homepage:** [LAReQA](https://github.com/google-research-datasets/lareqa) - **Repository:** [XQuAD-R](https://github.com/google-research-datasets/lareqa) - **Paper:** [LAReQA: Language-agnostic answer retrieval from a multilingual pool](https://arxiv.org/pdf/2004.05484.pdf) - **Point of Contact:** [Noah Constant](mailto:nconstant@google.com) ### Dataset Summary XQuAD-R is a retrieval version of the XQuAD dataset (a cross-lingual extractive QA dataset). Like XQuAD, XQUAD-R is an 11-way parallel dataset, where each question appears in 11 different languages and has 11 parallel correct answers across the languages. ### Supported Tasks and Leaderboards [More Information Needed] ### Languages The dataset can be found with the following languages: * Arabic: `xquad-r/ar.json` * German: `xquad-r/de.json` * Greek: `xquad-r/el.json` * English: `xquad-r/en.json` * Spanish: `xquad-r/es.json` * Hindi: `xquad-r/hi.json` * Russian: `xquad-r/ru.json` * Thai: `xquad-r/th.json` * Turkish: `xquad-r/tr.json` * Vietnamese: `xquad-r/vi.json` * Chinese: `xquad-r/zh.json` ## Dataset Structure [More Information Needed] ### Data Instances An example from `en` config: ``` {'id': '56beb4343aeaaa14008c925b', 'context': "The Panthers defense gave up just 308 points, ranking sixth in the league, while also leading the NFL in interceptions with 24 and boasting four Pro Bowl selections. Pro Bowl defensive tackle Kawann Short led the team in sacks with 11, while also forcing three fumbles and recovering two. Fellow lineman Mario Addison added 6½ sacks. The Panthers line also featured veteran defensive end Jared Allen, a 5-time pro bowler who was the NFL's active career sack leader with 136, along with defensive end Kony Ealy, who had 5 sacks in just 9 starts. Behind them, two of the Panthers three starting linebackers were also selected to play in the Pro Bowl: Thomas Davis and Luke Kuechly. Davis compiled 5½ sacks, four forced fumbles, and four interceptions, while Kuechly led the team in tackles (118) forced two fumbles, and intercepted four passes of his own. Carolina's secondary featured Pro Bowl safety Kurt Coleman, who led the team with a career high seven interceptions, while also racking up 88 tackles and Pro Bowl cornerback Josh Norman, who developed into a shutdown corner during the season and had four interceptions, two of which were returned for touchdowns.", 'question': 'How many points did the Panthers defense surrender?', 'answers': {'text': ['308'], 'answer_start': [34]}} ``` ### Data Fields - `id` (`str`): Unique ID for the context-question pair. - `context` (`str`): Context for the question. - `question` (`str`): Question. - `answers` (`dict`): Answers with the following keys: - `text` (`list` of `str`): Texts of the answers. - `answer_start` (`list` of `int`): Start positions for every answer text. ### Data Splits The number of questions and candidate sentences for each language for XQuAD-R is shown in the table below: | | XQuAD-R | | |-----|-----------|------------| | | questions | candidates | | ar | 1190 | 1222 | | de | 1190 | 1276 | | el | 1190 | 1234 | | en | 1190 | 1180 | | es | 1190 | 1215 | | hi | 1190 | 1244 | | ru | 1190 | 1219 | | th | 1190 | 852 | | tr | 1190 | 1167 | | vi | 1190 | 1209 | | zh | 1190 | 1196 | ## Dataset Creation [More Information Needed] ### Curation Rationale [More Information Needed] ### Source Data [More Information Needed] #### Initial Data Collection and Normalization [More Information Needed] #### Who are the source language producers? [More Information Needed] ### Annotations [More Information Needed] #### Annotation process [More Information Needed] #### Who are the annotators? [More Information Needed] ### Personal and Sensitive Information [More Information Needed] ## Considerations for Using the Data [More Information Needed] ### Social Impact of Dataset [More Information Needed] ### Discussion of Biases [More Information Needed] ### Other Known Limitations [More Information Needed] ## Additional Information [More Information Needed] ### Dataset Curators The dataset was initially created by Uma Roy, Noah Constant, Rami Al-Rfou, Aditya Barua, Aaron Phillips and Yinfei Yang, during work done at Google Research. ### Licensing Information XQuAD-R is distributed under the [CC BY-SA 4.0 license](https://creativecommons.org/licenses/by-sa/4.0/legalcode). ### Citation Information ``` @article{roy2020lareqa, title={LAReQA: Language-agnostic answer retrieval from a multilingual pool}, author={Roy, Uma and Constant, Noah and Al-Rfou, Rami and Barua, Aditya and Phillips, Aaron and Yang, Yinfei}, journal={arXiv preprint arXiv:2004.05484}, year={2020} } ``` ### Contributions Thanks to [@manandey](https://github.com/manandey) for adding this dataset.

提供机构：

google-research-datasets

原始信息汇总

数据集卡片

数据集描述

数据集摘要

XQuAD-R 是 XQuAD 数据集的检索版本（一个跨语言的抽取式问答数据集）。与 XQuAD 类似，XQUAD-R 是一个 11 种语言的平行数据集，每个问题在 11 种不同语言中出现，并在这 11 种语言中各有 11 个平行正确答案。

支持的任务和排行榜

[更多信息需补充]

语言

数据集包含以下语言：

阿拉伯语
德语
希腊语
英语
西班牙语
印地语
俄语
泰语
土耳其语
越南语
中文

数据集结构

数据实例

以下是 en 配置的一个示例： json { "id": "56beb4343aeaaa14008c925b", "context": "The Panthers defense gave up just 308 points, ranking sixth in the league, while also leading the NFL in interceptions with 24 and boasting four Pro Bowl selections...", "question": "How many points did the Panthers defense surrender?", "answers": { "text": ["308"], "answer_start": [34] } }

数据字段

id (str): 上下文-问题对的唯一 ID。
context (str): 问题的上下文。
question (str): 问题。
answers (dict): 答案，包含以下键：
- text (list of str): 答案的文本。
- answer_start (list of int): 每个答案文本的起始位置。

数据分割

每个语言的 XQuAD-R 数据集的问答对数量如下：

语言	问题数量
ar	1190
de	1190
el	1190
en	1190
es	1190
hi	1190
ru	1190
th	1190
tr	1190
vi	1190
zh	1190

数据集创建

数据集策划理由

[更多信息需补充]

源数据

[更多信息需补充]

标注

[更多信息需补充]

使用数据的注意事项

数据集的社会影响

[更多信息需补充]

数据集的偏见讨论

[更多信息需补充]

其他已知限制

[更多信息需补充]

附加信息

数据集策展人

数据集最初由 Uma Roy, Noah Constant, Rami Al-Rfou, Aditya Barua, Aaron Phillips 和 Yinfei Yang 在 Google Research 工作期间创建。

许可信息

XQuAD-R 数据集在 CC BY-SA 4.0 许可下发布。

引用信息

bibtex @article{roy2020lareqa, title={LAReQA: Language-agnostic answer retrieval from a multilingual pool}, author={Roy, Uma and Constant, Noah and Al-Rfou, Rami and Barua, Aditya and Phillips, Aaron and Yang, Yinfei}, journal={arXiv preprint arXiv:2004.05484}, year={2020} }

贡献

感谢 @manandey 添加此数据集。

搜集汇总

数据集介绍

构建方式

XQuAD-R数据集是XQuAD的检索版本，由Google Research团队精心构建。其核心构建思路是将原有的跨语言抽取式问答数据集转化为检索式问答任务。具体而言，该数据集保留了XQuAD的11种语言平行结构，即每个问题均以11种语言呈现，并配备11种语言的平行正确答案。数据源基于SQuAD和XQuAD，通过专家标注的方式生成，确保了答案的准确性和语言间的语义一致性。每个语言配置下包含1190个验证样本，以及相应数量的候选句子，构成了一个多语言检索池。

特点

XQuAD-R最显著的特点是其多语言平行检索架构。数据集覆盖阿拉伯语、德语、希腊语、英语、西班牙语、印地语、俄语、泰语、土耳其语、越南语和中文等11种语言，每种语言均提供1190个问答对。每个样本包含唯一的ID、上下文、问题以及答案的文本和起始位置。该数据集专为语言无关的答案检索任务设计，支持从多语言候选池中检索正确答案，是评估跨语言语义理解与检索能力的理想基准。

使用方法

使用XQuAD-R时，可通过Hugging Face Datasets库便捷加载。用户需指定所需语言配置，例如'ar'代表阿拉伯语，'en'代表英语。数据集仅包含验证集，可直接用于评估模型的跨语言检索性能。典型应用场景包括：给定一个问题的某种语言表述，模型需从包含多种语言句子的候选池中找出正确的答案句子。研究者可利用该数据集进行零样本跨语言检索实验，或将其作为多语言语义相似度计算的测试平台。

背景与挑战

背景概述

XQuAD-R数据集由Google Research的研究团队于2020年创建，主要研究人员包括Uma Roy、Noah Constant等，旨在解决跨语言检索式问答中的核心问题。该数据集是XQuAD的检索版本，保留了11种语言并行对齐的特性，每种语言包含1190个问题和对应的候选句子，覆盖阿拉伯语、德语、英语等广泛语种。其研究背景植根于多语言自然语言处理领域，特别是语言无关的答案检索任务，通过提供平行语料支持模型在跨语言场景下的泛化能力。XQuAD-R的发布推动了多语言问答系统的评估标准，成为LAReQA（Language-agnostic answer retrieval from a multilingual pool）研究的重要基准，对低资源语言的检索性能提升具有显著影响力。

当前挑战

XQuAD-R面临的核心挑战包括：首先，跨语言答案检索的领域难题在于语言间的语义鸿沟，模型需在11种语言中准确匹配问题与答案，而不同语言的句法结构和词汇差异导致检索精度受限，尤其在泰语、土耳其语等资源稀缺语言上表现不稳定。其次，数据集构建过程中，从XQuAD的抽取式问答转换为检索式格式需保持平行答案的一致性，人工标注的11种语言答案需严格对齐，但语言间翻译偏差和上下文歧义增加了标注难度，部分语言（如泰语）的候选句子数量显著少于其他语言（852对1222），造成数据不平衡，进而影响模型在低资源语言上的鲁棒性。

常用场景

经典使用场景

XQuAD-R数据集的核心应用场景在于跨语言答案检索任务，其设计初衷是评估模型在11种语言构成的平行语料库中，根据任意语言提出的问题检索出正确答案的能力。该数据集通过将XQuAD中的抽取式问答样本转化为检索式格式，每个问题均配备11种语言的平行正确答案，从而构建了一个多语言答案池。研究者常利用该数据集训练和评测多语言编码器或跨语言表示模型，检验其在语言无关的语义匹配与跨语言信息检索上的表现，尤其关注模型能否突破语言壁垒，从混杂的多语言候选集中精准定位答案。这一场景不仅考验模型对多语言语义对齐的理解，还挑战其在噪声环境下的鲁棒性，是衡量跨语言自然语言理解系统综合能力的标杆。

衍生相关工作

XQuAD-R衍生了一系列推动跨语言表示学习发展的经典工作。其直接关联的LAReQA论文提出了语言无关的答案检索框架，并基于该数据集系统比较了多语言BERT、XLM等预训练模型在检索任务上的表现，揭示了跨语言表示对齐的局限性。后续研究如mBERT的跨语言迁移分析、XLM-R的改进训练策略，以及基于对比学习的多语言句子嵌入方法（如LaBSE），均将XQuAD-R作为关键评估基准。此外，该数据集还启发了对检索式问答中答案多样性问题的探索，例如通过多语言数据增强提升低资源语言性能的工作，以及利用XQuAD-R验证跨语言知识蒸馏有效性的研究。这些工作共同构建了一个以XQuAD-R为核心的学术生态，持续推动着多语言自然语言处理领域的边界拓展。

数据集最近研究