XQuAD-R: Cross-lingual Answer Sentence Retrieval Benchmark, 11-Way Parallel Questions and Answers (Google Research LAReQA)
- Provider
- Google Research (Roy et al.)
- License
- CC-BY-SA-4.0 (source)
- Size
- 13 files
- Format
- Parquet (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/google-research-datasets/xquad_r)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Natural Language Processing
- Source
- https://arxiv.org/abs/2004.05484
About this dataset
XQuAD-R is the retrieval version of XQuAD, released with LAReQA (Language-Agnostic Answer Retrieval from a multilingual pool). XQuAD's 240 SQuAD paragraphs and 1,190 questions are professionally translated into 11 languages (Arabic, German, Greek, English, Spanish, Hindi, Russian, Thai, Turkish, Vietnamese, Chinese); XQuAD-R splits the paragraphs into candidate answer sentences, so every question has 11 parallel correct answers, one per language, inside a pooled multilingual candidate set. The task is to retrieve the answer sentence for a question from the whole pool regardless of language, which tests whether multilingual sentence embeddings and dual-encoder retrievers are language-agnostic rather than biased toward same-language matches. The repository has one folder per language code (ar, de, el, en and so on), each with a validation Parquet file holding contexts, questions and answers with sentence offsets. Load with the Hugging Face datasets library (load_dataset('google-research-datasets/xquad_r', 'en')). Distributed under CC BY-SA 4.0; cite Roy et al., LAReQA: Language-Agnostic Answer Retrieval from a Multilingual Pool, EMNLP 2020.
How to load this dataset in Python
from datasets import load_dataset
ds = load_dataset("google-research-datasets/xquad_r")
print(ds)Data structure
File types in sampled files: Parquet
Files
Example file: ar/validation-00000-of-00001.parquet
Analyze this dataset without downloading it
On Scigantic, XQuAD-R: Cross-lingual Answer Sentence Retrieval Benchmark, 11-Way Parallel Questions and Answers (Google Research LAReQA) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Natural Language Processing datasets
- Helpful Sentences from Reviews
- Google Books Ngram Viewer Datasets Version 3 (20200217): Yearly 1-gram to 5-gram Counts across Eight Languages and English Subcorpora (Google Books)
- CourtListener Case Law Embeddings: Chunked US Court Opinion Text with 768-Dimensional ModernBERT Vectors for Semantic Legal Search (Free Law Project)
- MultiCoNER Datasets
- Belebele
- XQuAD (Cross-lingual Question Answering Dataset): 1,190 SQuAD Questions Professionally Translated into 11 Languages (Google DeepMind)
- Md Gender Bias
- S2AND: Author Name Disambiguation Benchmark, Eight Bibliographic Datasets plus Semantic Scholar Production Models (Allen Institute for AI)