Add dataset card for XBCP

README.md

@@ -0,0 +1,63 @@
+---
+task_categories:
+- text-retrieval
+- question-answering
+language:
+- en
+- zh
+- fr
+- de
+- ja
+- ko
+- pt
+- es
+- sw
+- wo
+- yo
+- zu
+---
+
+# XBCP: Cross-lingual BrowseComp-Plus
+
+[**Paper**](https://huggingface.co/papers/2606.15345) | [**GitHub**](https://github.com/paddler2022/XBCP)
+
+XBCP (Cross-lingual BrowseComp-Plus) is a controlled benchmark designed to evaluate deep research agents and retrievers. While deep research agents are typically evaluated on their ability to reason over retrieved sources in the same language as the query, XBCP tests their performance when evidence appears in different languages.
+
+The dataset preserves the English question-and-answer space of [BrowseComp-Plus](https://github.com/texttron/BrowseComp-Plus) but translates supporting evidence documents into 12 languages spanning high-resource and low-resource regimes.
+
+## Dataset Summary
+
+XBCP instantiates two complementary evaluation settings:
+- **Cross-lingual**: Each query is paired with evidence in a single assigned non-English language.
+- **Multilingual**: The full evidence corpus is distributed equally and randomly across 12 languages.
+
+### Supported Languages
+The benchmark includes evidence translated into:
+- **High-resource**: Chinese (`zh`), French (`fr`), German (`de`), Japanese (`ja`), Korean (`ko`), Portuguese (`pt`), Spanish (`es`).
+- **Low-resource**: Swahili (`sw`), Wolof (`wo`), Yoruba (`yo`), Zulu (`zu`).
+- **Reference**: English (`en`).
+
+## Sample Usage
+
+You can download the translated corpora and pre-built retrieval indexes (supporting various retrievers like BM25, Qwen3-Embedding, and Multilingual-E5) using the Hugging Face CLI:
+
+```bash
+# Download translated corpora
+huggingface-cli download UTokyo-Yokoya-Lab/XBCP data --repo-type dataset --local-dir .
+
+# Download pre-built indexes
+huggingface-cli download UTokyo-Yokoya-Lab/XBCP indexes --repo-type dataset --local-dir .
+```
+
+For full reproduction scripts and evaluation pipelines, please refer to the [official GitHub repository](https://github.com/paddler2022/XBCP).
+
+## Citation
+
+```bibtex
+@article{lu2026beyond,
+  title={Beyond Monolingual Deep Research: Evaluating Agents and Retrievers with Cross-Lingual BrowseComp-Plus},
+  author={Lu, Yuheng and Zeng, Qingcheng and Qi, Heli and Yu, Puxuan and Zhao, Fuheng and Yang, Rui and Yanaka, Hitomi and Yokoya, Naoto and Xuan, Weihao},
+  journal={arXiv preprint arXiv:2606.15345},
+  year={2026}
+}
+```