Metric-AI/ms_marco_hye

Armenian MS MARCO Dataset

This dataset is part of ArmBench-TextEmbed (Armenian Embedding Model Benchmark). We translated a 10K subset of the MS MARCO (Bajaj et al., 2018) validation set (v2.1) using Gemini 2.0 Flash. This dataset is used for Armenian retrieval evaluation tasks.

Citation

If you use this dataset in your research, please cite our paper:

@inproceedings{navasardyan2026less,
  title={Less is More: Adapting Text Embeddings for Low-Resource Languages with Small Scale Noisy Synthetic Data},
  author={Navasardyan, Zaruhi and Bughdaryan, Spartak and Minasyan, Bagratuni and Davtyan, Hrant},
  booktitle={Proceedings of the Workshop on Low-Resource Language Models (LoResLM) at EACL 2026},
  year={2026},
  url={https://arxiv.org/abs/2603.22290}
}

License

This dataset is a translation/derivative of MS MARCO and is licensed under CC BY-NC 4.0. Use is restricted to non-commercial research purposes in accordance with the Microsoft Research License Terms