MERLIN Dataset and Baselines

This repository contains baselines and experiments on the MERLIN (Multilingual Entity Recognition and Linking) dataset.

Dataset Description

MERLIN is a multilingual dataset for entity linking that includes content in multiple languages with associated images. Each entry in the dataset contains:

{
    "Article Title": "बिहार: केंद्रीय मंत्री अश्विनी चौबे के बेटे अर्जित 'गिरफ्तार'",
    "Entity Name": "अश्विनी चौबे",
    "Wikidata ID": "Q16728021",
    "English Wikipedia Title": "Ashwini Kumar Choubey",
    "Image Name": "hindi_601.jpg"
}

Key features:

• Multilingual article titles and entity names
• Wikidata IDs for entity linking
• Cross-lingual mapping to English Wikipedia titles
• Associated images for multimodal learning

Dataset Statistics 📊

Our multilingual dataset spans 5 languages with comprehensive entity annotation coverage:

• Total Articles: 5,000 (1,000 per language)
• Languages: Hindi, Indonesian, Japanese, Tamil, and Vietnamese
• Average Article Length: 11.10 words
• Entity Coverage:
  • 2,480 unique entities identified
  • 7,287 total entity mentions
  • Average of 1.39 words per entity mention

Baselines

1. GEMEL Baseline

A modified version of GEMEL (Generative Multimodal Entity Linking), adapted for MERLIN dataset.

Usage

GEMEL Instructions are here

2. GENRE Baseline

A modified version of GENRE (Generative ENtity REtrieval), adapted for MERLIN dataset.

Environment Setup

Usage

mGENRE Instructions are here

Visualizing dataset

We have hosted dataset in ZenoML, visit these links:

1. Hindi
2. Indonesian
3. Vietnamese
4. Japanese
5. Tamil

Huggingface

Please find the huggingface repo for MERLIN here

Citation

@misc{ramamoorthy2025merlintestbedmultilingualmultimodal,
      title={MERLIN: A Testbed for Multilingual Multimodal Entity Recognition and Linking}, 
      author={Sathyanarayanan Ramamoorthy and Vishwa Shah and Simran Khanuja and Zaid Sheikh and Shan Jie and Ann Chia and Shearman Chua and Graham Neubig},
      year={2025},
      eprint={2510.14307},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.14307}, 
}

Baseline Papers

If you use these baselines, please also cite the original works:

@article{shi2023generative,
  title={Generative Multimodal Entity Linking},
  author={Shi, Senbao and Xu, Zhenran and Hu, Baotian and Zhang, Min},
  journal={arXiv preprint arXiv:2306.12725},
  year={2023}
}

@inproceedings{decao2021autoregressive,
  author    = {Nicola {De Cao} and Gautier Izacard and Sebastian Riedel and Fabio Petroni},
  title     = {Autoregressive Entity Retrieval},
  booktitle = {9th International Conference on Learning Representations, {ICLR} 2021},
  year      = {2021},
}

License

[To be added: License information]