ifm23/NAMES

The NAMES language dataset is generated using the Faker Python package (https://faker.readthedocs.io/). The dataset contains first and last names from a range of different countries, and the task is to classify a particular name instance into the correct country category. As a result, this can be considered a classification task for the language data domain.

The dataset consists of first and last names from 15 country categories:

• English (en_GB)
• German (de_DE)
• French (fr_FR)
• Italian (it_IT)
• Spanish (es_ES)
• Portuguese (pt_PT)
• Russian (ru_RU)
• Polish (pl_PL)
• Dutch (nl_NL)
• Japanese (ja_JP)
• Norwegian (no_NO)
• Chinese (zh_CN)
• Czech (cs_CZ)
• Greek (el_GR)
• Turkish (tr_TR)

It is important to note that for the Japanese and Chinese names, the romanised form was used. To ensure class balance for each country category, we generated 1000 unique first and last name combinations with a maximum length of 20, totalling 15000 unique names. Each category was split into an 80-20 train (800 samples) and test (200 samples) split to ensure class balance during training and evaluation. The dataset is used for the paper "A Function-Centric Perspective on Flat and Sharp Minima" which appears in the Transactions on Machine Learning Research. If you use this dataset, please cite the paper below:

@article{
mason-williams2026function-centric,
title={A Function-Centric Perspective on Flat and Sharp Minima},
author={Israel Mason-Williams and Gabryel Mason-Williams and Helen Yannakoudakis},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2026},
url={https://openreview.net/forum?id=LV1ffpXXeV},
note={}
}

Find the paper on arXiv here: https://arxiv.org/pdf/2510.12451

Additionally this repository contains a dataloader under: names_dataloader.py which can be used to load the train and test split of the dataset.

---

language: - English pretty_name: "NAMES" tags: - text classification - synthetic data license: mit task_categories: - text classification