SMOL and GATITOS: Professional Translations into 221 Low-Resource Languages (Google Research)

Provider
Google Research (Caswell et al.)
License
CC-BY-4.0 (source)
Size
870 files
Format
JSON Lines (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/google/smol)
Access
Free and public. No account with the data provider is needed.
Category
Natural Language Processing
Source
https://arxiv.org/abs/2502.12301

About this dataset

SMOL (Set for Maximal Overall Leverage) is a set of professional human translations into 221 low-resource languages, released by Google Research to support machine translation and language technology for under-represented languages. It has four parts. SmolDoc is document-level translation for about 130 language pairs. SmolSent is sentence-level translation for about 114 pairs. GATITOS is a token- and phrase-level lexicon for about 181 pairs, which in 2026 gained 1,000 Cantonese food dish names. The fourth part is factuality ratings with rationales for 661 SmolDoc documents. Volunteer translations were added in 2026 for Abaza, Brahui, Chechen, Juhuri, Meadow Mari and Tuvan, and many languages were raised to a larger data tier. There are 867 JSONL files under smoldoc/, smolsent/ and gatitos/, one per language pair, named by ISO code (for example en_tig, ru_kjh). Linguists and MT researchers use it as small, high-quality aligned data for fine-tuning, for lexicon-based augmentation, and for evaluating translation into languages with almost no web text. Load it with Hugging Face datasets by config name. The data is released under CC BY 4.0. Cite Caswell et al., SMOL (arXiv 2502.12301) and Jones et al., GATITOS (arXiv 2303.15265).

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("google/smol")
print(ds)

Data structure

File types in sampled files: JSON Lines

Files

Example file: gatitos/aa_en.jsonl

Analyze this dataset without downloading it

On Scigantic, SMOL and GATITOS: Professional Translations into 221 Low-Resource Languages (Google Research) 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