Back to Gap Map Model components / Language-specific datasets

SMOL

Google
open / Overall score: 1.3

SMOL is a Google suite of professionally translated training data for machine translation into more than 200 low-resource languages. It combines SmolSent, sentences picked for broad vocabulary coverage; SmolDoc, whole documents chosen for topic coverage, with factuality ratings; and the GATITOS word and phrase lexicon. Volunteers have added translations for further languages since the first release.

Openness

5 high confidence
5.0
license
cc-by-4.0
access
public
dataset_card
present

The translations carry an attribution-only license and download from Hugging Face without a gate.

Adoption

2 high confidence
2.0

Hugging Face downloads over the trailing month for the one SMOL repository, which also carries GATITOS.

Capability

1 medium confidence
1.0

SMOL gives a long list of low-resource languages professionally translated sentences and documents, documented in two papers, though no model or benchmark beyond its own experiments is known to be built on it. Its papers and card count tokens, sentences and documents but never sentence pairs, and what they do count is minute beside the billions of pairs in the largest parallel collections and smaller than the IIT Bombay English-Hindi corpus for a single language pair.

  • https://arxiv.org/abs/2502.12301 recorded 2026-11-12

    "SMOL has been translated into 124 (and growing) under-resourced languages (125 language pairs), including many for which there exist no previous public resources, for a total of 6.1M translated tokens."

  • https://arxiv.org/pdf/2502.12301 recorded 2026-11-12

    Table 2, whole dataset: SmolSent 88 languages and 76k examples, SmolDoc 106 languages and 27k examples; Table H.1 notes that a SmolSent example is a sentence pair and a SmolDoc example a document or paragraph.

  • https://huggingface.co/datasets/google/smol/raw/main/README.md recorded 2026-11-12

    "a collection professional translations into 221 Low-Resource Languages"; "sentence-level translations into 114 language pairs"; "for a 5x increase in dataset size", with no new total.

Verified 2026-09-24