Back to Gap Map Model components / Benchmark / eval datasets

IndoNLU

IndoNLP
gated / Overall score: 1.0

IndoNLU is a benchmark for Indonesian natural language understanding with twelve tasks, from emotion and sentiment classification to part-of-speech tagging, named-entity recognition, keyphrase extraction, textual entailment and extractive QA. The task data comes from social media, reviews, news and Wikipedia, some of it labeled by Indonesian linguists. It shipped with the Indo4B pretraining corpus and the IndoBERT models, from a collaboration including Institut Teknologi Bandung, Universitas Indonesia, Gojek and Prosa.AI.

Openness

2 medium confidence
2.0
license
mit (HF card: 'The licensing status of the IndoNLU benchmark datasets is under MIT License'; repo LICENSE file is Apache-2.0)
access
public (ungated)
dataset_card
present (per-task descriptions; creation and annotation sections are marked 'Needs More Information')
answers
held-out (GitHub test-set labels are masked; predictions are scored on CodaLab)

The task data downloads freely under a permissive license, but the official test labels are masked and scored through a CodaLab competition. The Hugging Face card leaves its collection and annotation sections unfilled, so provenance comes from the paper.

Adoption

1 high confidence
1.0

Adoption is Hugging Face downloads of the repository. Users of the GitHub copies and of the CodaLab test sets are not counted.

Capability

not assessed

A benchmark is not scored for capability, so that score is left blank. IndoNLU tests Indonesian, a language with a large written web, so it sits with general evaluation sets rather than with sets built for an underrepresented language.

  • https://arxiv.org/abs/2009.05387 recorded 2026-11-11

    "we introduce the first-ever vast resource for the training, evaluating, and benchmarking on Indonesian natural language understanding (IndoNLU) tasks. IndoNLU includes twelve tasks"

Verified 2026-09-27