Native SEA-Language Text (25,000 pieces)

Native SEA-Language Text (25,000 pieces)

TEXT

COMMERCIAL

LABELED

An opinion, a story, an explanation, written by a native or fluent speaker in the language they think in, with that attestation and a language tag on every piece, as plain text. The screening is the hard part: text that reads as machine-translated, copied or written by a learner is refused, because much web text in these languages is translated, and models trained on it write the way a translation reads. The task is open only in 11 Southeast Asian countries, so expect Filipino, Bahasa Indonesia, Bahasa Melayu, Javanese and Sundanese to dominate, with Thai, Vietnamese and others alongside; five pieces per author at most.

Volume

25,000+

Pieces

Contributors

8,300+

Regions

Southeast Asia (11 countries)

Updated

Technical specs

Technical specs

What’s in the collection and how it’s delivered.

Data modality

Text

Annotated

Labeled (language tag, native or fluent attestation)

Exclusivity

Non-exclusive rights

Capture tool

Typed in the Kled app

Use cases

Multilingual LLM pretraining, native-quality generation evaluation, machine-translation detection

Format

Plain text

Resolution

Plain text, UTF-8

Total volume

25,000+ pieces (approx. 125 MB)

Regions

Southeast Asia (11 countries)

Purity level

99%+ passed QA review

Ready to test
this dataset?

Ready to test
this dataset?

Get a sample set to evaluate, or talk to our data team about volume, exclusivity, and custom collection.

Get a sample set to evaluate, or talk to our data team about volume, exclusivity, and custom collection.