How does language identification work?

Every language leaves clues in writing. The script narrows it down fast, since Cyrillic, Greek or Arabic each rule out most languages at once. After that come letters like ñ in Spanish or ß in German, and small words that show up everywhere, such as "the", "le" or "und".

Closely related languages are the hard part. Spanish and Portuguese, or Danish and Norwegian, share so many words that a short sentence can fool a simple detector. An AI model can also weigh grammar and vocabulary, and split a text that mixes several languages.

Tool description

Paste up to 5,000 characters into Text and click Identify languages. The table lists every language found, from the largest share to the smallest, with these columns:

  • Language, the name of the language
  • Native name, its name in the language itself, such as Deutsch or 日本語
  • ISO code, the ISO 639-1 code, or ISO 639-3 for languages without a two letter code
  • Share, a rough percentage of the text in that language
  • Sample, a short excerpt copied from your text

Reasoning explains which words, spelling, grammar or script gave each language away, and notes where the AI isn't sure. Names, numbers, URLs and single borrowed words are ignored unless they're all the text has.

Example

For "Bonjour à tous! I just moved to Montreal y todavía estoy aprendiendo francés.", you'd expect something like English and Spanish at about 40% each and French at about 20%, with samples such as "Bonjour à tous!".

Features

  • Finds every language in a mixed text, with its share
  • ISO codes and native names for each language
  • A short explanation of how each language was recognized

Limitations

A word or two is often not enough to tell related languages apart, and the shares are estimates. AI answers can be wrong, so check results that matter.