AI Language Identifier (fastText, 176 Languages)
Detect which of 176 languages a text is written in, with probabilities, using the fastText lid.176 model in your browser.
Use the tool
Why this tool exists
Language identification routes text to the right translator, spell checker, hreflang tag or support queue, and flags pages in the wrong language. fastText lid.176 is a compact, well-known classifier trained on Wikipedia, Tatoeba and SETimes; this page reads its 917 KB file with a hand-written JavaScript reader and predictor and reproduces the reference implementation exactly on our test set.
Short strings, mixed-language text and transliterated text (for example Hindi typed in Latin letters) are the weak spots, and related languages such as Serbian and Croatian blur together; the probability column tells you when to doubt it. The model file is licensed CC BY-SA 3.0 and loads from this site once; your text stays in the tab.
How to use it
- Paste text; leave "each line separately" ticked to classify every line.
- Press Identify.
- Read the language, its probability and the next two candidates.
Frequently asked questions
How is this different from the Language Detector tool?
The other tool is rule-based over a few languages. This one runs the trained fastText model and covers 176 languages with probabilities.
Why is a short phrase wrong?
There is too little evidence in a few words. Lower probabilities flag these cases; add more text.
What licence applies?
The lid.176 model is CC BY-SA 3.0 (Facebook AI Research). Attribution is shown on this page and in data/SOURCES-gap5.md.