How does the language detector work?

It counts which Unicode script the letters belong to. If only one language uses that script (Greek, Korean, Thai, Georgian and so on) the answer is direct. Otherwise it splits the text into overlapping three-letter sequences (trigrams) and measures how far their ranking is from each language's trigram profile, the method used by the open-source franc library. Common function words such as "the", "und" or "que" add extra evidence for short text.

With a full sentence (about 40+ characters) the top guess is right for nearly all major languages. Very short inputs such as "Hello" or a single name are unreliable in any detector, so the confidence is capped for text under 20 letters. Related languages — Indonesian vs Malay, Danish vs Norwegian, Czech vs Slovak, Serbian vs Croatian vs Bosnian, Persian vs Dari — often need a paragraph.

What does the confidence percentage mean?

It is the relative probability of the top language among the candidates that use the same script, computed from the trigram distances. 95% means the text matches that language far better than any other; 55% vs 40% means two languages fit almost equally well, so check the runner-up.

Can it detect several languages in one text?

Yes. Turn on "Detect each sentence" to run the detector on every sentence or line separately. The table lists each sentence with its language, and the summary shows the share of characters per language.

How does it tell Chinese and Japanese apart?

Both use Han characters (kanji/hanzi). Japanese text almost always contains hiragana or katakana, so any meaningful share of kana means Japanese. Text written only in Han characters is reported as Chinese, with Japanese as the runner-up.

Does this tool send my text to a server?

No. The language models (about 60 KB) are downloaded once and all detection runs in your browser. Your text never leaves your device.

Which languages can be detected?

96 languages, including English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Ukrainian, Arabic, Persian, Hindi, Bengali, Chinese, Japanese, Korean, Vietnamese, Thai, Turkish, Indonesian, Malay, Tagalog, Swahili, Greek, Hebrew and all major European languages. The full list with ISO codes is in the table below the tool.

What are ISO 639-1 and ISO 639-3 codes?

ISO 639-1 is the two-letter code used in HTML lang attributes and BCP 47 tags (en, es, zh). ISO 639-3 is the three-letter code that covers every individual language (eng, spa, cmn). The result shows both.

Language Detector

Paste any text to see its language, a confidence score, the top-3 candidates, ISO 639-1 and 639-3 codes and the writing system. Mixed text can be split sentence by sentence. 96 languages, processed locally in your browser.

Input Text

Detect each sentence

Detected Language

Confidence

Script

ISO 639-1 / 639-3

Characters

Words

Top Candidates

Sentence by sentence

#	Sentence	Language	Conf.

Sample Texts

How the detection works

Script. Each letter is matched to a Unicode script. A script used by one language (Greek, Hangul, Thai, Georgian, Armenian, Tamil…) decides the language directly; any kana in Han text means Japanese.
Trigrams. The text is lower-cased, punctuation and digits are dropped, and every three-character sequence is counted: "the cat" gives " th", "the", "he ", "e c", " ca", "cat", "at ".
Distance. Each trigram is looked up in every language profile of that script (the 300 most frequent trigrams, ranked). Known trigrams add their rank difference; unknown ones add a 300 penalty. Common function words found in the text subtract from a language's distance, mainly helping short phrases.
Probability. The smallest distance wins. Distances are converted to relative probabilities (softmax), which gives the confidence and the top-3 list.

Example: "Wo ist der nächste Bahnhof?" has 26 trigrams. On trigrams alone Afrikaans and Dutch score close to German, but "wo", "ist" and "der" are German function words, so German wins. Longer text rarely needs this help.

Supported languages (96)

ISO 639-3 code as used by the trigram models, ISO 639-1 / BCP 47 tag for HTML lang attributes.

Language	ISO 639-1	ISO 639-3	Script
Afrikaans	af	afr	Latin
Albanian	sq	als	Latin
Amharic	am	amh	Ethiopic
Arabic	ar	arb	Arabic
Armenian	hy	hye	Armenian
Azerbaijani	az	azj	Latin
Belarusian	be	bel	Cyrillic
Bengali	bn	ben	Bengali
Bosnian	bs	bos	Latin
Bulgarian	bg	bul	Cyrillic
Burmese	my	mya	Myanmar
Catalan	ca	cat	Latin
Cebuano	ceb	ceb	Latin
Chichewa (Nyanja)	ny	nya	Latin
Chinese (Mandarin)	zh	cmn	Han (CJK)
Croatian	hr	hrv	Latin
Czech	cs	ces	Latin
Danish	da	dan	Latin
Dari	fa-AF	prs	Arabic
Dutch	nl	nld	Latin
English	en	eng	Latin
Esperanto	eo	epo	Latin
Estonian	et	ekk	Latin
Finnish	fi	fin	Latin
French	fr	fra	Latin
Galician	gl	glg	Latin
Georgian	ka	kat	Georgian
German	de	deu	Latin
Greek	el	ell	Greek
Gujarati	gu	guj	Gujarati
Haitian Creole	ht	hat	Latin
Hausa	ha	hau	Latin
Hebrew	he	heb	Hebrew
Hindi	hi	hin	Devanagari
Hungarian	hu	hun	Latin
Igbo	ig	ibo	Latin
Indonesian	id	ind	Latin
Italian	it	ita	Latin
Japanese	ja	jpn	Kana (Japanese)
Javanese	jv	jav	Latin
Kannada	kn	kan	Kannada
Kazakh	kk	kaz	Cyrillic
Khmer	km	khm	Khmer
Kinyarwanda	rw	kin	Latin
Korean	ko	kor	Hangul
Kyrgyz	ky	kir	Cyrillic
Lao	lo	lao	Lao
Latvian	lv	lvs	Latin
Lingala	ln	lin	Latin
Lithuanian	lt	lit	Latin
Macedonian	mk	mkd	Cyrillic
Malagasy	mg	plt	Latin
Malay	ms	zlm	Latin
Malayalam	ml	mal	Malayalam
Marathi	mr	mar	Devanagari
Mongolian	mn	khk	Cyrillic
Nepali	ne	npi	Devanagari
Norwegian Bokmål	nb	nob	Latin
Norwegian Nynorsk	nn	nno	Latin
Pashto	ps	pbu	Arabic
Persian (Farsi)	fa	pes	Arabic
Polish	pl	pol	Latin
Portuguese	pt	por	Latin
Punjabi	pa	pan	Gurmukhi
Romanian	ro	ron	Latin
Russian	ru	rus	Cyrillic
Serbian	sr	srp	Latin + Cyrillic
Shona	sn	sna	Latin
Sinhala	si	sin	Sinhala
Slovak	sk	slk	Latin
Slovenian	sl	slv	Latin
Somali	so	som	Latin
Spanish	es	spa	Latin
Sundanese	su	sun	Latin
Swahili	sw	swh	Latin
Swedish	sv	swe	Latin
Tagalog (Filipino)	fil	tgl	Latin
Tajik	tg	tgk	Cyrillic
Tamil	ta	tam	Tamil
Tatar	tt	tat	Cyrillic
Telugu	te	tel	Telugu
Thai	th	tha	Thai
Tibetan	bo	bod	Tibetan
Tigrinya	ti	tir	Ethiopic
Turkish	tr	tur	Latin
Turkmen	tk	tuk	Latin
Ukrainian	uk	ukr	Cyrillic
Urdu	ur	urd	Arabic
Uyghur	ug	uig	Arabic
Uzbek	uz	uzn	Latin
Vietnamese	vi	vie	Latin
Wolof	wo	wol	Latin
Xhosa	xh	xho	Latin
Yiddish	yi	ydd	Hebrew
Yoruba	yo	yor	Latin
Zulu	zu	zul	Latin

Trigram profiles: franc 6.2 (MIT licence), trained on the Universal Declaration of Human Rights.

Summary

The detector first finds the writing system (script) from Unicode ranges. Scripts used by one language — Greek, Korean Hangul, Thai, Georgian, Armenian, Tamil and others — identify the language directly; Japanese is recognised by its kana even when kanji dominate. For shared scripts (Latin, Cyrillic, Arabic, Devanagari, Hebrew, Ethiopic) it compares the text's character trigrams with frequency profiles of each language (the open-source franc models) and adds a small boost for common function words, which helps with short phrases. Distances are turned into relative probabilities, so the top-3 percentages add up to about 100%. Closely related pairs such as Indonesian/Malay, Danish/Norwegian, Czech/Slovak, Persian/Dari and Bosnian/Croatian/Serbian need longer text to separate.

How it works

Paste or type text into the input box — detection starts automatically.
Read the detected language, its native name, ISO codes and confidence percentage.
Check the top-3 candidates; close percentages mean the text is short or the languages are related.
Turn on "Detect each sentence" for documents that mix languages.
Use the sample buttons to see how different scripts and languages are recognised.

Use cases

Translators identifying source-language documents before starting work
Content moderators screening user-submitted text for language compliance
Linguists and researchers analyzing multilingual datasets
Developers checking which language tag (ISO 639-1 / BCP 47) to assign to content
Travellers working out what language a sign, menu or message is written in

Frequently Asked Questions

Last updated: 2026-10-10 · Reviewed by Nham Vu