How it works
Coverage is steeply front-loaded: a few hundred languages account for most speakers, while thousands of languages share the remaining sliver.
How many people speak what
The distribution of speakers is extraordinarily lopsided. The chart below plots the largest languages in the world by total speakers — first language plus second language — and colours each bar by how well the translation apps we benchmark actually cover it.
Two patterns show up immediately. The very largest languages are all comprehensively covered, so the apps look excellent on a population-weighted view. But several languages with tens of millions of speakers — Nigerian Pidgin, Wu Chinese, Western Punjabi — sit in the amber and red bands, supported thinly or not at all.
Coverage on the map
Speaker counts alone hide where the gaps are. The interactive globe below shades every country by the estimated share of its inhabitants who speak a language with production-grade app support. Rotate it and the pattern is geographic as much as demographic: Europe, East Asia and the Americas are solid green, while West Africa, the Sahel, Papua and parts of the Andes drop into amber and red.
How many languages are there?
Ethnologue counts roughly 7,100 living languages. That number is not stable: it shifts as linguists split or merge varieties, and around 40% of those languages are endangered, with one estimated to fall out of everyday use every few weeks.
Only about 4,000 of them have a developed writing system. That matters enormously for translation technology, because text corpora are the raw material that machine-translation and speech models are trained on. A language with no written record has almost no path into a conventional translation pipeline.
How many languages do translation apps cover?
Coverage varies by more than an order of magnitude between apps. Google Translate is the outlier at roughly 249 languages and language varieties, boosted by its 2024 expansion using large language models to bootstrap low-resource languages. Microsoft Translator and LLM-based tools like PolyPort.chat sit above 100. DeepL covers around 35, mostly European. Apple Translate is near 20, Reverso near 18, and Papago around 14 with an East Asian focus.
Raw counts overstate the picture. A language can appear in a dropdown while producing output that is barely usable in practice. Quality tracks training data, so the top 10 to 20 high-resource languages — English, Spanish, French, German, Chinese, Japanese, Portuguese, Russian, Arabic, Hindi and a handful more — get results far better than the long tail.
The gap to 100% coverage
The remaining 5–8% of people is not one gap but thousands of small ones. It includes speakers of Papuan, Australian Aboriginal, Amazonian, Caucasian and West African languages, many indigenous American languages, sign languages, and hundreds of regional varieties of otherwise supported languages — Arabic dialects, Chinese topolects, and regional Indian languages among them.
The last few percent is disproportionately expensive. Every additional language needs parallel text, speech recordings, evaluation data and native reviewers, and by definition the remaining ones have the least material available. This is why the curve flattens: going from 100 to 250 languages added maybe 8 percentage points of reach; going from 250 to 1,000 might add 3 more.
Two things are changing the economics. Massively multilingual models transfer knowledge from high-resource to related low-resource languages, so a new language no longer needs a corpus built from scratch. And speech-first models can skip text entirely, which is the only realistic route for the roughly 3,000 languages with no standard writing system.
What this means when picking an app
Coverage counts are only relevant for your specific pair. If you work in French, German and Spanish, DeepL's 35 languages are more useful than Google's 249, because quality on those pairs is better. If you travel widely or need Swahili, Amharic or Quechua, breadth wins and Google is the only serious option.
Check three things rather than the headline number: whether both of your languages are supported, whether the pair works in the modes you need — voice, camera, offline — and whether anyone reports usable quality on that pair. Many apps support a language for text but not for speech recognition or camera translation.
Frequently asked questions
How many languages are there in the world?
About 7,100 living languages according to Ethnologue, though roughly 40% are endangered and only about 4,000 have a developed writing system.
Which translation app supports the most languages?
Google Translate, with roughly 249 languages and language varieties after its 2024 LLM-assisted expansion. Microsoft Translator and LLM-based tools cover 100 or more.
What percentage of the world can use translation apps?
Apps covering around 250 languages nominally reach roughly 92–95% of people as first or second language speakers. A 100-language app reaches about 85%. That is a ceiling, not actual usage — it assumes device, connectivity and literacy.
Why don't apps support all 7,100 languages?
Machine translation needs large amounts of parallel text and speech data. Thousands of languages have little written material or none at all, and the cost per language rises sharply as the available data shrinks.
Does a bigger language list mean better translation?
No. Quality tracks training data per pair. A specialist app with 14 or 35 languages often produces better output on those pairs than a broad app that supports 249.
