Every one of the world's 7,674 spoken languages in Glottolog, on one map: how they are related, how many are in danger, how many people speak them, and where the great language families may have begun.
Nobody knows exactly how many languages there are, because nobody agrees exactly where one language ends and the next begins. But the best open attempt to list them all, Glottolog, counts 7,674 spoken first languages, living and dead, and places nearly all of them on a map. It also sorts them into families, the way a biologist sorts species: English sits inside West Germanic, inside Germanic, inside Indo-European, and Indo-European is just one of 238 families that cannot, on present evidence, be shown to be related to each other.
This page puts that whole tree on a map. Colour the world by family, then click a family to see its branches light up: Indo-European, then Germanic, then West Germanic, then English. Switch to endangerment to see how many languages are slipping away, to speakers to see how few languages most people speak, or to diversity to find the places where languages are packed most densely. A separate layer, clearly labelled, shows where specialists think the great families began.
Each dot is a language at the point Glottolog gives for it, usually the centre of where it is (or was) spoken; a language spoken in many countries still has one dot, at its historical heartland. 168 languages have no location. Click a dot for details, or browse the family tree in the panel.
Glottolog groups languages into families only where the evidence of shared descent is solid, which is stricter than many popular charts. On that standard the world's 7,674 spoken languages fall into 238 families and 183 isolates: languages such as Basque in Europe, Burushaski in the Karakoram, Nihali in central India and Kusunda in Nepal, with no relatives anyone has been able to prove. Some isolates are only known from writing, like Sumerian. Together that is 421 separate lineages, of which 314 still have at least one living language.
The families are very unequal. Five of them, Atlantic-Congo, Austronesian, Indo-European, Sino-Tibetan and Afro-Asiatic, hold 54.3% of all languages between them. At the other end, 135 of the 238 families have five languages or fewer.
The largest language families
Number of spoken languages in each family, living and extinct, Glottolog 5.2.1. Darker part: living languages. Tap a bar to show the family on the map.
Glottolog's trees are deep. English sits 11 levels down: Indo-European, Classical Indo-European, Germanic, Northwest Germanic, West Germanic, North Sea Germanic, Anglo-Frisian, Anglic and then several historical stages of English itself. The deepest language in the tree, Barama, a Bantu language of Gabon, is 26 levels down. Below the languages are 13,706 named dialects.
Languages are not spread evenly. Counting the living languages Glottolog lists in each country, Papua New Guinea comes first with 850, then Indonesia (702), Nigeria (522), India (450) and China (395). Papua New Guinea is also the deepest in history: its languages belong to 77 different lineages, against 55 for Indonesia and 31 for Brazil. Relative to land area, Vanuatu is in a class of its own, with 102.2 languages per 10,000 square kilometres.
Countries with the most living languages
Living spoken languages that Glottolog lists in each country (a language spoken in several countries counts in each). Right: number of separate lineages (families and isolates) they belong to.
To compare places fairly I also divided the world into squares of 200 by 200 kilometres on an equal-area projection and counted the languages whose point falls in each. The densest square, in Madang, Papua New Guinea, holds 148 living languages; the square with the most separate lineages, in Sandaun, Papua New Guinea near the border with Indonesia, has 23. Just 45 squares, the densest 1% of the world's land, hold 27.6% of all living languages. The densest squares lie in a tropical belt: New Guinea and Melanesia, the Nigeria and Cameroon borderland, southern Mexico, and the eastern Himalaya where India, Myanmar and China meet.
Glottolog's Agglomerated Endangerment Status combines the assessments of Ethnologue, UNESCO and the Endangered Languages Catalogue into one six-step scale. Only 35.0% of spoken languages are rated not endangered. 20.1% are threatened, 22.8% are shifting (children are no longer learning them as a matter of course), 5.4% are moribund and 3.8% nearly extinct. Together that makes 3,998 endangered languages, 52.1% of the total, or 59.7% of those still spoken. Another 981 (12.8%) are already extinct.
Endangerment by world region
Share of spoken languages in each of Glottolog's macroareas by Agglomerated Endangerment Status. Hover for counts.
Loss has been heaviest where colonial settlement was heaviest. In Glottolog's Australia macroarea 63.9% of languages are extinct, in South America 35.1% and in North America 28.2%, against 4.2% in Africa. Whole lineages are at stake: of the 421, 324 (77.0%) have no language left that is rated not endangered, and 98 of the 103 living isolates are endangered.
Only 35.0% of the world's spoken languages are rated safe. 52.1% are endangered and 12.8% already extinct.
Glottolog does not count speakers, so for numbers I turned to Wikidata, which holds a count for 1,627 of the 7,674 spoken languages (21.2%). The counts are a patchwork: they come from different years, from 174 counts made before 2000 to 257 made since 2015, with 495 undated, and some are of first-language speakers only while others include people who speak the language as a second language. Where a language has several counts I took the most recent (the smallest, if several share that date, as that is usually the first-language figure), or the largest if none is dated. Treat them as orders of magnitude.
Even so, the pattern is unmistakable. The counts add up to 6.9 billion. The 10 largest languages, from Mandarin Chinese (918 million) to Japanese, account for 50.9% of that, the 20 largest for 62.6% and the 100 largest for 89.8%. Half the languages with a count have fewer than 20,000 speakers, and those without a count are mostly smaller still. By family, Indo-European languages account for 45.4% of counted speakers and Sino-Tibetan for 19.6%.
The 20 largest languages by Wikidata speaker count
Millions of speakers, coloured by family; year of the count in brackets. Counts mix first-language and total figures (Standard Arabic, for instance, is mostly learned as a second language).
Every family had a first speech community somewhere. Where, and when, is one of the great detective stories of the human past, pieced together from reconstructed vocabulary, the shape of family trees, archaeology and, lately, ancient DNA. The answers are hypotheses, and some are fiercely contested. The table below sets out the main proposals for 11 large families, with the places and dates their authors give and a reference for each; the map shows them as stars when you switch on Proposed homelands. For Indo-European, Uralic, Afro-Asiatic and Uto-Aztecan the literature is split, and each main hypothesis is shown.
Alongside the literature I computed two neutral reference points for every family from Glottolog alone. The present centre is the centre of gravity of where the family's living languages are spoken today. The diversity hotspot is the place where the family's deepest branches crowd most closely together: the language with the most distinct top-level branches within 500 kilometres. That follows an old rule of thumb in linguistics, that a family has usually been longest where its deepest divisions are found side by side.
How far is each proposed homeland from the family's diversity hotspot and present centre?
Great-circle distance in kilometres from each proposed homeland to the hotspot of the family's deepest branches (dot) and to the centre of its living languages today (ring).
For 5 of the 11 families the hotspot falls within 500 kilometres of a proposed homeland: Austronesian (340 km from Taiwan, against 3,368 km for its present centre in New Guinea), Pama-Nyungan (178 km), Quechuan (288 km), Bantu (497 km) and the northern Uto-Aztecan proposal. For Indo-European the hotspot lands in the Balkans, where Greek, Albanian, Slavic, Romance and Romani meet, far from both the steppe and Anatolia: three thousand years of later movement have reshuffled Europe. For Sino-Tibetan it lands in north-east India, nearer the minority view of a south-western homeland than the Yellow River. Hotspots are a heuristic, not proof, and where they disagree with the literature, the literature usually has more to go on.
The homelands are hypotheses from the cited studies, not results of this one; the points mark the approximate centre of a proposed region, and the dates are the authors' ranges. Niger-Congo is not accepted as a proven family by Glottolog, so its star stands for a proposal, not for a family on the map. The computed centres depend on where Glottolog puts each language and on which languages survived to be recorded.
Languages and families. Glottolog 5.2.1 (Hammarström, Forkel, Haspelmath and Bank), in its CLDF release, lists 8,618 language-level entries. I dropped 380 'bookkeeping' entries, which are names kept for reference rather than languages, and mapped the remaining 8,238. Figures on families, endangerment, diversity and speakers use the 7,674 spoken first languages; the other 564 (227 sign languages, 87 pidgins, 128 unclassifiable, 68 unattested and a few artificial and mixed languages and speech registers) are on the map in grey. A family is a top-level genealogical family in Glottolog; an isolate is a language with no proven relatives. 'Living' means any status other than extinct.
Endangerment is Glottolog's Agglomerated Endangerment Status (not endangered, threatened, shifting, moribund, nearly extinct, extinct); I count threatened to nearly extinct as endangered. Speakers are Wikidata's 'number of speakers' for items with a Glottolog code, chosen as described above; they cover a minority of languages and mix years and definitions. Diversity uses Glottolog's country lists for each language and an equal-area grid of 200 km squares (Equal Earth projection); a language counts in the square that holds its point. Country areas are from Natural Earth. Origins: homelands are typed in from the studies listed below; the present centre is the spherical mean of a family's living languages; the diversity hotspot takes the family's top-level branches (splitting further where one branch holds 80% or more of the languages and there are no more than two), and picks the language with the most distinct branches within 500 km.
What this is not. A dot is one point per language, so the map shows where languages come from more than where their speakers live now: English, Spanish and Portuguese have one dot each, in Europe. The language and dialect line is a judgement, and other catalogues count differently. Endangerment ratings are for whole languages and can be out of date. Homelands are contested, and the map shows proposals, not facts.