A language family is a group of languages that descend from a common ancestor, called a proto-language. Members of a family are related by inheritance, shown by regular sound and word correspondences, not by looking alike or being spoken nearby.
Key takeaways
- Languages are placed in a family when regular patterns in their sounds and words point to a shared ancestor.
- The map has 211 named families plus 103 isolates, which have no known relatives.
- Indo-European has 485 languages on the map, in 8 branches.
- Family trees simplify. Borrowing between neighbours can make unrelated languages look alike.
How families are worked out
The homepage explains what a family is in a paragraph (see language families on the map). This page goes one step further: how linguists decide that languages belong together.
The method is called the comparative method. Researchers collect likely cognates, words in different languages with similar sounds and meanings, and test whether the similarities are better explained by chance, by borrowing, or by common descent [1]. Chance and borrowing are filtered out. If what remains fits a shared origin, the languages are treated as related and the ancestor is called a proto-language [1][2]. Linguists then use the same comparisons to reconstruct features of that ancestor.
Large families split into smaller units. Germanic, for example, is a subfamily of Indo-European, and its members share innovations that arose after the larger family's ancestor existed [1]. Those subunits are branches.
The tree is a simplification. Critics point out that tree shapes depend on the criteria used and that experts disagree about which languages belong in some families [1]. An alternative, the wave model, groups varieties by features that spread between neighbours and can overlap, which suits languages that stayed in contact [1]. Borrowing does not prove kinship, because unrelated languages can influence each other [1]. That is why language contact and the sprachbund are kept separate from family membership.
An example from our data
The map follows Glottolog's classification [3]. It has 211 named families and 103 isolates. Indo-European has 485 languages on the map in 8 branches, and its largest languages on the map are Spanish, English and Hindi. Sino-Tibetan has 496 languages in 17 branches, and Uralic has 41 languages in 9 branches.
The largest Uralic languages on the map are Hungarian, Finnish and Estonian, and its top country is Russia. Family membership is about descent, so it does not follow national borders. Use the family filter on the map to see how far apart relatives can be.
The map has no data on grammar or vocabulary, so it cannot show why a language is placed in a family. Each family page lists its languages, and the sources are on the data sources page.
The thing people get wrong
The common error is to treat resemblance as proof of relationship. Languages can share many words because one borrowed from the other, and unrelated languages can look alike by chance [1]. The reverse also holds: distant relatives can sound very different after thousands of years of separate change.
A related mistake is to think that a family is a ranking. It is a statement about shared ancestry only. It says nothing about which languages are older, harder or more developed. Every living language has an equally long history behind it.
Frequently asked questions
How do linguists know two languages are related?
They compare likely cognates, words with similar sounds and meanings, and look for regular correspondences. They rule out chance and borrowing before inferring a common origin.
Are all languages in a family?
No. Isolates have no demonstrated relatives. The map has 103 of them.
Does a family tell you where its languages are spoken?
Not necessarily. Related languages can be far apart, so the map colours languages by family and lets you show one family at a time.
Sources
- Language family, Wikipedia.
- Proto-language, Wikipedia.
- Glottolog, Max Planck Institute for Evolutionary Anthropology.
- Every figure from our own map is worked out from the datasets on the data sources page (checked against Glottolog and Glottography, retrieved 13 October 2026).