What is a language family? How languages are related

A language family is a group of languages that all descend from a single ancestor language, much as cousins descend from shared grandparents. English, Hindi, Russian, Spanish and Persian, for example, are all members of the Indo-European family, descended from a language spoken roughly 6,000 years ago. Linguists prove a relationship not by spotting similar words but by finding regular, repeated sound correspondences that chance and borrowing can't explain.

In short

• Languages change a little in every generation; when groups of speakers separate, their speech drifts into different languages that still share an origin.
• The proof of kinship is regularity: the same sound in one language matches the same sound in another across hundreds of words.
• Being related doesn't mean being understandable to each other, and the language a people speaks says nothing certain about their ancestry.

How do languages become a family?

Every living language changes. Sounds shift, words change meaning, grammar simplifies here and grows there. When speakers of one language spread out and lose regular contact, the changes in each group pile up separately, and after some centuries they no longer understand each other. That is how Latin became Spanish, French, Italian, Portuguese and Romanian, as the note on whether Latin is a dead language describes. Latin itself was one branch of the larger Indo-European family.

The Romance languages are a family within a family whose ancestor we can read, because Latin was written down. For most families, the ancestor was never written. Its words have to be worked out from its descendants, which is the job of the comparative method.

How do linguists prove that languages are related?

The comparative method compares languages feature by feature, starting with basic vocabulary that is rarely borrowed: numbers, body parts, family members, natural things. The key is not that words look alike, since any two languages share some lookalikes by chance, but that differences repeat in a pattern. The idea took hold after 1786, when the British judge William Jones observed that Sanskrit, Greek and Latin resembled each other too closely to be chance and must have "sprung from some common source," as the history of Proto-Indo-European recounts.

Worked example: mother, three and night

English	German	Latin	Greek	Sanskrit	Reconstructed ancestor
mother	Mutter	māter	mētēr	mātar-	*méh₂tēr
three	drei	trēs	treis	trayas	*tréyes
night	Nacht	nox (noct-)	nyx (nykt-)	nakt-	*nókʷts

Read the table column by column.

1. All five languages start "mother" with m, "three" with a t-like sound plus r, and "night" with n. Similar, but that alone could be luck.
2. Look closer at "three." Latin, Greek and Sanskrit have t; English has th; German has d. Now check other words: Latin tū, English thou, German du; Latin tenuis, English thin, German dünn. The same mismatch repeats across hundreds of words, and a mismatch that repeats is evidence; a match that happens once is not.
3. The pattern extends to other sounds. Where Latin has p, English has f: pater and father, pēs and foot, piscis and fish. Where Latin has c (a k sound), English often has h: canis and hound, centum and hundred.
4. These correspondences were set out by Rasmus Rask in 1818 and by Jacob Grimm in 1822 and are known as Grimm's law: the ancestor of the Germanic languages shifted p to f, t to th and k to h in one systematic change.
5. Then the exceptions. By Grimm's law, the t of Latin pater should have become th in Germanic, yet Old English had fæder, with a d; modern father gained its th only a few centuries ago. In the 1870s the Danish linguist Karl Verner showed why: the shifted sound became voiced when the ancestor word's stress did not fall on the syllable just before it. In the word for "father" the stress came after the t; in the word for "brother" it came before, and Old English brōþor kept its th. English mother went through the same step as father, Old English mōdor, before its d also became th. Even the exceptions follow rules.
6. Working backward from all the descendants at once gives a reconstruction of the ancestor's word, marked with an asterisk because no one ever wrote it down.

That is the difference between kinship and coincidence. English have and Latin habere look alike and mean the same, but by the correspondences above, a Latin h should not match an English h, so they are not cognates, words with a shared origin. In fact have matches Latin capere, "to take," exactly as the rule of Latin c for English h predicts. Lookalikes fool; patterns don't.

What are the largest language families?

Linguists recognize well over 100 families, plus many isolates, languages with no known relatives. The five largest by number of languages, with Trans–New Guinea (about 380) close behind, are also among the largest by number of speakers:

Family	Languages (Glottolog 5.0, 2024)	Speakers (Ethnologue, 2019)	Includes
Atlantic–Congo	about 1,410	about 500 million	Swahili, Yoruba, Zulu
Austronesian	about 1,270	about 330 million	Malay, Tagalog, Malagasy, Māori
Indo-European	about 590	about 3.2 billion	English, Hindi, Spanish, Russian, Persian
Sino-Tibetan	about 510	about 1.4 billion	Mandarin, Cantonese, Burmese, Tibetan
Afroasiatic	about 380	about 500 million	Arabic, Hebrew, Amharic, Hausa

The figures come from the Wikipedia list of language families, which takes its language counts from Glottolog and its speaker numbers from Ethnologue. Counts vary between catalogs, because deciding where one language ends and another begins is partly a judgment, a problem the note on what language is explains. Austronesian is the most spread out: its languages run from Madagascar to Easter Island. Indo-European has the most speakers: it already stretched from Europe to India some 3,000 years ago, and European empires and migration over the last 500 years carried it to the Americas and beyond.

What is a language isolate?

An isolate is a language whose relatives, if it ever had any, have all disappeared. Basque, spoken by about 806,000 people in northern Spain and southwestern France in 2021, is Europe's only known isolate: it is older in its region than the Indo-European languages that surround it. Korean is sometimes treated as an isolate or as the main member of a tiny family, and Japanese belongs to the small Japonic family. An isolate is a family of one, and it shows that the families we can trace are only the survivors.

Why are proposed superfamilies argued?

Sound change works like erosion. After enough millennia, the correspondences that prove a relationship wear away until they can't be told from chance. Most linguists hold that the comparative method reaches back somewhere around 6,000 to 10,000 years, and no further with confidence.

Still argued

Some linguists have proposed superfamilies that would join known families: Nostratic, linking Indo-European with Uralic, Altaic and others; or a single ancestor of all languages. Supporters point to shared pronouns and roots across families. Critics answer that at that depth, chance resemblances, borrowing and loose matching standards can produce lists that look convincing but fail the test of regularity. A few long-range links have won wider support, such as Dene–Yeniseian, which would connect Na-Dene languages of North America with the Yeniseian languages of Siberia, though even that is often described only as plausible. Where Proto-Indo-European itself was spoken is also debated: the steppe north of the Black Sea, backed by ancient DNA evidence such as a 2015 study tracing a large migration from the steppe into Europe about 4,500 years ago, or Anatolia, with the spread of farming.

What a language family doesn't tell you

Two warnings keep the idea honest. First, related doesn't mean mutually intelligible. English and German share an ancestor, but an English speaker can't follow a German conversation; English and Hindi are distant cousins, yet the relationship is visible only through careful comparison like the table above, starting from Sanskrit.

Second, a language family is a history of languages, not of peoples. Languages spread by conquest, trade, schooling and prestige, and whole populations switch. Hungarians speak a Uralic language related to Finnish, yet are genetically much like their Central European neighbors. Even a built language like Esperanto borrows most of its roots from Indo-European languages without belonging to the family by descent at all.

Check yourself

• Without looking back: why does a sound mismatch that repeats across many words prove more than a single word that looks the same?
• How does Verner's law turn an apparent exception to Grimm's law into further evidence?
• Why can't the comparative method show whether all the world's language families share one ancestor?

Sources

• Massive migration from the steppe was a source for Indo-European languages in Europe — Haak and colleagues, Nature, 2015
• Grimm's law — Wikipedia
• Proto-Indo-European language — Wikipedia
• List of language families — Wikipedia, from Glottolog 5.0 and Ethnologue
• Glottolog: a catalogue of the world's languages, families and dialects — Max Planck Institute for Evolutionary Anthropology
• Basque language — Wikipedia