How to read these pages

Every number on this site comes from one source, WALS Online, and from calculations made on it. This page explains what the source covers, what each kind of page shows, and how the computed figures are defined.

The source

The World Atlas of Language Structures (WALS) is a database of structural properties of languages, edited by Matthew S. Dryer and Martin Haspelmath and published by the Max Planck Institute for Evolutionary Anthropology. Its chapters were written by specialists, each describing one feature and coding a sample of languages for it. We use the CLDF release v2020.4 of the 2013 online edition (doi:10.5281/zenodo.13950591), licensed CC BY 4.0. Citation: Dryer, Matthew S. & Haspelmath, Martin (eds.) 2013. The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. (Available online at https://wals.info)

WALS lists 2,662 languages and 192 features. It is a sample, not a census: each chapter coded the languages its author could document, so some features cover about a hundred languages and others more than fifteen hundred. All counts here are therefore phrased as “of the languages WALS codes for this feature”. A value that no other language in the atlas shares is only rare within the sample; we never claim it is unique in the world.

Profile features

Language pages use 144 of the 192 features. We leave out sub-features that repeat another feature for a subset of languages, features that combine two others, the two features on sign languages, the writing-system feature, and the detailed negation chapters whose categories need a long explanation to be read correctly. Feature pages exist for every profile feature with at least 100 coded languages (140 features).

Which languages have a page

A language has a page when WALS codes it for at least 40 profile features (409 languages) and when at least two of the distinctive facts listed below apply to it (368 languages). The contrast tool covers every language with at least 30 coded features (581 languages). Coded values can be partial: a language page shows the features WALS has coded for that language and nothing else.

Reading a language page

The page opens with the language’s basic order of subject, object and verb, followed by its strongest distinctive facts. The diagram shows nine pairings of a head word (verb, adposition, noun, adjective) and its dependent. The dependent is drawn on the side the language uses; a dashed pair means WALS records no dominant order or a mixed type. The thin bar under each row shows how all languages coded for that pairing divide: the blue part puts the dependent first, the orange part after, and the grey middle covers mixed and other types. The tables below list every coded value with the number of languages in the atlas that share it.

Computed figures

• Languages sharing a value: the number of languages WALS codes with the same value for that feature, out of all languages coded for it.
• Uncommon value: a value shared by fewer than 10% of the languages coded for a feature with at least 100 coded languages. Below 5% it counts as a distinctive fact.
• Family majority: for each feature, the most frequent value among the other members of the language’s family (WALS’s classification) that are coded for it, when there are at least three. A language breaks with its family when at least five other members are coded, at least 80% of them share one value, and none has the language’s own value. The same test is applied within its branch (WALS “genus”) when the branch is smaller than the family, with at least four other members coded.
• Agreement with the family: the share of those features on which the language has the family’s most frequent value. Among families with at least four profiled members, the members with the highest and lowest rates are noted.
• Ordinariness score: the average, over all the language’s coded profile features, of the share of languages in the atlas that have the same value. A low score means the language often has values few others share. It depends on which features are coded, so it is compared only among the 409 languages with at least 40 coded features; the ten lowest and ten highest are noted as distinctive facts.
• Closest profiles: for every pair of languages with at least 40 coded features in common, the share of those features with the same value. Pairs are ranked by the lower bound of a 95% Wilson interval on that share, so that a high agreement on few features does not outrank a slightly lower agreement on many.
• Most fully coded: the number of profile features WALS codes for the language; the ten highest are noted.

Feature, family and area pages

Feature pages show one world map per value. Each dot is a language at the point WALS gives for it; coastlines are from Natural Earth (public domain), and the map is centred on the Pacific so that no language area is split. Family pages compare members that have at least 30 coded features with all languages coded for each word-order pairing, and list the features on which at least eight coded members agree or disagree most. Macro-area pages use WALS’s six areas: Africa, Eurasia, Papunesia (New Guinea, the Pacific and island South-East Asia), Australia, North America and South America.

What the data cannot say

WALS describes the standard or best-documented variety of each language as found in its sources, which date from many decades. A match between two languages means they fall into the same category, not that they share words or history. The atlas does not rank languages by difficulty, and nothing here is advice on which language to learn.