Skip to content

MMethodology

How every number is made.

An accessible overview of our method — the metrics, the proficiency threshold and the data sources behind every figure on this site.

Full document

Methodology

The complete technical methodology, with the full equations, worked examples and dataset schemas.

Read the full PDF ↗

01 — The three metrics

Three numbers carry the whole argument.

Everything here reduces to three quantities. Each has a precise definition and a fixed way of being displayed across the site.

ALPS

Average Language a Person Speaks

The average number of languages a person speaks at A2+ in a given population. The global, population-weighted value is 1.57, with a plausible range of 1.42–1.70.

Displayed: country to 2 dp · global 1.57 · uncertainty as the band 1.42–1.70

GLCI

Global Linguistic Connectivity Index

The probability that two randomly chosen people share a language at A2+, counting mutual intelligibility. Currently 11.3% — the basis of the “89% disconnected” framing.

Displayed: as a percentage (11.3%) · inverse as 89% · cross-country companion 7.5%

EGCT

Effective Global Connectivity Threshold

The average repertoire at which connectivity reaches a stated level under a stated distribution rule. The reference level is 90% connectivity under the bridge rule, which works out at 2.70 languages per person. Both halves matter: the same 90% costs about four times as much under a proportional rule, and is unreachable under a uniform one.

Displayed: the level and the rule together — never a bare number

02 — The proficiency threshold

Why we count at A2+.

A person “speaks” a language, for our purposes, when they reach A2+ on the Council of Europe's CEFR scale — the point at which someone can sustain everyday conversation.

Setting the bar here is deliberate. Lower, and a few words of holiday vocabulary would count as connection; higher, and we would miss the millions who communicate perfectly well without fluency. A2+ keeps the connectivity graph honest about who can actually talk to whom.

The same threshold is why liturgical languages are not counted as second languages: they are recited rather than used in everyday conversation, and the everyday conversation is what this index measures.

CEFR scale · the edge is drawn here

A1 · Beginnerexcluded
A2+ · Conversational↞ threshold
B1–B2 · Independentcounted
C1–C2 · Proficientcounted

03 — How we measure

From a list of languages to a single probability.

01

Count speakers

For each of the 934 languages and language groups in the index, count L1 and L2 speakers at A2+ from the Language Speakers Index.

02

Build the graph

Treat each person as a node; connect any two who share a language at A2+.

03

Apply the model

The observed language distribution is turned directly into a pairwise-sharing probability. A random-graph null model is computed alongside it, and rejected: it is off by more than an order of magnitude.

04

Publish the figures

ALPS and GLCI are computed from national sources — censuses and language surveys — each checked against its source. How strong the evidence behind each national estimate is, and whether it comes from a direct survey, a census, a proxy or a model, is documented in the methodology (Paper 1).

All heavy computation happens upstream in our data pipeline. The site renders pre-computed scores from versioned JSON datasets — it never calculates the graph in your browser.

04 — Data sources

Triangulated, and re-checked.

No single source covers the whole world reliably, so every jurisdiction is built from several and reconciled. When the CIA World Factbook shut down in February 2026, every figure that had relied on it was revalidated against the sources below.

01EthnologueLanguage inventories
02UNESCOEducation & literacy
03EurobarometerEU language self-report
04National censusesPer-country detail
05Regional barometersAfro / Latino / Asian
06Language Speakers Index934 languages and language groups at A2+

Want the complete method?

Methodology has the full equations, worked examples and dataset schemas.