Questions and answers

ALPS stands for Average Language a Person Speaks. It is a population-weighted average: the number of languages each person speaks, averaged across a population. The current global estimate is 1.57 (band 1.42–1.70), built for 237 countries and territories home to 8.2 billion people.

Unlike many instruments, ALPS counts an unlimited number of languages per person instead of stopping at a first and second language or capping the count at three or four.

Conversational competence at CEFR A2 and above, the level of the Common European Framework of Reference at which a person can handle routine everyday exchanges. The threshold is deliberately stricter than "knows some words" and deliberately looser than "native-level mastery".

Fixing it is what makes the comparison possible. Because the censuses and surveys we draw on apply different and often unstated thresholds, a large part of the methodology consists of harmonizing them to this single standard, with each adjustment documented.

Country by country, from that country's best available evidence (censuses, large survey programmes, national statistical offices, documented language surveys), each source harmonized to the CEFR A2+ threshold and to common counting rules, then checked against independent evidence. Where sources conflict, the conflict is reported and the likely direction of the remaining error is stated.

The connectivity index adds a further layer: partial comprehension between related languages, handled through a documented mutual intelligibility matrix.

The full methodology is published with the papers: the ALPS paper and the GLCI paper on SocArXiv. The assumption registers, the sensitivity of the headline figures to alternative conventions, and the data and code needed to reproduce every number are on the Open Science Framework. Specific questions are welcome at research@lingacon.com.

Mostly because published numbers rarely measure the same thing. A person counted as bilingual under one survey's threshold may not qualify under another's; some sources count languages "known", others languages used at home, others only a mother tongue plus one foreign language.

Our methodology differs from these in three specific ways. We fix a single proficiency threshold (CEFR A2+) and harmonize every source to it. We count an unlimited number of languages per person, where many instruments stop at two to four. And we check each national figure against independent evidence instead of adopting one source wholesale.

The result is comparable across all 237 jurisdictions, and therefore sometimes different from figures produced under other definitions. Where our number departs from a familiar one, the reason is documented, and readers are free to disagree with the convention once they can see it.

It is the probability that, of two people drawn at random, the one listening understands the one speaking. Aggregated over all pairs it becomes an index of how connected a population is: the Global Linguistic Connectivity Index.

The global value is 11.3%. Within a single country it is about 63%; between two different countries it falls to 7.5%, so in more than nine of every ten cross-border pairs the listener would not understand the speaker.

One precision worth keeping: the index measures comprehension, whether one person can understand the other. A two-way conversation is a stricter test, and comprehension is often asymmetric; the index is built to reflect that.

Yes, and it matters a great deal. A Spanish speaker follows a good deal of Portuguese; a Czech speaker understands much of Slovak; between neighbouring Arabic varieties comprehension varies sharply depending on which varieties are involved.

LINGACON handles this with a mutual intelligibility coefficient, a documented matrix of partial comprehension between related languages. Measured values from the experimental literature are used where they exist; the remaining pairs carry assessed values under written rules and are marked as such in the published matrix. The matrix is asymmetric, because comprehension frequently is.

Because a global average should represent the globe. The dataset covers 237 countries and territories, dependent territories as well as UN member states, home to 8.2 billion people (UN World Population Prospects 2024). 216 of them carry language-by-language detail.

It means the average person on Earth speaks between one and two languages, a little over one and a half. Behind that average lies wide variation: some societies are overwhelmingly monolingual, while in Luxembourg, the highest value backed by a direct survey, the average resident speaks 3.33 languages.

The average is also a weak predictor of connectivity. Which languages people learn matters a great deal: a country can be highly multilingual and still weakly connected to people outside it.

Yes. The data and code behind the ALPS and GLCI papers are on the Open Science Framework, with licences stated for each component. Please cite the paper the figures come from.

For uses the deposit licences do not cover, or for detail beyond the deposit, see Partnerships or write to research@lingacon.com.

ALPS figures: Hovhannisyan, A. (2026a), doi.org/10.31235/osf.io/3mtsx_v1. Connectivity figures: Hovhannisyan, A. (2026c), doi.org/10.31235/osf.io/g3vxj_v1. The standard: Hovhannisyan, A. (2026b), osf.io/preprints/socarxiv/n96g3. Full references are listed under Publications.

LINGACON is independent and receives no external funding. You can support it financially, sponsor one of the planned studies, collaborate on data collection, or help connect the project with survey institutions; see Contribute and Partnerships.