Glottolog counts 7,674 languages — and dates the answer

"There are about 7,000 languages in the world." It is one of the few numbers in this field that is basically right, which makes it a good case for a different question: right to what precision, and decided by whom?

One catalogue, one versioned count

Glottolog, maintained at the Max Planck Institute for Evolutionary Anthropology, states its figure with a date attached: as of 2 March 2026 there are 7,674 spoken L1 languages, sorted into 246 families and 183 isolates.

That is not the total of every entry in the database. Glottolog separately lists sign languages, pidgins, mixed languages, artificial languages and other categories. Its 7,674 figure has a defined scope: spoken languages traditionally used by a community as a first language.

The date matters as much as the figure. The count can change when new documentation is added or classifications are revised. A number quoted without a year and a source is a snapshot presented as a constant.

Where judgement enters

Before a variety contributes one to the total, editors have to decide whether the evidence supports treating it as a distinct language. Glottolog publishes a decision procedure and says plainly that its classification is the editors' best estimate. It also warns that classifications can change as historical-comparative research improves.

If a putative language does not meet the criteria for a distinct language, Glottolog treats it as a dialect or records it as based on a misunderstanding. This is not the same as counting people or countries. It is a maintained scholarly classification built from the linguistic documentation available at that version.

That is why a precise worldwide total without a catalogue name, version and scope is false precision. Another catalogue can reach another defensible result by defining its inventory differently; the number alone does not tell you which decision changed.

The version of this decision that costs money

This is not an abstract taxonomy problem for a learner, because course and difficulty tables also have to decide which varieties get separate entries.

FSI lists Cantonese and Mandarin as separate training entries, but that operational choice does not settle how every catalogue should classify them. It does show why a learner cannot plan from the English word “dialect” alone.

The same shape appears within a single language: Arabic's varieties differ enough that choosing which one to learn is a real fork, and a course sold as "Arabic" may not be the one you needed.

The useful takeaway

7,000 is fine as a round figure. If precision matters to an argument, take the count from a catalogue that publishes its date, version, scope and criteria — and report all four.

That is the same standard we aim to hold our own figures to. A source URL is not enough by itself: the source can change, so the fetch date and the wording actually verified matter too.