SIG-2026-017 · A tax on rare languages

In brief: The Token Tax study of 10 models and 16 African languages found that inefficient tokenisation makes texts in low-resource languages cost more and come out worse.

Date / period: 2026.
Geography: Africa.
Type and stage: Research or data.
Scenario: CS — Centralized — Synthetic.

Why this scenario: A system, platform or infrastructure makes the decision, and the person receives a ready result.

Collection trend: The new inequality — Access to technology and upgrades splits people into castes. Segregation is back in conversations about the future.

Macrotrend drivers: MT-038 — Inequality and resource distribution; MT-057 — Radical longevity; MT-065 — Transformational economy (value vertigo).

Life domains: Finance; Values, meaning and community.
Thematic tags: inequality

Description

The Token Tax study of 10 models and 16 African languages found that inefficient tokenisation makes texts in low-resource languages cost more and come out worse.

Why this is a signal

Inequality is built into the technology: you pay more for your native language.

Where the signal comes from

Research or data — aclanthology.org.

Limits of the conclusion

Figures come from the study or survey given; sample and method are the authors’. Assigning it to a Decoder world is our reading.

Source: aclanthology.org.


All signals in this collection: Signals: big trends, November 2026