Terrior

Localisation is usually done at the level of a language or a country, i.e., one Hindi version is used in both Delhi and Lucknow, and one German version in both Munich and Hamburg. Terrior is intended for the region where the text will be read. More about it is available at terrior.app.

Terrior measures three kinds of difference between regions. First, word choice: where a language has several competing words for the same thing, a region tends to settle on one of them. Second, mixing: where a region writes one language with words borrowed from another, it uses the borrowed word instead of the native one at a rate specific to that region. Third, divergence: some words are shared by a region and its neighbours, but are understood differently on either side of the boundary. This third measure is not yet used in localisation.

Two samples of writing never match exactly, even when they come from the same place, and some difference between two regions is therefore expected by chance. For this reason, we calibrate each measurement against a within-region noise floor, i.e., the difference observed when a sample from one region is compared with another sample from the same region. A word is reported as regional only when the difference between the regions is clearly larger than this floor. We then check the results against words that dictionaries already record as regional.

Word choice in German: Germany, Austria and Switzerland

September 2026

We first tested the measure of word choice on words that dictionaries already record as Austrian or Swiss. Of those that occurred in the material, 98% were assigned to the correct country. On a control set of common words with no known regional alternative, 4% were marked as regional in error.

We then applied the measure to everyday commercial vocabulary, and found words that one country uses considerably more often than the other two, e.g., brutto in Austria, Gutschein and Versandkosten in Germany, and Kundendienst and Abo in Switzerland.

We also measured differences in meaning, i.e., cases where a word is understood differently on either side of a border. Although some words clear the within-region noise floor, most of them are discourse adverbs such as wiederum and daneben, whose use varies within every country rather than between countries. Almost no commercial vocabulary is among them.

English in Hindi: Delhi and Lucknow

September 2026

Delhi and Lucknow share a written standard. We measured 113 concepts, each of which can be written either with an English loanword or with a Hindi word, e.g., रिव्यू or समीक्षा for review. The two cities discuss these concepts equally often. However, they differ in which word they choose, with Delhi using the English word 40% of the time and Lucknow 28%. Of the 102 concepts where the cities differ at all, Delhi uses more English for 87 (sign test, p = 1.7 × 10⁻¹³). The difference is not limited to commercial words, and is also present in general vocabulary.

The difference between the cities appears in two independent sources of text. However, the absolute level of English does not carry over from one source to the other, and within one city it varies between sources about as much as it varies between the cities. We believe that, for this reason, the difference between the cities is the reliable result, rather than the level of English in either city.

Terrior is being built. We have completed the measurement for German in Germany, Austria and Switzerland, and for Hindi in Delhi and Lucknow. We have also designed a study in which reviewers from each region compare text localised by Terrior with standard localisation, although it has not yet been run. To know more about Terrior, or to stay in touch while it is being built, you can contact us through terrior.app.