Which languages does AI Uniti monitor?
The behavioural half of AI Uniti's detection works in any language, because tempo, timing and network shape are numbers rather than words. The semantic half reads the major world languages and reports its findings in English. We list only languages we have collected, scored and clustered on a real production corpus.
Short answer: the behavioural half of our detection works in any language. The semantic half reads the major world languages and reports its findings in English.
Most of what we do is language-independent by construction
Signal scores how accounts behave, not what they say. Roughly two thirds of the risk instrument is deterministic behavioural measurement: volume, velocity, timing concentration, and how a conversation moves against a brand’s own prior window.
Tempo, timing and network shape are numbers, not words. They read identically in Serbian, Japanese or Arabic, and they need no per-language work at all. That is the layer coordinated manipulation cannot hide from, because content improves every month and behaviour does not.
The semantic layer reads broadly and reports in English
Sentiment, tone and narrative clustering are produced by a large language model reading the post text. It handles a wide range of languages and scripts, including right-to-left scripts and corpora that mix languages.
Findings are always rendered in English. An analyst who speaks no Serbian still reads the Serbian conversation: the cluster labels and summaries arrive in English regardless of the source language.
Languages we have run in production
| Region | Languages |
|---|---|
| Europe | English · Spanish · Czech · Turkish · Serbian, and with it Bosnian, Croatian and Montenegrin |
| Asia | Japanese · Korean · Chinese, both Simplified and Traditional · Bengali · Hindi · Urdu |
| Middle East and North Africa | Arabic |
Every language listed here has been collected, scored and clustered on a real production corpus. We do not list a language we have only assumed would work.
Scripts
Latin including diacritics · Cyrillic · Han, simplified and traditional · Kana · Hangul · Arabic, right to left · Bengali · Devanagari.
Brands are seeded in every script their audience actually uses. Serbian, for example, is collected in both Cyrillic and Latin as a matter of course, because a term estate in one script silently misses half the conversation.
Serbian, Bosnian, Croatian and Montenegrin
These four standard languages are near-identical in written form, and for monitoring they behave as one language across two scripts. A single term estate covers all four. The discipline that matters is scripts, not languages: Serbian and Montenegrin use both Cyrillic and Latin, while Bosnian and Croatian are Latin-dominant.
What we do not claim
- Absence from the list means unobserved, not unsupported. The underlying models read far more languages than we have run production corpora in. If a language matters to you and is not listed, ask us, and we will tell you what we know rather than what we hope.
- We treat per-language sentiment quality as validated only where we have verified it with a customer on their own data, during an evaluation. Observed in production and validated with a customer are different claims and we keep them apart.
- Analyst-facing output is English. We do not currently produce localised summaries.
A language that matters to you
Every evaluation begins by seeding your term estate in the languages and scripts your audience actually uses, and we validate detection quality on your own data before either of us calls it proven. If the language you care about is not on the list above, that is a conversation worth having rather than a dead end.
Frequently Asked Questions
Which languages does AI Uniti support?
AI Uniti has run production corpora in English, Spanish, Czech, Turkish, Serbian (and with it Bosnian, Croatian and Montenegrin), Japanese, Korean, Chinese in both Simplified and Traditional, Bengali, Hindi, Urdu and Arabic. The behavioural half of the detection is language-independent by construction, so it works in any language; the list describes where we have production evidence, not the limit of what the platform can read.
Does coordination detection need to understand the language?
No. Coordination detection scores how accounts behave rather than what they say: volume, velocity, timing concentration and network shape. Those are numbers, not words, and they read identically in Serbian, Japanese or Arabic. That is the layer coordinated manipulation cannot hide from, because content improves every month and behaviour does not.
What language are the findings reported in?
English. Cluster labels and narrative summaries are always rendered in English regardless of the source language, so an analyst who speaks no Serbian can still read the Serbian conversation. We do not currently produce localised summaries.
What if the language I care about is not listed?
Absence from the list means unobserved, not unsupported. The underlying models read far more languages than we have run production corpora in. Every evaluation begins by seeding your term estate in the languages and scripts your audience actually uses, and we validate detection quality on your own data before either of us calls it proven.
How does AI Uniti handle Serbian, Bosnian, Croatian and Montenegrin?
These four standard languages are near-identical in written form and behave as one language across two scripts for monitoring purposes, so a single term estate covers all four. The discipline that matters is scripts rather than languages: Serbian and Montenegrin use both Cyrillic and Latin, while Bosnian and Croatian are Latin-dominant.