# Multilingual lexical uptake Part of [The AI Tells Index](https://feedsquad.com/ai-tells). A tell signals low effort. It does not identify an author. Skilled writers produce every shape listed here, some of them daily, and automated detectors misread those writers for it at rates measured above 60 percent on non-native English prose. Nothing in this index proves that a machine wrote anything. Read an entry as one piece of evidence to weigh against the false-positive notes printed beside it. ## Facts - Id: multilingual-lexical-uptake - Category: Lexical (lexical) - Subcategory: non-english - Also known as: cross-language overuse profile, non-English marker lists, LexA language coverage - Status: Active. Currently signals low-effort writing. - Severity: medium - Evidence grade: corroborated (named independently by multiple credible secondary sources) - Languages: en, af, ar, bg, cs, de, el, es, et, fa, fi, fr, hi, hr, id, is, it, ja, kk, ko, ky, lt, lv, mr, nl, pl, pt, ro, ru, sr, ta, tr, uk, zh - Added: 2026-08-15 - Updated: 2026-08-17 - Page: https://feedsquad.com/ai-tells/multilingual-lexical-uptake ## Description The post-2022 lexical shift is not an English problem. Juzek tested 34 languages and found post-ChatGPT lexical uptake in 26 of them, with a mean prevalence increase of 15.1 percent. The companion CC0 dataset covers those languages in a news register. The English word lists that circulate do not transfer: a Finnish or German text does not become suspect because it contains a translation of delve, and no per-language marker list is published that we could responsibly reprint. This entry exists to record the finding and to stop the English list being applied where it does not belong. ## Why it reads as machine-written The same pull produces the same shape in any language: a verb of depth, a noun of significance, and no date anywhere in the paragraph. What changes across languages is the vocabulary, not the emptiness. ## Detection Type: statistical (a measured metric against a threshold with a stated basis) Metric: distinct-per-language-uptake-markers-per-1000-words Threshold: 3 (fires when above) Threshold basis: > Carried over from the English density trigger, and weaker here. Juzek reports 26 of 34 languages showing post-2022 uptake with a mean prevalence increase of 15.1 percent, all at corpus level, and publishes no per-document cutoff and no per-language marker list. The metric is named separately from the English one because the marker list differs by language and the two numbers are not comparable. Three distinct markers per 1,000 words is therefore a FeedSquad review trigger with no per-language calibration behind it at all. Anyone applying this outside English must build the marker list from a measured corpus in that language first. Running the English list through a translator produces nothing worth reading. ## Examples Constructed specimens. Written for this index. Never quoted from anyone's posts. ### Specimen 1 Before, exhibiting the tell: > Tämä artikkeli syventyy etätyön monimutkaiseen kokonaisuuteen ja korostaa selkeän viestinnän merkitystä. Luottamus rakentuu hitaasti ja katoaa nopeasti, minkä useimmat tiimit oppivat vasta jälkikäteen. Toimivat käytännöt syntyvät harvoin kerralla, vaan ne muotoutuvat matkan varrella. Lopulta ratkaisee se, miten arjessa oikeasti toimitaan, ei se, mitä ohjeeseen on kirjoitettu. After, repaired: > Siirryimme etätyöhön maaliskuussa. Palaverit vähenivät neljästä kahteen viikossa, ja kirjoitettu muistio korvasi loput. Muistio on nyt ensimmäinen asia, jonka uusi työntekijä lukee. Note: The specimen is Finnish. Figures in this repair are invented for the specimen. The before text makes the same move as the English version: a verb of depth and a noun of significance where a month and a number would do. ### Specimen 2 Before, exhibiting the tell: > Dieser Beitrag beleuchtet die vielschichtigen Herausforderungen der Fernarbeit und unterstreicht die Bedeutung klarer Kommunikation. Vertrauen entsteht langsam und geht schnell verloren, was die meisten Teams erst im Nachhinein bemerken. Gute Gewohnheiten lassen sich nicht verordnen, sie wachsen mit der Zeit. Am Ende zählt, wie im Alltag tatsächlich gearbeitet wird, und nicht, was in einem Leitfaden steht. After, repaired: > Wir haben im März auf Fernarbeit umgestellt. Aus vier Meetings pro Woche wurden zwei, den Rest übernimmt ein schriftliches Protokoll. Note: The specimen is German. Figures in this repair are invented for the specimen. ## False positives Who legitimately writes this way. Translators and staff at multilingual institutions write a formal register in every language they publish in, and that register is a large part of what the measurement picks up. European Union and United Nations texts are drafted to be translatable, which flattens exactly the features a marker list keys on. Second-language writers in any target language produce the most predictable register available to them, which is the same mechanism behind the 61.3 percent average false-positive rate Liang and colleagues measured across seven detectors on non-native English essays. Languages with rich morphology break naive word matching outright, so a Finnish marker count that ignores inflection measures tokenisation rather than writing. A reviewer working outside English should treat this entry as a warning about method, not as a detector. ## Model attribution Corpus-level and cross-model. The LexA news data was generated with a small 2025-era model and the science data with several families, so no single vendor sits behind the finding. ## Sources 1. Juzek, LLM lexical uptake across 34 languages, preprint (arXiv:2605.25358) https://arxiv.org/abs/2605.25358 (tier: primary-doc; accessed 2026-08-14) 2. LexA-Index, CC0 per-language overuse dataset https://github.com/fsu-nlp/lexa-index (tier: primary-doc; accessed 2026-08-14) 3. Kobak et al., Delving into LLM-assisted writing in biomedical publications, Science Advances 11(27) https://www.science.org/doi/10.1126/sciadv.adt3813 (tier: peer-reviewed; accessed 2026-08-14) 4. Liang et al., Widespread LLM adoption in society, Patterns (arXiv:2502.09747) https://arxiv.org/abs/2502.09747 (tier: peer-reviewed; accessed 2026-08-14) 5. Liang, Yuksekgonul, Mao, Wu, Zou: GPT detectors are biased against non-native English writers, Patterns 4:100779 (PubMed Central copy) https://pmc.ncbi.nlm.nih.gov/articles/PMC10382961/ (tier: peer-reviewed; accessed 2026-08-14) ## Status history Ids are permanent. A retired tell keeps its id and its page. - 2026-08-15, Active: Active. Juzek measures uptake in 26 of 34 languages and the companion dataset covers them; Liang and colleagues find the same shift in consumer complaints, corporate communications and United Nations press releases at population scale. The source is a preprint committed to EMNLP 2026 and is labelled as such on the entry. ## License CC BY 4.0. https://creativecommons.org/licenses/by/4.0/ Attribution: The AI Tells Index, feedsquad.com/ai-tells Reuse the data, including commercially. Keep the attribution line. --- Dataset version 1.0.0. Schema version 1. Part of [FeedSquad](https://feedsquad.com). Built by [Herman Foundry](https://hermanfoundry.com) from Levi, Finnish Lapland.