Twenty-five years of Turkish knowledge — read not for its words but for its
behaviour: what a society chose to record, what it fought over, and how it knows the
things it claims. Three independent chronicles fall out of the same archive.
An independent corpus of 277,000 Turkish news articles (2008–2019). Taking the words
whose share of articles surges in a given year — corrected against every prior year —
reconstructs the era from the wire, with no markup and no bots.
01 · the news chronicle
A quarter-century, in the words that spiked
Each year names itself. The surging vocabulary is a clean fingerprint of what the country was living through — revolutions, disasters, a coup, an election.
Watch the word before “erdoğan” in the news. Başbakan (prime minister) Erdoğan falls to literally zero by 2016 as cumhurbaşkanı (president) Erdoğan takes over — they cross in 2014, the year he won Turkey's first popular presidential election. The office of prime minister then passes to Davutoğlu, to Yıldırım, and after the 2018 switch to a presidential system evaporates from the language entirely. A constitutional epoch in a single collocation.
Part II
What we fought about
Wikipedia records its own conflict. A revision that restores an earlier exact state of a page
is a revert — the trace of an edit war. Their rate over time is a contention index; their
targets are a society's fault lines.
03 · the contention curve
When the encyclopedia fought hardest
The share of article edits that undo a prior state climbed to a 12.7% peak in 2014 — the height of political polarisation — fell to 4% by 2018–19, then rebounded. Conflict has a history of its own.
04 · the edit-war chronicle
What got fought over, year by year
Reverts concentrated in a single year name that year's flashpoints — an event timeline built from conflict rather than news, and it rhymes: Gezi, ISIS, the coup, then K-pop fandom, the earthquake, the İmamoğlu protests.
2010
Rey MysterioKurtuluş HareketiWWEThe Undertaker2010 Eurovision Şarkı YarışmasıKavak Yelleri (dizi, 2007)Devrimci Liseliler
2011
Beşiktaş TVHikmet Kıvılcımlı2011Türkiye'de engelli haklarıEngellilikNihat Doğan
2012
Fenerbahçe-Galatasaray derbisi2012 Eurovision Şarkı YarışmasıMahidevran HâtunMoussa SowŞehzade Mehmed (I. Süleyman'ın oğlu)2013 Eurovision Şarkı YarışmasıFelipe Melo
2013
Gezi protests + the Ottoman TV culture war
Didier Drogba2013 Eurovision Şarkı YarışmasıGezi Parkı olaylarıMuhteşem YüzyılYounès BelhandaÓscar CardozoPis Yedili (dizi)
2014
ISIS hits the record
Demba BaIrak ve Şam İslam DevletiKerem Bürsin
2015
The HDP and an election year
Robin van PersieNaniZlatan IbrahimovićHalkların Demokratik PartisiKaçak (dizi, 2013)Eskitorunobası, EvrenHaziran 2015 Türkiye genel seçimleri için yapılan anketler
2016
The coup, hour by hour
15 Temmuz Darbe GirişimiPoyraz KarayelFırat Kalkanı HarekâtıHande ErçelVasip ŞahinÖmer HalisdemirEmre Mor
2017
Vladimir Lenin bibliyografyasıGumball bölümleri listesi
Survivor 2022: All StarZafer PartisiDataizmÜlke çapında 2023 Türkiye genel seçimleri için yapılan anketlerCan YamanKuruluş Osman karakterleri listesiAy Yapım
2023
The Kahramanmaraş earthquake + the election
2023 Kahramanmaraş depremleriBen Bu Cihana SığmazamÜlke çapında 2023 Türkiye genel seçimleri için yapılan anketlerGelinim MutfaktaNazmiyeSurvivor 20232023 Türkiye cumhurbaşkanlığı seçimi için yapılan anketler
2024
Reality TV
Survivor 2024: All StarHudutsuz Sevda2024'te Türkiye'de televizyonYalı ÇapkınıKızılcık ŞerbetiKirli SepetiCNBC-e
2025
İmamoğlu and the protests
ZippoSahipsizler (dizi)Survivor 2025Ekrem İmamoğlu2025 Türkiye protestoları
05 · the fault lines
The permanently contested
Across all 25 years, the most edit-warred articles map a society's third rails — national history, the Armenian question, the Kurdish question, religion, the leadership — interleaved with football and television. And these never settle: of the oldest, most-developed articles only 0.3% ever go dormant — the national-identity core (Türkler, Öcalan, Kıbrıs, the eastern provinces) is reopened every single year.
Osmanlı İmparatorluğu2,637Türkiye2,544Cristiano Ronaldo1,957Anasayfa1,950Arka Sokaklar (dizi)1,843Ermeni Kırımı1,826Fenerbahçe SK1,755Adalet ve Kalkınma Partisi1,725Mustafa Kemal Atatürk1,623İstanbul1,561PKK1,550Fethullah Gülen1,535Kurtlar Vadisi Pusu1,510Vikipedi1,495Lionel Messi1,491Recep Tayyip Erdoğan1,478Muhammed1,405Alex de Souza1,387Arda Turan1,346Süper Lig1,336
Part III
What we catalogued
The words that surge in newly-created article titles show the frontier of recorded
knowledge advancing — and reveal that the encyclopedia grew less by human curation than by
machine cataloguing.
06 · the documentation frontier
The frontier was mostly machines
Villages, then star catalogues and tick taxonomy, then every US county, then — overwhelmingly — plants, insects and spiders. The 78,000 articles created in 2024 are mostly species stubs. The growth is automated cataloguing, not idea-curation.
New articles per year. The surges are import campaigns — the settlement wave, the taxonomy waves — not bursts of human writing.
08 · the dictionary that exploded
And next door, a dictionary tripled overnight
The sister project Turkish Wiktionary shows the same force at its most extreme. One bot (CavlakBOT) created roughly 984,000 entries in 2025–26 — 73% of the entire dictionary in eighteen months. And most of it isn't Turkish: the flood is Greek, French and Italian word-lists. The “Turkish dictionary” is now only 24% Turkish words — a reference work mostly not about its own language.
Part IV
Who writes it
And who does all this? The revision log names every actor. The answer is a tiny human core,
a rising tide of bots — and a three-year hole punched straight through the community by the state.
09 · hands on the encyclopedia
A few humans, ever more machines
The mix of who edits, over time. Anonymous (IP) editing decayed from 39% (2008) to 3% (2025); bots surged through the 2010s and again into the 2020s — by 2024 the majority of edits are bots (the species-stub waves). Human editing is a thinning slice of a growing whole.
10 · the block crater
When the state switched the lights off
From April 2017 to January 2020 Turkey blocked Wikipedia entirely. Monthly human editors collapse from ~2,600 to ~650 overnight and stay flat for 33 months — the dedicated few who kept editing over VPN. In that period just 50 people made 91% of all human edits (vs 64% before, 50% after): censorship didn't only shrink the community, it concentrated it. The unblock in January 2020 is a vertical cliff — a 6× jump in one month.
11 · the night shift
A nation that edits at midnight
And when do they write? Human editing is a late-night act — it bottoms out at 06:00 and peaks at 23:00, an 8.8× swing; bots run almost flat around the clock. Tellingly, late-night's share of human edits rose from 24% to 40% during the block years (circumvention editing, done furtively at home) and fell back after. The anonymous IP editors behind so much of this are a paradox: their edits are reverted 6.7× more than registered ones, yet the village articles they overwhelmingly write are the least-reverted on the whole wiki — the main vandalism stream and the irreplaceable local-knowledge source at once.
Part V
How it knows
Finally, the epistemics. Comparing which words distinguish the encyclopedia from the news
exposes two different ways of holding a fact.
12 · assertion vs attribution
“It is” versus “someone said”
The words most over-represented in news are attribution and event verbs — dedi, belirtildi, açıklamada (‘said’, ‘was stated’) — plus security and politics. The words most over-represented in Wikipedia are the timeless -dir copula — mevcuttur, bulunmaktadır, sahiptir (‘it is’, ‘it exists’). Journalism sources its claims; the encyclopedia asserts them. The epistemic split, measured.
Source: the complete Turkish Wikipedia revision history (2002–2026, 33.5M
revisions) and an independent corpus of ~277k Turkish news articles (2008–2019). Chronicles use
year-over-year surge in document frequency (news, article titles) corrected against all prior
years; reverts are detected as restorations of a prior page SHA-1; the register split is the
log-ratio of per-corpus document share. These are the project's robust signals —
metadata and clean news prose — distinct from the wiki text co-occurrence on the companion page.
Editorial year-labels summarise the surging terms.