← Zaman
Zaman · the lexical history

The lexical history

Twenty-five years of Turkish writing, read word by word for what it reveals about the language, the culture, and the people a society chooses to remember — from nearly a billion words of dated text. Part of Zaman; see also connections & meaning, which reads the same corpus pair by pair.

947Mwords analysed
611kdistinct words (2025)
2004–2026year by year
13 of 18pairs went native
Part I

The Turkish language, reshaping

A century after Atatürk's reform set out to purge Arabic and Persian from Turkish, you can still watch the lexicon move — word by word, year by year.

01 · the reform, in one line

Turkish is still shedding its loanwords

Averaged across competing native-vs-loanword pairs, the native (öztürkçe) word's share keeps climbing: 13 of 18 pairs moved toward the native form over 25 years. Above the dashed line, the native word is the majority.

2026-06-26T23:06:17.375848 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/
02 · which words win

…but not on every word

Pair by pair the native word usually wins — yanıt overtook Arabic cevap for ‘answer’ (26%→52%), imkan all but surrendered to native olanak. But it's no clean sweep: a few entrenched loanwords resist or win back — Persian şehir holds off native kent, Arabic kelime beats sözcük, and for ‘political’ the Arabic siyasi beats the reform's own coinage siyasal.

2026-06-26T23:06:17.498606 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/
03 · the proof

Not a Wikipedia quirk — the news agrees

Does this reflect Turkish, or just Wikipedia's house-style? We checked against a completely independent genre: 277,000 Turkish news articles (2009–19). Across the same 12 pairs the native form gains in the news too (56%→61%), tracking Wikipedia's rise. Journalism is more loanword-conservative (a lower baseline) but moves the same way — so this is the language changing, validated across two genres.

2026-06-26T23:07:02.014104 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/
04 · new words, new world

What arrived in Turkish, and what faded

New vocabulary lands with world events. Indexed to each word's own peak: koronavirüs appears from nothing in 2020, pandemi and karantina with it; bitcoin from 2012, tiktok from 2023. And internet peaks around 2011 then fades — no longer worth saying once it's everywhere.

2026-06-26T23:07:02.971436 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/
Part II

Who, and what, Turks write about

Strip the grammar away and the culture shows through — who a society thinks worth recording, and where its attention points. Measured by how many distinct texts use each kind of word.

05 · a culture of people

More and more, it's about individual lives

Turkish writing turns biographical: life-event words climb steadily — doğdu (‘born’) 151→259 per 1,000 texts, öldü (‘died’) 140→239, evlendi (‘married’) 99→175. By the 2020s roughly a quarter of all writing is the story of a single person's life.

2026-06-26T23:07:03.706131 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/
06 · strikers over scientists

A nation that writes about its footballers and stars

And which lives? Entertainers and athletes. Film, music and TV words rise steadily; football vocabulary tripled into the mid-2010s — while hard science (molecules, enzymes, quanta) stays a flat sliver the whole 18 years, never growing. A culture's attention in one chart: the pitch and the screen, not the lab.

2026-06-26T23:07:04.543997 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/
07 · women, slowly

Women edge into the record

In 2008 kadın (woman) and erkek (man) are named about equally — but kadın pulls ahead (171 vs 148 per 1,000 by 2026), its share rising 50%→54%, while aktris (actress) more than doubles. The gap narrows — though fathers (babası) still outnumber mothers (annesi) when a life is told.

2026-06-26T23:07:05.296990 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/
08 · from Turkey to the world

A widening gaze

Early on, Turkish writing is about Turkey. But its mentions of the wider world climb steadily and overtake home around 2011 — by the 2020s a Turkish text is likelier to name a foreign country than its own. The horizon widened from the nation to the globe.

2026-06-26T23:07:06.000715 image/svg+xml Matplotlib v3.11.0, https://matplotlib.org/

Source: the complete Turkish Wikipedia revision history (2002–2026), tokenised locally into a ~947M-word diachronic corpus (year-end text of every article, every year). The culture findings use a prose subset (articles > 2,000 chars) and document frequency (distinct articles using a word) to resist the bot mass-imports that dominate raw counts. The language-reform finding is cross-genre validated against 277k independent Turkish news articles (2009–19), where the same native-word gain appears. Honest caveats: word-pairs are hand-chosen and senses overlap; topic buckets carry some polysemy, so trust the trends over the exact levels.