The Romanian language online · 1999–2026
Romanian diacritics online, 1999–2026: who wrote without ă â î ș ț, and when the marks came back
At first there were two Romanian internets, side by side. Between 1999 and 2007, 43.4% of the kept homepages (76 pages from 12 sites) carried at least 5 marked letters per 100, that is, were already written with the marks — with a cedilla, from the old character sets — while 47.4% carried almost no mark at all (0.5 per 100 or fewer). Only in 2009 does the middle page pass half of the edited-text level and stay there; between 2022 and 2026, 90.9% of pages clear the 5-per-100 mark and none is bare. The correct comma below ș and ț moves ahead on most sites in 2016; counted letter by letter across the panel, the cedilla holds on until 2022–2024, mostly through Adevărul's long homepage.
In short. There was no single year when every site lost the marks and then found them again together. Each site had its own year, and where the archive keeps consecutive years the switch happens from one year to the next: Mediafax was writing in full as early as 2001, Ziarul Financiar had the marks, lost them between 2006 and 2009 and regained them in 2010, Libertatea switched in 2009, HotNews stayed almost bare until 2018 and switched in 2019, and Ziare.com only in 2021. On the three sites present in almost every year, the halfway point also falls in 2009, and the comma also moves ahead in 2016 (Capital and Libertatea; Ziarul Financiar stays with the cedilla). The years before 2008 rest on few sites (at most 7 in any year) and on what the archive kept.
Workshop-ul internațional „Recent Developments in Analysis and PDEs (APDE 2026)”
uaic.ro, 13 July 2026 · 5.8 marked letters per 100 letters across the page
The kiosk of years
Move the year on the kiosk, or play the years: each headline comes from a real archived homepage, chosen near the ordinary page of that year. Where the archive keeps the bare letter, the mark that should have stood on it appears faintly, in red. By the late years, almost no mark needs the pencil.
The curve of the return
Each point is the median of the homepages kept from that year: how many letters in 100 carry a Romanian mark (ă, â, î, ș or ț, with either mark below). Edited Romanian text carries about 7 per 100; between 6 and 8, depending on the kind of text. The dotted line shows the share of pages declaring UTF-8. The wide band of the early years is the sign of the two internets: complete pages and bare pages stand in the same year, so the median jumps from year to year with whichever pages the archive kept. The crossing years printed here are therefore the sustained ones — the first year after which the median does not fall back — and the 90% line against the 7-per-100 level is never reached: today's plateau sits at about 6.0 marks per 100 letters.
Inside the return sits a second, later change. For years the mark under s and t was the Turkish cedilla (ş, ţ), because the older character sets had no comma. On most sites in the panel the comma (ș, ț) moves ahead in 2016: that year 6 of 10 sites write it more often than the cedilla. Counted letter by letter across the whole panel, though, comma letters move ahead only in 2022 and stay ahead from 2024, because Adevărul's homepage, the longest in the panel, keeps the cedilla until 2022, and Ziarul Financiar still keeps it in 2026. The two marks look almost identical on screen; the difference lives in the letter's code and, at times, in searches and in copying between programs.
Your text, set against each year
Paste a sentence or a paragraph. The page counts, on the spot, how many marked letters you have per 100 letters and names the year whose ordinary homepage resembled it most closely. Your text stays in the browser: a shared link carries only the year and a density band, never your words.
Which mark your keyboard makes
Type ș and ț in the box below. Some keyboards and older layouts produce the cedilla instead of the comma, and the eye can barely tell them apart. The page reads the codes of the letters you type and says which of the two you have.
șț
No ș or ț yet. Type the two letters in the box.
Comma or cedilla
The comma (ș, ț) hangs apart from the letter; the cedilla (ş, ţ) grows out of it, like a small tail. In many fonts the difference barely shows at reading size, which is why the check beside it reads codes, not shapes.
ș ş
Every year, in numbers
The panel asked stays the same every year: 18 domains — news sites, a portal, institutions and universities. Of those, 15 left usable pages in the archive; for 1998 none survives, so the series starts in 1999. A missing year means the archive keeps no usable panel homepage for that year. The cedilla column appears only when a year gathers at least 20 ș and ț letters; below that threshold the percentage would be noise.
| Year | Pages | Marks per 100 letters (median) | Against edited text | Cedilla of all ș+ț | Sites with the comma ahead | Declared UTF-8 |
|---|---|---|---|---|---|---|
| 1998 | no usable copy | |||||
| 1999 | 1 | 0.00 | 0.0% | — | — | 0.0% |
| 2000 | 2 | 3.25 | 46.4% | 100.0% | 0 / 1 | 0.0% |
| 2001 | 6 | 4.17 | 59.6% | 100.0% | 0 / 1 | 0.0% |
| 2002 | 5 | 0.00 | 0.0% | 100.0% | 0 / 1 | 0.0% |
| 2003 | 13 | 1.05 | 15.1% | 100.0% | 0 / 3 | 15.4% |
| 2004 | 11 | 5.66 | 80.9% | 100.0% | 0 / 4 | 0.0% |
| 2005 | 13 | 5.08 | 72.5% | 100.0% | 0 / 3 | 15.4% |
| 2006 | 13 | 0.00 | 0.0% | 100.0% | 0 / 3 | 23.1% |
| 2007 | 12 | 0.08 | 1.2% | 99.1% | 0 / 3 | 25.0% |
| 2008 | 12 | 0.84 | 11.9% | 99.7% | 0 / 3 | 83.3% |
| 2009 | 11 | 5.99 | 85.5% | 99.4% | 0 / 6 | 90.9% |
| 2010 | 12 | 5.62 | 80.2% | 99.8% | 0 / 8 | 100.0% |
| 2011 | 13 | 6.17 | 88.2% | 93.9% | 0 / 8 | 84.6% |
| 2012 | 6 | 5.63 | 80.5% | 98.7% | 0 / 4 | 83.3% |
| 2013 | 12 | 5.61 | 80.1% | 92.1% | 0 / 8 | 100.0% |
| 2014 | 16 | 5.82 | 83.2% | 89.1% | 0 / 8 | 100.0% |
| 2015 | 20 | 5.61 | 80.2% | 89.5% | 2 / 9 | 100.0% |
| 2016 | 20 | 5.93 | 84.7% | 79.9% | 6 / 10 | 100.0% |
| 2017 | 14 | 5.87 | 83.8% | 66.6% | 7 / 9 | 100.0% |
| 2018 | 23 | 5.79 | 82.8% | 68.5% | 7 / 11 | 100.0% |
| 2019 | 14 | 5.91 | 84.4% | 51.6% | 7 / 9 | 100.0% |
| 2020 | 23 | 5.88 | 84.0% | 54.1% | 8 / 11 | 100.0% |
| 2021 | 20 | 6.19 | 88.4% | 58.4% | 8 / 11 | 100.0% |
| 2022 | 15 | 6.03 | 86.2% | 39.7% | 7 / 9 | 100.0% |
| 2023 | 13 | 6.02 | 86.1% | 51.8% | 7 / 9 | 100.0% |
| 2024 | 18 | 5.88 | 84.1% | 44.5% | 8 / 10 | 100.0% |
| 2025 | 18 | 6.12 | 87.5% | 44.8% | 6 / 8 | 100.0% |
| 2026 | 13 | 5.88 | 84.0% | 15.9% | 9 / 10 | 100.0% |
Method and limits
The copies come from the Wayback Machine (Internet Archive): for each of the 18 domains asked and each year, up to three homepage captures near 1 March, 1 July and 1 November, taken in raw form. 369 pages with at least 800 letters of visible text remain; menus and short lines leave the count, and older encodings (ISO-8859-2, ISO-8859-16, Windows-1250) are read by each page's own declaration. Pages that, once read, do not come out as clean Romanian (under 90% plausible Romanian words; every correctly read page clears 91%) leave the series: 16 pages, mostly UTF-8 with one broken byte read as Windows-1250, and Windows-1250 declared as Latin-1. The 5 and 0.5 per 100 cutoffs move the early shares little: at 4 and 1 they read 48.7% and 47.4%; a cutoff of 6 would count many of today's pages as incomplete, since the plateau sits right at 6. The comparison level, 7 marked letters per 100, is the middle of the usual range for edited text; at 6 and at 8 the crossing years move by a few years at most, and the shape of the curve stands. The archive favours popular pages and recent years, and a homepage stands for its whole site only approximately: the measure is about the panel's homepages, not about the entire Romanian internet.
Sources: Wayback Machine, Internet Archive (raw homepage copies, 1999–2026). The panel: hotnews.ro, adevarul.ro, libertatea.ro, jurnalul.ro, evenimentul.ro, zf.ro, capital.ro, mediafax.ro, gsp.ro, ziare.com, gov.ro, mae.ro, edu.ro, bnr.ro, insse.ro, unibuc.ro, ubbcluj.ro, uaic.ro.
The kiosk headlines are reproduced from the archived copies, mistakes and all; the red marks show the edited form of the same words.