The most quoted line in French poetry ends on a small, ordinary word: antan. Its literal sense is last year, the year just gone. In 1870 Rossetti used yester-year to carry it into English. The word was already in print: George H. Calvert used yesteryear in 1846.
Villon is counting the dead — not soldiers or kings, but les dames du temps jadis, the beautiful women of bygone time. Flora, Thaïs, Echo; Heloise, for whom Abelard was gelded and made a monk; the queen who had Buridan thrown in a sack into the Seine; Jehanne la bonne Lorraine, Joan of Arc, whom the English burned at Rouen. Where are they now? Each stanza asks, and each stanza answers with the same line, the one everyone remembers even when they have forgotten the poem: Mais où sont les neiges d'antan?
The line is built to be unanswerable, and it does it with the plainest possible image. Not the snows of ages; the snows of last year — the snow that fell one winter ago and is, of course, already gone, already water, already run to the sea. Villon does not reach for the deep past to make you feel loss. He reaches for the nearest past there is. Antan is a homely, almost rustic word — it is what a farmer says — and that is the whole devastation: the great queens of history are as recoverable as the slush from a single thaw.
Antan comes, worn smooth by a thousand years of mouths, from the Latin ante annum — ante, before, + annum, the year: the year before this one. Pull it apart.
In Villon's fifteenth-century French antan still
carried its literal force — the previous year — though by his day it was
already drifting toward the general bygone it means in the rare modern
survivals. Etymology per the Trésor de la Langue Française and Wiktionary
(s.v. antan); both derive it from ante annum.
That precision is the problem. English has both last year and yesteryear; the latter can also mean time gone by. The choice is one of phrasing and register. A compact word can toll at the end of every stanza. Yore is vague and grand; last year's is two words and a possessive, with a different music. Payne's refrain shows that the longer phrasing can work too.
In 1870, in the volume simply called Poems, Dante Gabriel Rossetti printed his version, The Ballad of Dead Ladies, and at the foot of every stanza he set yester-year. He printed it with a hyphen:
The word is a small fusion: yester- (as in yester day, from Old English ġeostran, “the day before”) + year, echoing the bones of antan. Merriam-Webster gives 1870 as the first known use in the sense “last year”; etymonline calls it Rossetti's coinage for this refrain. But Calvert's Scenes and Thoughts in Europe (1846), p. 5, already prints yesteryear, and Wiktionary records that use. Rossetti printed it hyphenated, yester-year; the closed form predates his use. Here it carries one French word across.
A translation can enrich a reading without inventing a word. Rossetti's choice gives the refrain a different music. The needle Villon threaded left a thread in English.
And yet the word that saved the line also enlarged it.
Here is the critical reading. Taken literally, Villon's antan is small and near: last year, this past winter, a thaw you could remember. In Rossetti's yesteryear we can hear the grand and the far, the romantic, irrecoverable past. The phrasing lifts the poem up a register: Villon's nearest snow becomes the snow of all lost time. That is a reading of this refrain, not a date for a change in the word's meaning. Merriam-Webster still gives both last year and time gone by. The selected translations make the difference of register audible.
These two selected Victorian translations of this ballade, both long out of copyright,
have their refrains checked against embedded transcriptions in
research/villon-antan/verify.mjs. They make the split visible in its purest
form. They sit eight years apart and choose opposite walls.
Rossetti reaches further (yester-year) and so
lifts the line into the romantic past; Payne keeps it
literal (last year's snow) and so keeps the homely, near thing Villon
actually wrote. Above them sits the source itself. Tap a chip to light one strategy;
tap a card to read the hand.
“Strategy” classes only the treatment of antan in
the refrain, not the merit of the whole translation. Each card's full source —
volume, year, page or scan — is listed under Sources.
Read the literal hand and you hear what Villon wrote: last year's snow, blunt and near. Read the romantic one and you hear a wider resonance: the snows of yesteryear, the whole vanished past in one bell-note. Neither is wrong. The poem is a question about where the lost go; Rossetti's phrasing can make the loss larger. That enlargement is the translation, visible, here, as the gap between the two selected hands. Later translators widen the same fork: Richard Wilbur keeps it literal (last year's snow); Ken Knabb generalises (years gone by). Those versions are still in copyright. The gallery holds two selected public-domain translations, not every available rendering.
There is a scholarly argument that makes Villon's last year sharper still. The Dutch Romanist Paul Verhuyck proposed that the neiges d'antan may also recall something real: the carved snow statues of the Low Countries and northern France. His proposal supplements the metaphor of transience; he explicitly keeps that reading. He documents medieval snow figures at Tournai and Arras, with later continuations at Lille and Antwerp: kings, saints, and famous women sculpted in the streets. The winter of 1457–58 was one such freeze, when the Seine itself iced over. On that reading the refrain is colder and more exact than the romantic English lets on: last year's snow because the snow-ladies, like the real ladies the stanzas name, melt within the season — a double death, the women and their snow images both gone by spring. It is a minority reading and not the only one, but it turns on the same fact every literal translator kept and every romantic one softened: antan means last year, and last year's snow is the most certain absence there is.
The refrain strings are checked against embedded transcriptions by
research/villon-antan/verify.mjs. It also checks the displayed strategy
labels and the presence of the etymological account and source citations.
The derivation of antan from ante annum / “last year”
is sourced below. The dictionaries' attribution of yesteryear to Rossetti
is set beside Calvert's earlier printing; their attribution does not establish
first invention. The offline check does not independently establish these facts.
Mais où sont les neiges d'antan ?
Villon's own manuscripts and the early prints vary in spelling and accent
(fifteenth-century French is pre-orthographic); the apparatus below records the
edition used and its punctuation difference from the banner.!. Modernised editions (e.g. the INRIA
text) set ?; the famous form quoted in the banner uses the question
mark. The wording les neiges d'antan is stable across all of them.node research/villon-antan/verify.mjs runs offline, no network, and
compares the refrains with embedded transcriptions, checks the displayed strategy
labels, and checks that the etymological account and citations are present. The live page is driven in a real browser by
verify-yesteryear.mjs (no console errors, the instruments respond, no
horizontal overflow at phone / tablet / desktop widths).
Written 2026-06-21. Claims re-read against their sources on 2026-10-14: 109 checked, 78 confirmed, 30 wrong, 0 unverifiable, 1 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/queue-1014c/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex.
This was the directed 2026-10-14 correction pass for the 32 items in this layer's queue-1014c triage, starting from c8289bd340. The fixer re-read each item's evidence before editing, retained X28 as literary criticism, and carried forward the 77 CONFIRMED claims that the N2 checker re-read in its pinned f5e5f84f34 audit; their notes are explicitly attributed to that checker rather than presented as 77 new fixer checks. The counts include 22 WRONG and 8 MINOR lines in wrong, plus the STALE metadata field. The seven public corrections cover only WRONG findings; the MINOR fixes are recorded here, and the dateModified exception has no separate public line. The repair review checked all three fix-checker problems and all seven mirrored dated corrections. It corrected the remaining CSS comment and retained the correction texts. Ten item dispositions below remain open for surviving copies in other layers, grouped into the two NOTE NEEDED paragraphs in research/assay-line/queue-1014c/the-snows-of-yesteryear/repair-report.md. The claim counts are unchanged: 78 CONFIRMED + 22 WRONG + 8 MINOR + 1 STALE = 109.
This record is long (109 claim lines and 31 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.
The assay office: what a record is, and every layer re-read so far.