The Romance phase: Latin's daughters take a desk
25 July 2026 · Nabu news
The library’s Latin has always been classical. This phase follows it forward — through the late-antique transition band, the medieval chronicles and charters, into Old French — and gives the continuum a desk of its own: romance, the twenty-first, beside the classical desk it grows out of. The Romanist — the Latin-to-vernacular continuum, charter Latin to Roland to the troubadours.
Four shelves opened it in one day. openMGH brings the critical
edition backbone of medieval Latin — the Monumenta Germaniae
Historica in TEI, CC BY 4.0, texts themselves free of copyright — with
a first wave of the complete SS rerum Germanicarum series: Einhard,
the Frankish annals, Widukind, Adam of Bremen, fifty-seven volumes.
digilibLT covers the approach: 370 late-antique secular prose
texts, the second through seventh centuries, machine-lemmatized and
honestly labelled [silver]. BFM, the Base de français médiéval,
carries the vernacular side — 6.4 million words of Old and Middle
French under France’s open Etalab license, from the 842 Serments de
Strasbourg onward. And the UD Romance treebank pack adds the gold
annotation: Tuscan charters of the 770s, Dante’s Latin, Old French,
Old Romanian — four new gold-lemma languages in one config change.
CroALa’s Croatian Latin, last phase’s arrival, dual-tags onto the desk
as the continuum’s Latin side.
The phase’s riders sharpened two instruments. The meter layer
grew its maps from evidence against the live catalog — 82 Latin works
crosswalked and 79 of 88 Greek scansion files matched, which is how a
mismapped seed entry (Lucretius pointed at Catullus) was caught and
its false rows purged — and the layer doubled to 167,732 scanned
lines, with a first query surface: nabu search --meter H filters
hits to hexameters, honestly footnoted with where the scansions come
from. Iliad 1.1 now carries its meter line. And the Pleiades
gazetteer moved into a derived SQLite index: a place lookup that
cost five seconds and four gigabytes of memory now answers in
milliseconds, byte-identical, rebuilt from canonical bytes like
everything else.
The library stands at 813,257 documents / 65.1 million passages in 115 language codes, 89 sources all wired, gold lemmas in thirty-two languages. Local, license-honest, rebuildable — now from Sumer to the troubadours’ doorstep.