Paleography

about

Caroline minuscule — a reader's primer

Track 1 of the MVP. Written for someone who reads Latin and has never read a manuscript. Frequencies are real, counted over 12,314 lines of ground truth from five 9th-century witnesses (see corpus/sources.yml). Nothing here is invented for illustration.

---

1. Why this script first

Caroline minuscule (c. 780 – c. 1150) is the easiest medieval Latin bookhand and the reason your eye already half-knows it: the humanists of the 15th century mistook it for Roman writing, copied it, and it became the model for roman type. You are, in a real sense, already reading a descendant of it right now.

What that buys the beginner: clearly separated letters, generous word spacing, minimal ligature, one letter per sound. Almost everything that makes Gothic and cursive hard is absent. What remains hard is the abbreviation system — and that is where the actual work of this track lies.

2. What will actually trip you

Not the letterforms. Four things, in order of how often they bite:

(a) a, and the reason you will misread it. Caroline a is usually a two-storey form close to modern printed a, but the open "cc"-form a (looking like two c's side by side) is common in early and Insular-influenced hands. Beside a u, that gives you au looking like cui. This is the single most common misreading for beginners.

(b) Long ſ. Word-initial and medial s is written tall, like an f without the crossbar — or with a half-bar, which makes ſ/f genuinely ambiguous. Round s appears at word-end. ſi next to fi is the classic trap. (Rescribe's GT modernizes this to plain s; Eutyches' preserves it. Same script, different transcription layer — this is why sources.yml declares the layer.)

(c) e caudata — ę. 2,251 occurrences in our corpus. The little tail (ogonek) is the medieval way of writing ae. ecclesię = ecclesiae. It is not a spelling error and it is not e.

(d) The abbreviations. See §4. They are the real curriculum.

3. What the page looks like

Real lines from Wien ÖNB Cod. 940 (Salzburg/Saint-Amand, s. VIII ex.), an Irish commentary on Matthew:

    pinguitudo · HVCVSQVE ORDO NVNC ACCEDIT ID · EST ·
    infernus ‧ et sanctis fidelibus praeparatur regnum ut est illud ‧

Three things to notice, all of them typical:

• Punctuation is a dot, and its height carries meaning. The middle dot · (4,865x)

and the raised point ‧ (8,340x) mark pauses of different weight — the ancestors of comma and full stop. Do not read them as periods.

• Display capitals interrupt the text. HVCVSQVE is in rustic capitals for

emphasis; scribes switch script grade mid-line. V is u.

• No j, no w; u and v are one letter. uerbum, not verbum.

And line-ends break words without ceremony — our corpus has 1,790 line-break hyphens (¬), often mid-syllable.

4. The abbreviation system — the actual work

Medieval scribes abbreviated constantly, to save parchment. The system is not random; it has four mechanisms, and once you know them most abbreviations are readable on sight.

(i) Suspension by a stroke — the nasal bar

A horizontal stroke over a vowel means a following m or n is omitted. This is the commonest abbreviation in Latin manuscripts and our counts show it:

⚠ The same stroke is also a general "something is missing here" mark, not only a nasal. Context decides.

(ii) Special letterforms with a stroke through the descender

⚠⚠ Two corrections that only came out of reading Cappelli, and both were mine.

**ł is not simply vel.** Cappelli gives it as vel, ul…, …el — crossing the ascender of an l in the middle of a word it can just supply those letters. In a grammar book, forever saying a form may be this or that, vel remains much the likeliest; but "ł = vel" as a flat rule is wrong.

ꝗ depends on the century, and ours is the awkward one. Cappelli's footnote: "Nel sec. VIII lo stesso segno vale que" — in the eighth century this same sign means que, and in Anglo-Saxon hands quam or quia. Our manuscripts are ninth-century, sitting directly on that boundary. Read the sentence before trusting the sign. This is the concrete case of Thompson's warning that a sign's meaning shifts with the period.

(iii) Superscript letters

A small letter written above the line supplies what is missing — usually with an r understood: qͣ = qua, qͥ = qui, IIͦIͬI = ordinal forms. Our corpus has combining a e i o u r t above the line, 700+ occurrences together.

(iv) Conventional signs standing for whole words

⭐ That last row is the lesson of this whole track, and Cappelli gives it a name. His section IV (p. XXIX) is headed segni abbreviativi con significato relativo — abbreviative signs with relative meaning, whose value is not fixed but changes with the letter they are attached to. Expansion is not a character map. The same sign means different things after different letters, and only reading Latin tells you which. This is exactly the fusion of vision and language prior that paleography trains — and the reason a learner who already knows Latin can be taught this quickly, while a model needs the language model bolted on.

⚠ What is NOT yet settled

Of the 64 characters occurring ≥20 times in our Latin corpus, 9 of the 16 taught here now have an expansion verified against a source page read as an image — the dataset's own character table, Thompson 1912, and **Cappelli's Lexicon Abbreviaturarum (1929), whose page for every sign is indexed in research/cappelli-index.md. 23 more carry expansions I proposed from standard convention — they need Wilson's ratification before they teach anyone** (corpus/latin-abbreviations.json, field expansion_proposed). 37 have no expansion recorded at all.

⛔ This caution is not decorative. Reading four private-use codepoints from context, I got three of four wrong — the contexts were suggestive and the reasoning was sound and it made no difference. See corpus/INGEST-NOTES.md §7.

5. Nomina sacra

Contractions of holy names, inherited from Greek practice, always with a stroke over:

DS = deus · DNS = dominus · IHS = Iesus · XPS = Christus (from Greek ΧΡΙΣΤΟΣ — the X and P are chi and rho, not Latin x and p) · SPS = spiritus · SCS = sanctus.

⚠ XPS is the one that catches people: it is a Greek word in Latin dress.

6. Dating a Caroline hand — rough orientation

Not an MVP exercise (Level 5), but useful to know the axis you are on:

• s. IX — round, wide, well-separated, few ligatures, a often open. Our Wien 940

and Eutyches witnesses sit here.

• s. X–XI — more compact, slight lateral compression begins.
• s. XII — "late Caroline": letters narrowing, feet acquiring serifs, curves

flattening. The road to Gothic.

• Then Gothic textualis (s. XIII), where adjacent curves fuse ("biting"), and the

beginner's difficulty jumps a level. That is a later track.

7. Practice order for this track

1. Script ID — tell Caroline from Gothic textualis and from Insular. (One session.)
2. Glyph cards — a open vs two-storey · ſ vs f · r vs n · ę vs e.
3. Expand — the four mechanisms of §4, drilled in frequency order: nasal bar first

(it is 8,800+ occurrences), then ꝑ/ꝓ, then word-signs.

1. Line — Wien 940 MainZone only. Filter out MarginTextZone or you will be

handed a four-word scrap of gloss (INGEST-NOTES §6).

1. Page — then, and only then, the glossed pages of Lat7499, where main text and

three layers of commentary share a folio.

Sources

All ground truth CC-BY 4.0; see corpus/sources.yml for shelfmarks, licences and URLs. Frequencies computed by tools/ingest.py over corpus/normalized/*.jsonl. Character identities from the chocomufin control table shipped with the Eutyches dataset (Clérice & Pinche), which is authoritative for that corpus.

Practise this hand: Latin — Caroline minuscule · Latin — a glossed grammar