Appearance
Transcribe medieval marginalia when the notes carry evidence your research question actually needs — reader glosses, ownership and provenance notes, scribal corrections, nota marks against specific passages, or annotations that reveal how a text was used. Skip full transcription when the marks are decorative, illegible without imaging you do not have, or so voluminous and peripheral that cataloguing them by position and hand serves you better. The decision is rarely all-or-nothing: most projects transcribe a defined subset in full and describe the rest.
What counts as marginalia worth transcribing?
Not all marks in a margin are equal. A practical hierarchy, from "almost always transcribe" to "almost never":
- Interpretive glosses and commentary — explanations, cross-references, theological or legal notes. High evidential value.
- Ownership and provenance notes — ex libris, prices, donation records. Often the single most valuable line on the page for collection history.
- Corrections and corrigenda — letters, words or lines added or struck through; essential if you are editing the text.
- Reader response and manicula — pointing hands,
NB, nota marks, underlining. Transcribe the words; record the symbols by position. - Pen-trials, scribbles, doodles, decorative flourishes — usually catalogue, do not transcribe verbatim.
If a note changes how you read the main text, transcribe it. If it only tells you that someone read the page, a positional note is enough.
When is full transcription not worth the cost?
Transcription is expensive: a dense glossed folio can take longer than the main text column it surrounds. Decline full transcription when:
- The marginalia are purely decorative or are repetitive pen-trials with no semantic content.
- They are illegible at your current image resolution and you have no access to raking-light or multispectral capture.
- The corpus is large and the margins peripheral to your question — for 4,000 folios you may sample, or transcribe only ownership notes, and describe the rest.
- Your publication model cannot display them usefully, so the effort would never surface to readers.
In these cases, record hand, language, position and a one-line résumé instead. You lose verbatim text but keep the analytically useful metadata at a fraction of the cost.
How should I anchor a marginal note to the main text?
Keep marginalia logically separate from the reading line but explicitly linked to the passage they comment on. In TEI P5 the usual pattern is a note (or add for additions) anchored at the relevant point:
xml
<p>...quod omnis homo<note type="gloss" place="margin-left"
hand="#hand2" xml:id="n12">id est mortalis</note> moritur...</p>For notes that comment on a span rather than a point, anchor with anchor/span so the note can target a range without breaking the base text:
xml
<anchor xml:id="a1"/>...omnis homo moritur<anchor xml:id="a2"/>
<note target="#a1" targetEnd="#a2" place="margin-bottom"
hand="#hand2">Augustinus contra hoc</note>This separation lets a publication toggle marginalia on or off, and lets you query "all notes in hand2" without disturbing the edited text.
How do I record position and hand consistently?
Adopt a controlled vocabulary before you start, not midway. A minimal scheme:
| Field | Controlled values | Example |
|---|---|---|
| Folio | <n>r / <n>v | f.34r |
| Position | head, foot, inner, outer, interlinear, around-text | outer margin |
| Hand | hand1, hand2, ... | hand2 |
| Type | gloss, ownership, correction, nota, pen-trial | gloss |
| Status | complete, partial, illegible | partial |
Describe each hand once in a handDesc, then reference it everywhere. Mixing free-text descriptions ("a later hand", "the second annotator") across 200 notes makes the data un-queryable and breaks any later attempt at dating by hand.
What about marginalia you cannot fully read?
Capture the legible portion and mark the rest honestly. Use a documented gap convention — square brackets for editorial uncertainty in plain transcription, or gap with reason and extent in TEI:
xml
<note place="margin-right" hand="#hand3">pretium
<gap reason="illegible" extent="2" unit="words"/> solidorum</note>Then flag the folio for imaging. Raking light recovers dry-point and faded ink; multispectral or UV can recover palimpsested or washed annotations. Discarding an unreadable note loses the location evidence permanently; a stub note preserves it.
Should you let HTR handle the margins?
Generally no, not without preparation. Handwritten Text Recognition pipelines are tuned to the main text block; marginalia are smaller, cramped, frequently in a second hand, and live in regions standard layout analysis ignores. For a few manuscripts, transcribe margins by hand. For a large annotated corpus, treat the margin as a separate text region in layout analysis and, if volume justifies it, train a dedicated model on marginal ground truth — otherwise the recognition error rate on margins will dwarf that of the body text.
Key Takeaways
- Transcribe marginalia in full when they carry interpretive, provenance or correction evidence; catalogue (do not transcribe) decoration and pen-trials.
- Decline full transcription when notes are illegible without imaging you lack, or peripheral to the research question in a large corpus.
- Anchor notes to the passage they comment on with TEI
note/addoranchor/span, keeping them logically separate from the reading line. - Fix a controlled vocabulary for position, hand, type and status before transcribing, and describe each hand exactly once.
- Record partial readings with a documented gap convention and flag the folio for raking-light or multispectral imaging.
- Do not rely on out-of-the-box HTR for margins; treat them as a separate region or train a dedicated model.
- Most real projects transcribe a defined subset in full and describe the remainder — that is a feature of good scoping, not a compromise.
Frequently Asked Questions
Should I transcribe every annotation in the margins?
No. Transcribe in full only the marginalia that carry interpretive weight for your research question — glosses, ownership notes, corrections, reader responses. Pen-trials, smudges and decorative flourishes are usually better catalogued by location and hand than transcribed verbatim.
How do I record where a marginal note sits on the page?
Use a controlled position vocabulary keyed to folio side and margin: for example f.34r, outer margin, lines 12-15. In TEI you can attach a place attribute to a note element and anchor it to the relevant text with an anchor or a span pointer.
Should marginalia be transcribed separately from the main text?
Keep them logically distinct but linked. Encode marginal notes as note or add elements anchored to the point in the base text they comment on, rather than splicing them into the reading line, so you can suppress or display them independently.
How do I handle marginalia in a different hand from the main scribe?
Record the hand explicitly. Assign each distinct hand an identifier (hand1, hand2), note it on every transcribed annotation, and describe the hands once in your manuscript description so dating and attribution stay consistent across the document.
Is it worth transcribing marginalia that I cannot fully read?
Yes, partially. Capture what is legible, mark gaps with a documented convention such as square brackets or a gap element with a reason and estimated extent, and photograph the passage under raking or multispectral light for later recovery rather than discarding it.
Does HTR work on marginalia?
Rarely out of the box. Marginalia are small, cramped, often in a second hand and sit outside the main text block, so generic models and layout analysis miss them. Treat margins as a separate region and, for large corpora, train a dedicated model or transcribe by hand.