How to read old cursive: a provenance and transcription worksheet

1. Record where the document came from, before you read a word

Fill in these fields while the record is in front of you or the image is open. They identify the source that the transcriber says was used and give a later reviewer a location to check.

Record at least the following, at the head of the same file that will hold the transcript:

repository

The archive, library, courthouse, parish, registry or website that holds the original, named as it names itself.

reference

Collection, series, volume, box, folder, page or folio, film or image number — every level of the citation the holder uses, copied character for character.

original or surrogate

Whether you are reading the original paper, a bound register copy, a microfilm frame, a digital scan, or a photograph you took yourself. Record which, so a later reader knows what you actually looked at.

image identity

The file name and, where one exists, the permanent URL or item identifier of the exact image you worked from.

access date

When you consulted or downloaded it. This records which version and access event your note refers to.

condition

Fading, bleed-through, water damage, tight binding, trimmed margins, later pencil annotations, stamps. Note this before transcribing so that later gaps have a recorded cause.

transcriber and date

Who produced the transcript and when. If somebody else revises it later, that revision gets its own line.

If permission and handling rules allow, capture the page and relevant neighbouring material such as adjacent entries, title pages and column headings. Keep an unedited copy of the capture you cite, and make crops, rotations or contrast changes on separate working copies.

2. Write down what you can establish, and mark the rest unknown

Three fields to fill in before decoding. Each one may legitimately be recorded as unknown; record the basis alongside the answer, so that a later reader can see whether it was evidence or assumption.

What language is this?

Record script identification and language identification as separate questions. List the visible clues you considered — recurring short forms, dates, names, opening or closing formulae and printed headings — and label the result as a candidate when those clues do not establish it.

Is there a date, and where does it come from?

Record explicit evidence only: a date within the text, a printed form number, a dated stamp, a filing endorsement on the reverse, a description supplied by the repository. Copy it exactly and note where on the document it appears. If you have no such evidence, record the document as undated rather than estimating one from the look of the script.

What kind of record is this, and what repeats in it?

Rather than assigning a genre from memory, read several neighbouring entries in the same hand and write down what actually repeats: which phrases recur word for word, which columns carry the same kind of value, which words appear in the same position in every entry. That list is evidence from your own volume, and it is what you will use in step five. Do it before attempting the entry you actually care about.

3. Transcribe exactly first; modernise later, in a separate file

This worksheet proposes two labelled documents: a literal transcript and, if useful, a separate normalised reading derived from it. Keeping separate files makes later edits easier to attribute; it does not establish that either reading is correct.

In the exact transcript, the rules are mechanical:

• Record source line breaks. Numbered source and transcript lines provide location references for later comparison. They do not by themselves prove completeness.
• Preserve spelling exactly, including variant spellings of the same name in the same document. Do not standardise a surname to the form you expect: record each spelling as written, with its line number.
• Preserve capitalisation and punctuation as written, including the absence of punctuation and capitals that fall mid-sentence.
• Leave abbreviations closed. Copy the abbreviation as written. Expansion belongs in the normalised version, or in square brackets if you are certain.
• Keep the layout information that carries meaning: column values in column order, marginal notes marked as marginal, interlined words marked as interlined, struck-through words recorded as struck rather than deleted.
• Do not reorder anything to make it read more smoothly, and do not silently correct an arithmetic error, an impossible date or a repeated word. Record it, then comment on it.

In a separately labelled normalised version you may expand abbreviations, regularise spelling, restate dates, join source lines or add notes. Keep explicit pointers to the literal transcript and provenance block so a reader can locate the stated source of each supplied form.

4. A four-mark notation for everything you cannot read

This worksheet defines four marks for four stated conditions. Use them consistently within this transcript and record why you selected one; other projects may use different notation.

• [unclear] — writing is present and you cannot resolve it, though it is physically legible in principle. Use this for hard handwriting, not for damage.
• [illegible] — writing is present but genuinely cannot be made out: ink loss, abrasion, heavy overwriting, a blot across the word. Add extent when you can, as in [illegible, 2 words].
• [gap] — no text is available at all, because the page is torn, trimmed, bound too tightly to open, obscured in the image, or the writer left the space blank. Say which: [gap: torn, approx. half a line] or [gap: blank in original].
• [?] — you have read the word but are not confident. Place it immediately after the reading, as Margaret[?], so the candidate survives while staying flagged. Use [unclear] instead when you have no candidate at all.

Three supporting conventions handle the rest: ^word^ for text interlined above the line, <word> for text the writer struck through but which you can still read, and [margin: ...] for anything written outside the main text block. If you expand an abbreviation inside the exact transcript, put only the supplied letters in brackets, as in Wm[?] left unexpanded or Sept[embe]r where you are sure.

The discipline that makes this work: a bracket is never removed by a later reader who merely feels more confident. It is removed when new evidence resolves it, and the note records what that evidence was. Margaret[?] becomes Margaret when the same entry number in the index reads Margaret — not when the name simply looks more like Margaret on the fourth reading.

5. Build a letterform key out of this document's own repetitions

A printed alphabet chart does not establish how the writer formed a particular mark. A specimen sheet made from candidate examples in this document gives you traceable comparisons, not proof of a letter identity.

Collect role hypotheses suggested by position

Using the list of repetitions you made in step two, mark words that may serve as a recurring column heading, a value in a date column, a phrase at the head of an entry, a signature at the foot of a page or a name in a running head. Position suggests a role but does not establish the letters. Record the positional clue, then test the candidate reading against visible letterforms and other occurrences.

Collect the repeated names

Find every occurrence of a name across the entries you have, including any index at the front or back of the volume and any printed form. Where one occurrence is clearer than another, use it as your reference and record which occurrence you used. Where two occurrences differ, record both spellings rather than choosing.

Write the sheet down with citations

Keep a plain list, letter by letter, with each candidate citing its source location, for example capital H — candidate from "Hampton", p.14 entry 3. Record the shapes observed and the number of examples. The count describes your observation set; decide separately whether it supports a reading.

One sheet per hand

A volume may contain several hands. Compare against a sheet only where you have established that the sheet and the page are in the same hand; when the ink, the slope or the signature at the foot changes, start a new sheet rather than carrying the old one across.

6. Shapes that stall a reading, and how to test candidates

This table lists shapes people get stuck on and gives you two or more candidate readings to test for each. It asserts nothing about what any period, place or writer used: a candidate is something to try against your own document, and it earns a place in your transcript only when your own evidence supports it. Where nothing decides between candidates, record them all in brackets.

For every row, compare the unfamiliar shape with cited instances elsewhere in the document when available. The comparison may support, contradict or leave open a candidate; retain brackets wherever the inspection does not resolve it.

7. Distinguishing a letterform from damage, by inspection

This worksheet assigns different labels to an unresolved letterform and to physical loss or obstruction. Record the inspection behind the label so a later reader can decide whether another image or the original might be useful.

Questions to answer about the mark

• Is the ink density the same as in the strokes around it, or thinner, browner, blurred, haloed or absent?
• Does the same shape appear elsewhere on the page or in the same hand, in comparable positions? How many times?
• Does it sit on the baseline and respect the line spacing of the surrounding text, or ignore both?
• Under magnification, can you follow a coherent pen direction through it?
• Does it affect one letter, or cross several letters, words or lines?
• Does it align with a fold, a tear, a stain edge, a binding shadow, a punch hole or a trimmed margin?
• Is it a mirror image or offset of writing you can find on the reverse of the leaf or on the facing page?

Two checks to run before deciding

• Change the light or the image. Raking light across the original, or raised contrast and greyscale on a copy of the scan, gives you a second view. Compare the two views and record what each showed.
• Look at the reverse of the leaf and the facing page. Do this before concluding anything about extra strokes, and note what you found there.

Write the answers next to the reading, then choose the notation from them. [illegible, 1 word: ink loss at fold] is a complete result: it records that there is writing, that you could not make it out, and why — which also tells a later reader that a different image or the original itself is the thing to try.

8. A worked example, start to finish

The short record below is invented for demonstration. There is no such volume, no such deposit and no such person; the names, the reference number and the wording were written for this guide to show the notation in use. Do not cite it, and do not treat any detail in it as a statement about how records from any particular place or period were actually worded.

repository : Northmoor County Records Office (invented)
reference  : Deeds Register, vol. 4, p. 112, entry 2 (invented)
surrogate  : digital scan, file NM-D4-0112.tif
accessed   : 7 August 2026
condition  : brown ink, faded at outer margin; fold crosses lines 5-6
language   : English; date and formula legible
hand       : Hand B (second hand in this volume, begins p. 108)
transcriber: L. H., 7 August 2026

01  Memorandum that on the twelfth day of
02  [unclear] in the yeare aforesaid Thomas
03  Ha[?]den of this parish husbandman did
04  acknowledge the sale of one close of
05  ground called Long [illegible, 1 word: fold]
06  conteyning fowre acres [gap: torn, approx. 3 words]
07  unto Wm Sadler of the same parish
08  ^and to his heires^ for the summe of
09  <fifteene> sixteene pounds paid in hand
10  [margin: entred and examined J.P.]

Read what the notation is doing line by line, remembering that the record itself is invented. Line 2 carries [unclear]: strokes are present but the transcriber has no candidate yet, so none is supplied. Line 3 uses [?] because a reading exists and is merely uncertain; the surname stays available for corroboration instead of being lost to a blank. Line 5 is [illegible] with its cause stated, so a later reader knows that a different surrogate or the original is the thing to try. Line 6 is a [gap] — nothing to read, because the paper is gone. Line 8 records an interlineation that the writer added, and line 9 preserves a struck word alongside its replacement, which in a financial clause is exactly the kind of detail that must never be silently tidied. Line 10 keeps the marginal endorsement out of the body text.

NORMALISED READING — derived from exact transcript
of Deeds Register vol. 4, p. 112, entry 2 (invented).
Spelling modernised; abbreviations expanded; supplied
words in brackets. See exact transcript for what is
actually on the page.

Memorandum that on the twelfth day of [month unread]
in the year aforesaid, Thomas Ha[?]den of this parish,
husbandman, acknowledged the sale of one close of
ground called Long [one word lost at fold], containing
four acres [approx. three words lost to a tear], to
William Sadler of the same parish and to his heirs,
for the sum of sixteen pounds paid in hand.
(Fifteen struck through and replaced by sixteen.)
Margin: entered and examined, J.P.

Unresolved: month (line 2); surname Ha[?]den (line 3);
field name (line 5); clause after "fowre acres" (line 6).

Notice what survived the modernisation and what did not. The spelling was regularised and Wm expanded, because that is what a normalised version is for. But the struck word, the interlineation and every unresolved reading are still declared — and the file says outright which document it derives from. A reader who only wants the sense stops at the first paragraph; a reader who needs the evidence knows precisely where to go and what remains open.

9. Check names and dates against records that do not derive from this one

List names, dates, ages, sums and place names separately for review. Sentence plausibility does not verify them, so record any comparison made against the image or another identified source.

Before treating another record as confirmation, establish whether it derives from the one you are reading. An index compiled from the same volume, or a published transcription of it, may simply repeat the reading you are testing. What you are looking for is a record created separately that names the same event or person — and where you cannot establish independence, record that too.

• Ask where the second source came from before you weigh it. Write down its provenance next to your note, exactly as you did for the first document.
• Test dates for internal consistency. Check an age against a stated year, a later event against an earlier one, an entry against the dates the volume covers. Where the result is impossible, re-read the numerals rather than amending the record.
• Do not assume a date format. Record the date exactly as written, note that its order is unestablished if you cannot establish it, and give an unambiguous form only in the normalised version, marked as supplied.
• Keep corroboration out of the exact transcript. When another record resolves a bracket, add a note — Ha[?]den read as Harden; compare 1851 census, household 12 — rather than editing the transcribed line.
• Record searches with no match too. A note such as "Searched the burial register 1840–1850, no matching entry" documents the scope and outcome of that search.

Capture rules, and experiments you can run on the image

The image is the object you will actually work from, so record how it was made and keep the original of it. The lists below are rules to follow and experiments to try; none of them states what the result will be on your document.

If you are photographing

• Shoot straight down, with the sensor parallel to the page.
• Fill the frame with the page, so the resolution goes to the ink rather than the table.
• Light the page evenly, and keep your own shadow and the device's shadow off it.
• Experiment: take a second capture with the light raking low across the surface, then compare the two images on the words you cannot read, and keep both.
• If you disable beautification, sharpening or document-scan filters, record that choice. If you use them, retain the unfiltered image because processing may alter visible edges.
• Capture the whole opening as well as the detail: the page in full, the facing page, the spine or title page, and the reverse where there is writing or show-through.
• Follow the repository's rules on photography, flash, handling and reproduction, and ask before setting anything up.

If you are scanning

• Scan at a high optical resolution and save a lossless master.
• Scan in colour, even for a document that looks monochrome, and convert copies to greyscale afterwards if you want to. You can convert a colour master later; you cannot recover colour from a greyscale one.
• Turn off automatic descreening, background removal and text enhancement, for the same reason as the photographic filters above.
• Do not force a tightly bound volume flat on a platen. Use an overhead capture, or ask staff.

Working on the image

Keep the master untouched and adjust copies. Treat contrast, greyscale, inversion, rotation and enlargement as experiments, apply one at a time, and record which view was used for a candidate reading. Compare it with the master because processing can alter the displayed edges.