Transcribing a Nineteenth-Century Accession Ledger: A Line-by-Line Guide

An accession ledger is more than a page of handwriting

A nineteenth-century accession ledger often combines several kinds of evidence in a compact space. One row may contain a date, donor name, object description, quantity, valuation and a reference to another document. Columns may be ruled by hand, entries may continue across multiple lines and later staff may have added notes in different ink. A transcription that treats the page as ordinary continuous prose can lose the relationship between these fields.

Before attempting to transcribe handwritten ledger material, establish the page structure. Record the repository reference, volume identifier, page number and image filename. Identify the columns and their order, then number the entries. If the original page contains ten rows, the working transcription should make it possible to return to each of those ten positions without relying on a search through a block of extracted text.

• Record the volume and page reference.
• Preserve the order of columns and rows.
• Distinguish original entries from later annotations.
• Retain a link between each transcription segment and its source image.

How handwritten text recognition approaches the page

Handwritten text recognition, commonly abbreviated as HTR, uses computational methods to identify and transcribe handwriting from images. Unlike conventional printed-text recognition, it must account for connected letters, variable letter shapes, individual writing habits and inconsistent spacing. Some systems process entire pages, while others use selected regions or line-level segmentation. Available functions and supported languages differ by provider and product.

The output may be a plain transcription, a structured representation of page regions or text associated with coordinates. Those distinctions matter in an accession ledger. A correct word attached to the wrong row can create an incorrect relationship between an object and its donor. Staff should establish whether the chosen tool preserves layout and whether its export can be connected to the original image before incorporating text into a catalog workflow.

Read each line as a structured record

A reliable transcription begins with the physical organization of the page. Consider a hypothetical entry with a date, a donor surname, an object description and a quantity. Transcribing the words without their column positions could turn the date into part of the description or attach the donor to the following object. Preserve the column labels and entry boundaries even when the handwriting recognition output is supplied as continuous text.

A practical working format assigns each entry an identifier and records each field separately. For example, entry 07 might contain '12 May 1889' in the date field, an uncertain surname in the donor field and 'two brass candlesticks' in the description field. The transcription should preserve the original wording. A later catalog record can normalize the date and object name according to institutional conventions.

• Transcribe the date without silently changing its format.
• Keep donor and recipient names in their original fields.
• Preserve original object descriptions and abbreviations.
• Mark empty fields as empty instead of supplying expected values.

Names, numbers and abbreviations need special attention

Personal names are among the most consequential transcription problems. A surname that resembles a familiar English word may be converted into that word, particularly when the surrounding sentence gives the system a misleading context. Place names and business names create similar difficulties. Historical handwriting recognition should therefore be treated cautiously when the resulting text contains names that will become searchable access points or donor references.

Numbers require their own review. A handwritten 3 can resemble an 8, and a poorly formed 1 can be confused with another character. In a ledger, a misread quantity or valuation can change the meaning of an entry. Abbreviations should be preserved in the transcription unless the project explicitly defines an expansion policy. An expansion should be marked as editorial intervention, especially when several interpretations are possible.

• Compare every personal and place name with the image.
• Review dates, quantities and monetary figures independently.
• Preserve unusual spellings in the transcription.
• Mark uncertain abbreviations without guessing their expansion.

Use a consistent notation for damaged and uncertain text

Transcription conventions make uncertainty visible to later researchers. A partially legible word should not be silently completed from context. If an institution uses square brackets for supplied letters, illegible passages and editorial comments, those conventions should be defined before work begins. The notation must distinguish a missing or damaged passage from an original blank space.

For example, a transcription might render an unclear surname as '[?]arson' if the first letter is illegible and the institution's conventions allow that form. A completely unreadable passage might be marked '[illegible]'. If a word is crossed out but remains readable, preserve it and identify the deletion. These conventions allow subsequent cataloguers to reconsider the reading without having to reconstruct what the transcriber believed.

Compare the transcription with the source image

A transcription requires visual review at the level of individual entries. Compare the extracted text with the corresponding line, then check whether the line belongs to the correct row and column. Review especially closely where ink is faded, the paper is stained, writing crosses a ruled boundary or several entries appear close together. Enlarging a crop may help, but it can also remove the visual context needed to identify a letter or distinguish an annotation from the original entry.

A useful correction table has four columns: source location, proposed reading, corrected transcription and unresolved question. For example, page 14, entry 07 might contain an uncertain surname. The correction could retain the visible letters and mark the remainder as unclear. The final column would record the need for comparison with another occurrence of the name. This creates a transparent account of editorial decisions without presenting an uncertain interpretation as a settled fact.

• Check the first and last word of every line for segmentation errors.
• Compare names and figures character by character.
• Inspect overwritten and crossed-out passages separately.
• Retain unresolved readings in the working transcription.

Keep transcription and catalog normalization separate

A transcription reproduces the historical document as closely as the chosen conventions allow. A catalog entry makes that information useful for collection management. These are related but different activities. An original description such as '2 brass candlesticks' might be normalized in a catalog as 'Two candlesticks; brass' while the transcription preserves the exact wording, spelling and position of the original entry.

The distinction also applies to names and dates. An original date can be retained in the transcription while a normalized date is entered in a separate structured field. A doubtful donor identification should remain qualified until documentary evidence supports it. An AI-use note can state: 'Handwriting recognition assistance used to prepare a draft transcription; uncertain readings and final text reviewed against the source image.' Before beginning the next ledger page, prepare a page template with fields for source reference, line identifier, transcription, correction and uncertainty.

Questions readers ask