Reading Old Words: OCR and Reliable Transcription

Decide What You Are Making

Transcription turns words on a document into text that can be searched and read. It differs from translation and from summarising. OCR, or optical character recognition, can provide a draft for printed material, but cannot decide whether a name belongs to your relative. Separate original images, literal transcription, and interpretation notes so readers can distinguish the document from your understanding.

Imagine a fictional family letter saying, “Arriving at the port next month; staying with Ah Wing first.” Preserve that wording in the transcription. An interpretation note can explain that the original word may mean a port or town, and list places still to be confirmed. Do not change it to “He arrives in Sydney next month,” even if that seems plausible. Family experiences deserve careful reading, and the text deserves its original boundaries.

Give the Image a Stable Identity

Before recognition, record who supplied the material, when you obtained it, how many pages it has, and whether you saw the original. Give each item an identifier and use explicit filenames for fronts, backs, and envelopes. Put the identifier and page number above the transcription and retain original photographs. Brightening, rotation, and cropping belong on working copies so processed images are not the only versions left.

Inspect the whole page before magnifying details, including margins, seals, table headings, and surrounding text. An isolated character meaning “four” is easy to misread; its position in a month column may clarify it. Record glare or missing corners and try to photograph the item again. A machine cannot turn uncaptured content into documentary fact. Clear input helps more than repeatedly switching recognition tools.

Treat Machine Output as a Draft

For a clearly printed page, run OCR and then compare every line with the image. For cursive handwriting, vertical text, variant characters, or mixed Chinese and English, test a small passage first to see where errors occur. Tidy output does not establish accuracy: columns may be joined incorrectly, dates may lose digits, and surnames may become common words. Label unchecked text “machine draft.”

Prioritise names, dates, places, amounts, and relationship terms while still reading every line. In a fictional ledger, misreading “borrowed ten dollars” as “borrowed a thousand dollars” changes the family story entirely. Record corrections in checking notes, including the machine result and your reason, so a later reader can understand your choice.

Set Rules for Unclear Passages

Write short project rules: retain original misspellings, use square brackets for editorial explanations, and mark illegible text as “[unclear].” Archive platforms may have their own conventions; read official instructions before contributing. In a family project, consistency matters, and your supplied letters must not appear to have been present in the document.

If only a surname is legible, write “Chen [given name unclear]” and list two possible readings with their supporting reasons in notes. Do not fill every blurred character as Wen because the pedigree already contains a Chen Wen. Leaving a gap is not failure. It identifies the need for a better image, other examples of the writer’s hand, or another reader familiar with that writing style.

Read by Comparison Rather Than Guesswork

For unfamiliar handwriting, find repeated characters on the same page whose readings are certain and build a small comparison table. A writer who makes the second tree element in Lin unusually short may help you recognise the surname elsewhere. Note the comparison’s line and avoid imposing handwriting from a different author or period. Similar shapes support a reading, but sentence and position must also fit.

Ask another reader to interpret a crucial passage independently before exchanging results. Saying first, “This should be grandfather’s name,” can steer the answer. If readings differ, keep both and describe the differences rather than replacing checking with a vote. When one character affects identification, pause the conclusion and focus on obtaining a clearer copy.

Save Transcription and Translation Separately

After literal transcription, make a modern Chinese reading or English translation and label it as another interpretive layer. Old terms of relationship and place names may have no single equivalent; retain the term with an explanation where needed. Translation must not remove ambiguity: a letter saying only “elder brother” does not establish birth order beyond that term, full sibling status, or a personal name. Let readers return from translation to its original sentence.

A fictional note saying “sister-in-law to receive” can be translated accordingly, with a note that the individual addressed remains unidentified. Do not add a wife’s name to smooth the story. Date and credit each version and preserve revision notes. These layers help English readers understand Chinese material while seeing where judgment remains unfinished.

Deliver Text That Can Be Checked

Before sharing, check the image, transcription, translation, and questions together, making sure each passage has an item identifier and page number. If sharing extracts, explain omissions and do not remove surrounding sentences that change the meaning. Searchable text helps find leads, but when using a fact, return to the image to confirm it is present and has not been changed by misplaced columns, missing characters, or mistranslation.

Completion does not mean every character has an answer; every answer must have an explainable origin. Give unclear passages next steps, such as photographing the envelope’s back or comparing three signatures. Save a checked version and record later discoveries separately. When old words become readable, we approach the people who wrote letters, kept accounts, and waited for news, rather than only a name on a pedigree. This patience lets later family members participate in checking, too.

For tables, write column headings first and then record rows, preventing adjacent households’ names from being joined. Explain treatment of blank fields and repetition marks: a blank means the original is unfilled, not that research has established the information does not exist. Expand a ditto mark only when its reference is clear. If converting vertical text to horizontal, explain reading order and retain markers locating original columns for checking. Also record whether a table continues across pages and verify that the next page belongs to the same item.