How to Check OCR Errors Before Quoting a Scanned Page

How do you check OCR errors before quoting a scanned page?

Compare the extracted text with the page image, word by word, before treating it as a quotation. Confirm the correct document and page, follow the printed reading order, and record any correction in your research notes. Correct a machine misreading only when the image supports the replacement. Preserve errors that genuinely appear in the source, and leave unresolved wording uncertain rather than guessing. A searchable transcription helps you find words; it does not independently verify them.

What is the difference between the image and the text?

Optical character recognition, or OCR, produces machine-readable text from an image. A digital collection may therefore show both a photograph or scan of a page and a separate text layer that you can search, select or copy.

The National Archives' guide to extracted text describes its machine-generated text as a starting point that still needs human review. That is an important distinction when one letter could decide whether a sentence says “not” or “now.”

Check what kind of text you are reading. A collection may distinguish machine extraction from a contributed transcription. Neither label means you have personally compared the passage with the image. The National Archives' Citizen Archivist FAQ also explains that it does not formally review every contributed transcription and does not treat one as permanently final. This describes that program, not every digital library.

Our workflow below concerns checking a short passage for quotation. It does not ask you to correct an entire archive or submit public edits.

Which page are you actually looking at?

Before comparing letters, make a small locator note. Record the work or document title, author or attributed creator, edition or issue date, repository and stable item link. Then record the particular page.

Keep the number printed on the page separate from a viewer's image counter. In a fictional example, printed page 12 might be image 15 because other images precede it. That does not establish a universal three-page offset; the next volume may be arranged differently.

Use labels such as “printed page 12; viewer image 15” in your working note. If there is no visible page number, say so and save the repository's image identifier or other locator. Do not manufacture a printed number from a counter.

For newspaper material, add the article heading and column location when useful. For a book, check the title and edition rather than assuming two similarly titled digital copies contain identical wording. The purpose is practical: another reader should reach the same passage, not merely the same website.

How do you compare the passage without losing your place?

Keep the page image beside the copied text. Work through a short section at a time, enlarging the image enough to distinguish its characters. Use these as our editorial checking passes:

1. Find the boundaries. Identify where the passage starts and ends on the page. Read the preceding and following sentence.
2. Follow the layout. Notice column boundaries, footnotes, headings and continuation notices before accepting the text layer's sequence.
3. Compare the words. Check every word rather than only the ones that look strange.
4. Compare small marks. Recheck punctuation, apostrophes, accents, hyphens and numerals.
5. Record the decision. Separate a confirmed machine error from an unresolved reading or an error printed in the document.

A smooth sentence deserves checking too. “Now” is a normal word, but it is wrong if the page says “not.” Spell-checking alone cannot decide which one the author printed.

Once the wording is secure, use the wider quote-verification guide to examine attribution and context. Reading a sentence accurately does not prove that the person named on a quotation graphic originally said it.

Can you practise with a deliberately faulty extraction?

The following passage and extraction are invented for this guide. They are not a real historical quotation, a library record or the output of an OCR test. Treat the first block as the legible source text for this paper exercise.

Fictional source text:

The reading room will not open on Monday. Twelve readers may request a seat for Tuesday, provided the catalogue work is complete.

Deliberately faulty extraction:

The reading room will now open on Monday. Twelve readers may request a seal for Tuesday, provided the catalogue work is complete.

Compare the versions before reading the answer key.

There are two changed word positions. Both replacements create ordinary English words, so neither can be dismissed as an obvious string of nonsense. The second sentence's condition is unchanged; it should remain in a quotation of that full sentence.

The corrected transcription is the first block exactly. Do not improve it to “The reading room opens on Tuesday.” That proposed rewrite adds a certainty the source does not supply: the passage allows requests subject to a condition. Accurate transcription and summarizing are separate jobs.

This exercise demonstrates a comparison method, not the frequency of these errors in any actual system.

What if the source itself contains an error?

First establish where the oddity originates.

The National Archives' transcription tips ask contributors to preserve original wording, including misspellings and punctuation, and to mark unreadable words rather than invent them. Its instructions address catalog transcription.

For a published quotation, follow the assigned quotation style when deciding how to explain a genuine source error. Do not attach “sic” to a mistake made only by the text extraction: that would wrongly suggest the oddity was printed in the document. The quotation-shortening guide covers editorial alterations separately.

What should you do with hyphens and columns?

Look at the layout before joining words. A word split at the end of a printed line is not necessarily a permanently hyphenated compound. Conversely, a hyphen inside a compound is not disposable merely because another nearby hyphen marks a line ending.

The National Archives' getting-started instructions permit removing line-ending word-division hyphens for searchability. That is not permission to remove every hyphen from a quotation or flatten a poem's line structure. Keep a note of any joining decision when the layout matters to your use.

For columns, draw a reading path in your own notes: article heading, first column, continuation. If the copied text appears to jump into an advertisement or neighboring report, locate each segment in the image. Do not splice them into one sentence because the text layer placed them together.

If the article continues elsewhere, inspect that continuation before deciding you have the whole statement.

What if enlargement still does not resolve the word?

Stop short of a confident quotation. Try another available scan of the same edition, compare clear examples of the relevant letter on the page, or ask the holding library for reading assistance. Context can suggest candidates, but a plausible candidate is not a legible letter.

If you inspect another edition, retain its separate citation. It may help you investigate without proving what an unreadable patch in the first image says.

Your working note can simply say “one word unresolved; exact quotation not ready.” Avoid publishing a guessed reading under quotation marks. If the uncertainty changes the claim, paraphrasing that guessed claim does not solve the problem either.

What should you save after checking?

Keep three separate records:

• the extracted text you initially encountered;
• your checked transcription, including unresolved portions; and
• the final quotation you intend to use, with its citation and context.

Add a short correction log: location, extracted reading, visible reading and reason for the change. In the fictional exercise, the log contains the two replacements in the answer key. It does not claim that a real archive was corrected.

A commonplace book can hold this evidence beside the passage you want to remember. Label the checking status specifically: “compared with this page image” is narrower and more informative than “verified.” It leaves attribution, context and any remaining uncertainty visible.

Sources

• National Archives: Get Started Transcribing, including extracted text and line-ending hyphen guidance; reviewed 6 February 2025.
• National Archives: Citizen Archivist FAQ, including transcription review status and extracted-text distinctions.
• National Archives: Transcription Tips, including source spelling, punctuation and unreadable text; reviewed 12 March 2025.