OCR Newspaper Clippings, Legal Notices and Old Articles

OCR can read a newspaper clipping when the scan is sharp enough for small newsprint type, but the article may come out in the wrong order. Narrow columns, neighboring stories, hyphenated line ends and jumps to another page all break the text. Check the reading order, and record the paper, date and page yourself.

Can OCR read a newspaper clipping?

Yes, as far as the letters go. Newsprint type is small and the paper is poor, so expect more wrong characters than on office print. The larger risks are different ones: sentences assembled out of sequence, and a block of text that no longer says which newspaper it came from or when.

The Library of Congress, which runs OCR over the newspapers in its Chronicling America collection, is plain about the ceiling. Its post on improving machine-readable text for newspapers in Chronicling America says "achieving 100% accuracy in an automated OCR process is virtually impossible" for historical newspapers on microfilm, and adds that such newspapers "often exhibit tight columns and tiny text sizes that further complicate the OCR process." A clipping in a legal file shares both features.

What you were handed matters as much as what it says.

Newspapers printed before modern typefaces add one more difficulty, with letterforms that differ from modern print. The post on OCR for old documents covers those.

Why do narrow newspaper columns come out in the wrong order?

A newspaper page packs several columns side by side with slivers of white between them, and runs headlines across several at once. Recognition has to decide where each column starts and stops before it reads a word. When it misjudges a gutter, it reads straight across two columns and splices unrelated half-lines together.

Reading order as a general subject, and how software finds column edges, is explained in the post on whether OCR can handle two-column documents. A newspaper is the same problem with every difficulty turned up.

What is different about a newspaper page

• The gutters are narrow. The white between two columns may be little wider than the space between two words, and a slight tilt in the scan closes it.
• Headlines span columns. A headline across three columns tells the layout step that the three belong together. It may then treat them as one wide block.
• Stories interlock. An article can run down one column, continue at the top of the next, and wrap around a photograph or an advertisement on the way.
• Neighbors intrude. A clipping cut along the article's edge still carries the last letters of the story to its left and the first letters of the story to its right.
• Small items sit inside the article. Bylines, datelines, photo captions, pull quotes and subheadings each look like a separate block.

Any engine has to assume some layout before it starts. Tesseract's documentation on improving the quality of OCR output lists its page segmentation modes, which range from "Fully automatic page segmentation" to "Assume a single column of text of variable sizes." Whichever assumption a tool makes, a clipping with one and a half columns of another story attached will test it.

Signs that the order is wrong

Read the converted text aloud, or read only the first three words of each line. Wrong order shows up as:

• a sentence that stops mid-clause and resumes on a different subject;
• two names from different stories in one sentence;
• a line twice as long as its neighbors, where two columns were joined;
• the headline appearing in the middle of the body;
• a caption stranded between two paragraphs.

Getting the columns back in order

1. Duplicate the scan. Everything that follows is done to the duplicate.
2. Crop the duplicate to the article, cutting off neighboring stories and advertisements.
3. Convert it and read the text against the image, paragraph by paragraph.
4. If columns are still spliced, crop each column into its own image, convert them separately, and join the text in reading order.
5. Put the headline, byline and dateline at the top by hand if they landed elsewhere.
6. Note at the head of the working copy that the article was isolated from a larger page and reassembled.

Cropping to single columns is more work than one conversion, and it removes the layout decision from the software altogether.

Hyphenated line breaks and jumps to another page in a converted article

A narrow column holds only a few words to a line, so words are split with hyphens at line ends all the way down. Converted text can keep every one of those splits. Longer stories also stop partway and continue on another page. Both leave you with text that looks whole and cannot be searched or quoted as it stands.

Stitching hyphenated words

"The defen- dant" in the text means a search for "defendant" finds nothing. In a narrow column, a surname or a company name can be one of the split words, and those are the words you search for.

To repair it in a Word export:

1. Search for a hyphen followed by a line or paragraph break. Your word processor's find dialog has a code for each kind of break; look it up in the program's help.
2. Also search for a hyphen followed by a space, to catch splits that conversion flattened onto one line.
3. Look at each hit before changing it. Do not use Replace All.
4. Where the hyphen only marks the line end, delete the hyphen and the break: "defen- dant" becomes "defendant".
5. Where the hyphen belongs to the word, keep it and delete only the break: "court- appointed" becomes "court-appointed".
6. For a name, check the spelling against another occurrence in the article or another document in the file.

Step 5 is the reason for step 3. A compound that happens to break at its own hyphen looks exactly like a split word, and joining it changes the text.

Following a jump

A front-page story may end with a line such as "See CONTRACT, Page A7", and the rest appears on that page under a short label. A clipping file may hold the first part, the second part, or both on separate scraps.

Take a fictional example. The Marlow County Ledger reports on a paving contract awarded to Delacroix Paving. The front-page portion ends mid-sentence with "the commission voted 3 to 2 to" and a jump line. The continuation on page A7 begins "award the contract despite". Converted separately, neither fragment contains the full sentence, and a search for "voted 3 to 2 to award" fails.

Handle a jump this way:

• Convert each part and put them in one working document in reading order.
• Mark the join in brackets where it occurs: "[continues on p. A7, col. 2]".
• Delete the jump line and the continuation label from the running text, and keep them in the bracket so the page reference survives.
• If you hold only one part, say so at that point: "[story continues on p. A7; continuation not in file]". Do not let half an article pass as the whole.

Yellowed newsprint, show-through and halftone photographs

Newsprint is thin, absorbent and quick to discolor. Three things follow for a scan. The background darkens toward tan, which narrows the contrast with the ink. Print on the reverse shows through as gray ghosts. And photographs, printed as patterns of dots, become fields of specks that software tries to read as characters.

Yellowed and browned paper

A black and white scan has to place the tan background on one side of its cutoff. Set light, the faintest print goes with it. Set dark, the paper becomes speckle. Scan a discolored clipping in grayscale or color so the choice between ink and paper is not forced at the scanner.

Show-through

Hold the clipping to the light. If you can read the other side through it, the scanner will see it too, because the white lid reflects light back through the sheet. Put a sheet of black paper behind the clipping on the glass. The reverse print then has nothing bright behind it and fades from the scan. This changes nothing on the clipping itself.

Halftone photographs

A printed news photograph is made of dots of varying size. Scanned, the dots can be read as full stops, colons, the letter "o" and stray symbols. In the converted text this appears as a block of nonsense where the photograph was.

• Delete that block from the working copy and put a bracketed note in its place: "[photograph: three people at a podium]".
• Keep the caption. It may name the people shown, and it is text the article itself does not repeat.
• Check that the caption did not end up fused to the paragraph beside it.

Handling a brittle clipping

• Scan on a flatbed. Do not send old newsprint through a sheet feeder.
• Unfold it once, gently, and scan it. Each unfolding costs a little more of the crease.
• Do not tape tears or stick the clipping to a backing sheet before scanning. If it is evidence or a client's property, leave repairs to whoever is responsible for it.
• Photocopy or scan it at intake, because newsprint keeps darkening.

Print that has paled, as opposed to paper that has darkened, is covered in the post on OCR for faded text.

Published legal notices and affidavits of publication as scanned exhibits

A published legal notice is a block of small type among a newspaper's notices, such as a summons, a notice of sale or a notice to creditors. The publisher's proof that it ran is an affidavit of publication, and the notice itself may be attached to it. Both reach a law office as scans.

What a notice must say, where it must run and how publication is proved are set by the statute or rule that requires the notice. Two examples show the kind of wording to look for in your own jurisdiction. Washington's RCW 65.16.030 provides that "the proof of such publication shall be the affidavit of the printer, publisher, foreman, principal clerk or business manager of the newspaper which published said notice." Florida's statute on proof of publication, section 50.041(2), describes a paper affidavit with a margin in which "shall be substantially pasted a clipping which shall be a true copy of the public notice or legal advertisement for which proof is executed." Other states and courts have their own provisions. Read the one that governs your notice.

Why the affidavit page converts badly

A paper affidavit of the kind the Florida statute describes has two texts side by side: the sworn statement in ordinary type, and a narrow clipping in tiny type beside it. That is a two-column layout with columns of different type sizes and different line spacing. Conversion can interleave them, giving a line of the affidavit, then two lines of the notice, then the next line of the affidavit.

The clipping may also be longer than the space allowed and folded up, attached at an angle, or pasted over part of the form. Unfold it for the scan if you hold the paper and can do so without damage. If you received a scan with the fold in place, the hidden part of the notice is not in your file, and you need a better copy.

A notice of sale clipped to that affidavit is where you verify the foreclosure date, the property, and any lien, right of first refusal, or partition language the order required. Tiny type splits those words across the column, so you read them from the clipping itself before you tell anyone the notice matches.

What to verify on a notice and its affidavit

Notice type is small, and much of what it spells out is names, numbers and dates. Treat the converted text as a draft and check these against the image:

If your conversion is for search only, errors in a notice cost little. If you are confirming that the published notice matched the order that authorized it, compare the notice with the order character by character from the images. Do not rely on a comparison of two converted texts for that.

How do you record a clipping's newspaper, date and page beside its text?

Type it at the top of the converted text yourself, from what the scan shows. Conversion gives you the words of the article and drops what made them a citation: the paper's name, the date, the page, and how the copy reached you. A clipping cut from the page has lost the first three already.

A header block for each clipping

Put a block like this above the text of every converted clipping:

Source:        Marlow County Ledger
Date:          14 March 1998
Page:          A1, cols. 3-4; continued A7, col. 2
Headline:      Commission awards paving contract in split vote
Byline:        [none shown]
Copy held:     Photocopy of clipping, client's file, received 2 June 2026
Source shown:  Paper's name and date handwritten on copy; no printed folio line
Converted by:  JR, 9 June 2026; columns cropped and rejoined; hyphenation repaired

Where each detail comes from

• The folio line. If your scan includes the strip at the top of the page, look there for the paper's name, the date and the page number, and copy what it shows.
• A handwritten or stamped note. Someone wrote "Ledger 3/14/98" in the margin. Record it, and record that it is an annotation and not part of the printed page.
• The archive or database. If the page came from an online archive, note the archive, the search date and any identifier it gave the page.
• The reverse side. A dated advertisement or the end of another story on the back can narrow down the issue when the front shows nothing.
• Nothing. Write "not shown". An honest gap is better than a confident guess, and it tells the next person what to look for.

Keep the image and the text together

Name the scan and the converted text so they sort together, for example 1998-03-14 Marlow Ledger A1 paving contract.pdf and the same name with .docx. Keep the uncropped scan as the record and the cropped duplicate as a working file. A searchable PDF is useful for the full page, because it keeps the picture of the page with its folio line. Word or plain text is the better home for the reassembled article. Schema Law lists both among its export formats and lets you export one upload in more than one of them.

When you quote the article, quote from the image, cite from the header block, and say what kind of copy you worked from. How a clipping is authenticated or admitted is a matter for the evidence rules of your court. The header block does not settle that question, but it collects the facts you will be asked for.