OCR Typewritten Documents: Reading Old Typed Pages

OCR can read most typewritten documents, because typewriter type is regular and evenly spaced. What goes wrong comes from the machine and the paper: uneven ink, clogged letters, corrections typed over mistakes, pale carbon copies, and characters the typist borrowed to stand for others. Convert old typed pages, then check figures, names and corrections against the image.

Can OCR read a page typed on a typewriter?

Yes. A ribbon original typed on decent paper is among the easier old documents to convert, since every letter has the same width and letters seldom touch. Results fall as you move from the original to its carbon copies, and from a well-kept machine to a worn one.

Sort a typed file into rough grades before you convert it, so you know where to spend your checking time.

No vendor's general description predicts how its software handles your typist's machine. Run a few pages. Schema Law's free plan reads one document a day of up to three pages, so you can upload a good page, a middling one and the worst carbon in the file, then read the parsed pages against the scans. The steps from upload to export are on the home page.

What makes typewriter type different from modern print?

Typewriter type differs from modern print in three ways that matter to recognition: every character occupies the same width, the ink varies from letter to letter, and worn or dirty type changes the shapes of the letters themselves. The first helps OCR. The other two are where the errors start.

Fixed pitch

On a standard typewriter an "i" takes as much room as an "m". The spacing looks loose to a modern eye, and it is a gift to software, because neighboring letters keep their distance. The pair "rn" is much less likely to fuse into "m" on a typed page than in a tight laser-printed font.

Fixed pitch also explains how typed tables were built. The typist lined up columns by counting spaces or setting tab stops. When converted text is shown in a proportional font, those columns collapse, even when every character is right. Put the text in a monospaced font such as Courier before you judge it. If the alignment is still lost, rebuild the table by hand from the image and check each row.

Uneven ink

Each letter on a manual typewriter was struck by a finger, through a ribbon that was wearing out. So within one line you may find:

• Light and dark letters side by side. Keys under the weaker fingers print faintly. A black-and-white scan can drop the faint ones entirely.
• Half-printed characters. A ribbon that has slipped leaves the tops or bottoms of letters missing along a whole line.
• Two-color ribbons. A character typed with the ribbon set between its black and red bands is part black, part red, and may scan as a broken stroke.
• Punctuation that punched through. Periods and commas hit hard enough to pierce the sheet, leaving a hole that scans as a dark or a white dot.

Changed letter shapes

Typewriter faces have heavy slab serifs, and the metal type collected ink and lint. Three faults show up on the page:

• Filled letters. The enclosed parts of "e", "a", "o", "d" and "g" print as solid blots. An "e" and a "c" with filled centers are hard to tell apart, and so are "o" and a filled "a".
• Letters out of line. One character rides above or below the rest because its type bar was bent. Software may treat it as a superscript or split the line.
• Damaged type. A nick in one letter appears every time that key was struck. The same wrong reading then recurs throughout the document, which at least makes it easy to search for.

Emphasis without bold or italics

A typist had capitals, underlining and spacing. Underlines run through the tails of "g", "p" and "y" and can turn into stray underscores. Headings typed with a space between every letter, such as "A G R E E M E N T", convert as nine separate letters and will not be found by a search for the word.

Strikeovers, correction fluid and typed-over errors

A typewriter has no delete key, so every correction stays on the page, and OCR reads the correction and the mistake together. Typed-over letters come back as one wrong character. Text hidden under correction fluid is gone from the scan. Treat any corrected passage as needing a read against the image.

Two of these deserve care.

Text that still reads as live. A word crossed out in pen, or struck with a line of hyphens, is usually legible to software. The output then contains a word the author removed. If you copy from the converted text, you may quote language that was deleted before the document was signed.

Corrections that change substance. In an old deed, will or contract, a changed figure or name can be the reason you are reading the document. Converted text flattens the page into one layer and cannot tell you that anything was altered. Look at each typed correction on the image, and consider scanning those pages in color so a white patch or a different ink stands out. What an alteration means legally is a question for the lawyer on the matter, and the answer varies by jurisdiction and document type.

When you prepare a typed working copy of a corrected page, show the changes openly. The post on transcribing handwriting to text sets out bracket conventions for strike-outs and insertions that work equally well for typed originals.

Carbon copies, onionskin and mimeograph sheets

Carbon copies, onionskin sheets and mimeographed pages are harder to convert than a ribbon original because each process produces a weaker image. Carbons are soft and thick, onionskin lets the next sheet show through, and duplicated sheets fade. In an old office file these are often the only copies that survive.

Carbon copies

The typist stacked letterhead, carbon paper and thin second sheets, and one keystroke marked them all. The top sheet went in the mail. The file kept the carbon. That is why the outgoing letters in a closed file are so often the poorest pages in it.

Each sheet further down the stack received less pressure through more layers. Letters spread, fine parts close up, and periods and commas may not print at all. Carbon also smears when handled, leaving gray clouds across the lines. Expect dropped punctuation, small words misread, and digits you cannot fully trust. If the recipient's file or a recorded copy holds the ribbon original, that is the page to ask for.

Onionskin

Onionskin is thin, translucent paper used for copies and airmail. On a scanner the writing on the sheet behind it, or on its own reverse, shows through as gray text between the lines. Scan each sheet singly with a sheet of plain white paper behind it. If the back of the same sheet carries the show-through, try a black backing sheet, which darkens the whole page evenly so the reverse-side text stands out less. Scan in grayscale and compare the two results.

Mimeograph and spirit duplicator sheets

Offices and courts ran off multiple copies from a typed stencil or master. Stencil copies have letters whose centers sometimes dropped out and whose strokes are uneven. Spirit duplicator copies, the purple ones, lose contrast with age, and a black-and-white scan of pale purple type on cream paper may capture almost nothing. Scan them in color and look at the result before feeding a stack. The guide to scanner settings that make OCR work covers color mode and resolution.

Old paper brings damage of its own: rust marks from paper clips, tape that has turned brown, water stains and brittle edges. The post on OCR for coffee stained and damaged pages deals with those.

Why did typists use a lowercase L for the number one?

Typists used a lowercase L for the number one because many early typewriters had no key for the digit. In typewriter faces the two shapes are close enough that the letter served for both. OCR then reads what is on the paper, an "l", and a year typed as "l9l4" becomes text no search for 1914 will find.

You can see the convention described in the instructions Cornell University Library posted beside a working machine in one of its exhibits. The page for its L.C. Smith typewriter, dated around 1915 to 1925, tells visitors that, like many early typewriters, the machine has "no #1 (use a lowercase L)". The same instructions explain that it has no exclamation point either, and that the typist makes one by typing another character, backspacing and typing a period.

That has an awkward consequence. When software returns "l9l4", it has read the page correctly. The character on the paper is the letter. The meaning is the digit. A faithful conversion and a useful one pull in different directions.

Borrowed characters you may meet

Not every machine or typist followed the same habits, so treat this as a list of things to look for on your own pages and not as a rule about all of them.

Handling it without damaging the text

1. Keep a faithful copy. Store the converted text as it came out, beside the scan.
2. Correct in a working copy only. Change "l9l4" to "1914" where you have looked at the page and the meaning is plain.
3. Never replace across the whole document. Swapping every "l" beside a digit will also rewrite "Apt. l" correctly and "Exhibit l" wrongly when the exhibit was lettered.
4. Search for the typed form too. When a date or amount matters, search for the version with the letter in place of the digit, and for the capital O version of any zero.
5. Check identifiers by hand. A lot number, book and page reference or docket number has no surrounding words to settle whether a stroke is a letter or a digit.

The wider subject of digits misread as letters, and the cross-checks that catch them, belongs to the post on why OCR gets numbers wrong.

Typewritten deeds, transcripts and closed case files

Typewritten deeds, transcripts and closed case files each put typewriter problems in a predictable place. In deeds it is the legal description. In transcripts it is the line numbers and the question and answer labels. In closed files it is the carbon file copies of outgoing letters.

Deeds and other recorded instruments

An old deed is often a pre-printed form with the parties, price and property typed in. The typed fill may sit on or across the printed lines, which is a form problem as much as a typewriter one. The post on OCR for scanned forms covers pairing each typed answer with its printed label.

The legal description is where a wrong character does the most harm. Bearings and distances such as "N 45o 30' E l20.5 feet" pack digits, borrowed characters and symbols into a few inches. Read every call against the image, one character at a time, and have a second person read it back. The copy you hold may also be a print from the recorder's microfilm, several generations from the typed original, with a book and page stamp over part of the text.

A typed quitclaim deed and a deed of trust fail in the same inch of the page. You read the fee simple language and any easement the typist fitted between the printed lines, one character at a time, because a borrowed lowercase L in a bearing changes the parcel.

Transcripts

Typed transcripts of hearings and depositions are long, regular and usually clean, which makes them good candidates for conversion. Watch three things. Line numbers in the left margin may be read into the text, so "12 Q. Did you" becomes part of a sentence. The "Q." and "A." labels may come out as "0." and "4." on a worn copy. And a carbon copy of a transcript has all the faults of any other carbon. The post on making deposition transcripts searchable covers citing by page and line.

Closed case files

A box from storage mixes everything: ribbon originals of incoming letters, carbons of outgoing ones, typed memos with pencil notes, court papers on legal-size sheets, and a few handwritten pages. A workable order of attack:

1. Separate handwritten pages from typed ones. Handwritten records need different treatment, and the post on old wills has a section on old court records, cursive and faded ink.
2. Scan onionskin and carbons singly with a backing sheet, in grayscale or color.
3. Convert the typed pages and grade them: dark ribbon original, worn or dirty original, early carbon, late carbon.
4. List the pages you could not read well, so a later search that finds nothing is understood correctly.
5. Keep the scans with the text. For a typed page with corrections and borrowed characters, the image is the document, and the text is how you find your way to it.