Convert Handwriting to Text: A Method for Legal Files

To convert handwriting to text, scan the page in grayscale or color, run it through software that recognizes handwriting, then correct the draft against the page image and export it as Word, plain text or a searchable PDF. The software gives you a first draft. The corrected version is the one you can search and edit.

What does it take to convert handwriting to typed text?

Converting handwriting to typed text takes four things: a clear image of the page, a recognition pass, a person who corrects the result, and a place to keep the typed version beside the image. Skip the third and you have a draft with unknown errors. Skip the fourth and nobody can check it later.

Each stage has a different owner and a different cost.

People who want to change handwriting to text often picture one step, the recognition pass. In practice that pass decides how long correction takes, and correction decides whether the job was worth doing. A page that comes back nine-tenths right is a quick edit. A page that comes back half right is slower to repair than to retype.

Whether a given tool can read a given hand is a separate question. The post on handwritten notes explains what handwriting recognition can and cannot read, and the short version is that no description of a product settles it. A test on your own pages does.

Set the goal before you start, because it changes how much correction you owe the page. A client's forty pages of journal entries that you only need to search for a name call for a light pass. A two-page handwritten statement you plan to quote in a letter calls for a word-by-word check.

A two-page living trust or pour-over will needs that word-by-word check. You compare the grantor's name with the trustee named to hold the assets, and you look for a revocable trust label on the cover and again beside the signature. Those repeats are how you catch a letter the software smoothed into a different surname.

Sorting a handwritten file into print, cursive and mixed pages

Sort the pages by handwriting style before you convert anything, and start with the printed-style pages. Separate block print, joined cursive and mixed pages into three piles. Recognition does best on the first pile, so it shows quickly whether the tool is worth using on the rest.

The sort takes a few minutes for a file of fifty pages. You are looking at the writing, not reading it.

Three more sorting rules save time later.

Sort by writer as well as by style. A file of handwritten notes to text is usually several people's notes. A tool that manages one adjuster's print may fail on the next adjuster's. Keep each writer's pages together so one test page speaks for the pile.

Pull out pages that are mostly numbers. Mileage logs, payment tallies and medication lists look easy because the entries are short. They are the riskiest pages, since a wrong digit reads as plausibly as a right one. Put them in a pile that always gets a full check. The post on why OCR gets numbers wrong covers the misreads to expect.

Set aside pages you will never need as text. Doodled phone-message slips, envelopes and duplicate copies can stay as images. Converting every sheet in a box is rarely the task you were given.

Pre-printed forms deserve a note of their own, because the handwriting on them is tied to typed field labels. The guide to OCR on scanned forms deals with keeping each answer beside its field.

Converting a handwritten page to typed text in five steps

Converting a handwritten page to typed text follows the same five steps whatever tool you use: capture the page, run recognition, judge the draft, correct it against the image, and export. The third step is the one people skip, and it is where you decide whether to keep going or retype.

1. Capture the page. Scan the original, not a photocopy, and keep the gray tones or the color so that light strokes survive. Include the whole sheet, edges and all. The guide to scanning handwriting into text covers the settings for pencil, colored ink and lined paper.
2. Run recognition. Upload the image to the tool you are testing. In Schema Law that means signing in with Google and dropping in a PDF, PNG or JPEG of up to 20 MB. The pages are read into text, and you see the result page by page on screen, where you can check it against the original.
3. Judge the draft before you fix anything. Read the first ten lines of the draft against the image and count the words you would have to change. If it is roughly one word in ten or fewer, correct the draft. If it is closer to one word in three, stop and retype from the image, because hunting for errors in bad text is slower than typing clean text.
4. Correct line by line. Put the image on one side of the screen and the draft on the other. Read a line of handwriting, then fix the typed line. Keep the writer's spelling and abbreviations. Where you cannot read a word, mark the gap and move on. Do not supply what the sentence probably said.
5. Export and label. Save the corrected text in the format the next person needs, and say at the top what it is: a typed version of a handwritten page, who corrected it, and from which image.

Step 4 has a few rules for marking gaps, strike-outs and inserted words that matter when anyone else will rely on your typed version. The post on how to transcribe handwriting to text sets out those conventions, so only the principle is stated here: the typed text should never look more certain than the page.

One fact about recognition output shapes step 3. Software rarely returns nonsense for handwriting it cannot read. It returns real words that happen to be wrong, so a draft can look clean and still say "received" where the writer put "refused". The Schema Law homepage says as much in its own questions section: recognition is weaker on handwriting, and nothing marks a passage as uncertain. Assume the same of any tool until its documentation says otherwise. Count errors by comparing with the image, never by scanning the draft for words that look odd.

A worked example

A paralegal has an eleven-page handwritten timeline from a client, Ms. Okafor, in a fictional wage claim. Seven pages are careful print and four are cursive written late at night. She scans all eleven in grayscale, converts the seven printed pages first and finds two or three corrections in the first ten lines of each. She corrects those seven. The first cursive page needs a change in almost every line, so she types the four cursive pages from the image and asks Ms. Okafor to confirm two words on page nine. The result is one Word file of eleven typed pages, with a note on which pages came from a recognition draft and which were typed by hand.

How long does handwriting conversion take compared with retyping?

Handwriting conversion is faster than retyping only when the recognition draft is mostly right. Correcting a good draft takes a fraction of the typing time. Correcting a bad one takes longer than typing, because you read every word twice. The honest way to know is to time both on a sample from your own file.

The comparison has three parts, and people usually count only the first.

• Typing from the image. A steady typist reading unfamiliar handwriting works far below normal typing speed, because most of the time goes on deciphering, not on keystrokes.
• Correcting a draft. You still read every handwritten word. What you save is the typing of the words that came back right.
• Checking. A retyped page and a corrected page both need a second read against the image if the text will be quoted. This cost is the same either way, so leave it out of the comparison.

Timing a sample

1. Pick two pages by the same writer that look alike in density and neatness.
2. Type the first page from the image. Note the minutes.
3. Convert the second page, correct the draft against the image, and note the minutes, including the upload and the export.
4. Compare the two times, then multiply the difference by the number of similar pages in the pile.
5. Repeat for each writer or style. One writer's result says nothing about the next.

Here is how the arithmetic reads, with invented figures. Suppose typing page one takes 14 minutes and converting and correcting page two takes 6. The pile holds 30 similar pages, so conversion saves about 8 minutes a page, or four hours across the pile. Now suppose a second writer's page takes 14 minutes to type and 17 to correct. For that pile you type, and the software's draft is useful only for searching. Your own numbers will differ, which is the reason to take them.

Two things distort a sample. The first page you correct is always slow, because you are learning the writer's letter shapes, so do not time your very first attempt at a new hand. And a page of names and figures takes longer to correct than a page of narrative, whatever the totals say.

Which export suits converted handwriting: Word, text or searchable PDF?

Choose Word when you will correct or reuse the text, plain text when you only need the words, and a searchable PDF when the handwriting itself must stay in view. Many handwritten pages need two exports: an editable one for the corrected text and a searchable PDF to keep with the image.

The third row matters more for handwriting than for print. A searchable PDF keeps the picture of the page and puts recognized text behind it, as the guide to making a scanned PDF searchable explains. If the recognition was wrong, search inherits the error. A search for "Okafor" will miss a page where the hidden layer says "Okator". Treat a searchable PDF of handwriting as a finding aid, and treat the corrected Word or text file as the reading you stand behind.

Every Schema Law plan, the free one included, exports Word, searchable PDF and plain text, and the homepage lists the full set of export formats. The pricing page shows how many pages each plan allows. One upload can be exported more than once, so you can take both a Word copy and a searchable PDF.

Spreadsheet formats come into play when the handwriting sits in a grid, such as a mileage log or a sign-in sheet. Expect to rebuild the rows by hand if the columns were drawn freehand. For a wider view of matching a format to the person who receives the file, see the post on legal document conversion.

Where should the typed version of a handwritten page be filed?

File the typed version in the same folder as the scan of the handwritten page, with a name that ties the two together. The image is the record of what the writer put on paper. The typed text is a reading of it. Anyone who opens one should be able to find the other in seconds.

A naming pattern that works is the same base name with a suffix for what the file is:

2024-03-18 Okafor timeline (scan).pdf
2024-03-18 Okafor timeline (typed, corrected by JR).docx
2024-03-18 Okafor timeline (searchable, uncorrected).pdf

The words in brackets do real work. "Corrected by JR" tells a later reader that a person compared the text with the page. "Uncorrected" warns that the hidden text is the software's draft.

Put the same facts inside the typed file, at the top, where a pasted excerpt will not carry them away by accident but a reader of the whole file will see them:

• the name of the image file the text was taken from
• who corrected it and on what date
• which pages began as a recognition draft and which were typed from the image
• any pages or passages that could not be read

A few filing habits prevent the common failures.

Keep the scan unaltered. If you crop, rotate or brighten a page to read it, save that as a separate working copy. The first scan stays as it came off the scanner.

Do not let the typed version travel alone. When you send the typed text to a colleague, an expert or a client, attach the scan or say where it is. A typed page with no image invites people to quote it as if it were the document.

Download what you need from any online tool. A converter is not your filing system. Schema Law, for instance, keeps a document in your account until you delete it and removes nothing automatically, but the account belongs to one person. Export the files and store them in the matter folder the same day, where colleagues can reach them.

Record the paper's location. If the original notebook or letter still exists, note in the matter file who holds it. The handwriting may matter later in ways the typed text cannot answer, and the paper is what a document examiner or a court would ask to see.