OCR Handwritten Will and Probate Records: Limits

OCR can read a handwritten will only in part. Recognition software does well on neat, separated print and poorly on cursive, old ink and cramped lines, so its output is a rough draft for a person to correct. Typed probate records and trust documents convert far better. In every case the original will controls, never the transcript.

That shapes how an estates practice should use the technology: freely on typed records, cautiously on handwriting, and always with the paper or its image as the authority. The sections below take each kind of document in turn.

Typed probate papers often share a scan with the handwritten pages. When you find a codicil or a petition beside the last will and testament, search the typed text for the executor's name and for any paragraph on testamentary capacity or undue influence. The probate proceeding cites those pages. Confirm every name and date on the image, and keep reading the handwriting from the page itself.

Can OCR read a handwritten will?

OCR can sometimes read a handwritten will well enough to give you a draft, and rarely well enough to rely on. Clear block letters on unlined white paper may come through mostly intact. Joined-up writing, a shaky hand or a crowded page will produce text with gaps and wrong words.

The trouble is that a will has no unimportant words. In a letter, a misread word costs little. In a will, the difference between "all" and "half", or between "my son Daniel" and "my son David", is the whole gift. The places recognition is weakest are the places that matter most:

• Names. Recognition leans on context to guess a word, and a surname has none.
• Figures. Handwritten amounts and percentages, especially a "1" against a "7" or a "4" against a "9".
• Changes. Words struck through, written over, or squeezed in above the line.
• The date. Often written quickly, sometimes in a format that could be read two ways.

Wills on printed forms raise a different problem. The printed wording converts cleanly, and the handwritten entries in the blanks, which are the only parts unique to this will, are the parts most likely to be wrong or missing. A transcript that looks tidy can hide that.

Schema Law makes no stated claim about how accurately it reads handwriting. A will is usually short, and the free plan takes one document a day of up to three pages, so you can try the actual document and judge the draft for yourself. What recognition can do with handwriting in general is covered in the post on OCR and handwritten notes.

Be realistic about the saving. A one-page will does not take long to type from the image. If the recognized draft needs every line corrected, starting from a blank page may be quicker and less error-prone, because you are not tempted to accept a plausible wrong word.

Why should the transcription never replace the original?

A transcription of a will is a reading aid, and the original document is the evidence. With a handwritten will the handwriting itself can be what makes the document valid, so a typed version removes the very thing a court would look at.

Arizona's statute shows why. Under Arizona Revised Statutes section 14-2503, a will that does not meet the ordinary execution requirements "is valid as a holographic will, whether or not witnessed, if the signature and the material provisions are in the handwriting of the testator." That is one state's rule. Whether a holographic will is recognized at all, and on what conditions, depends on the state, so check the law that governs the estate in front of you.

Where a rule like that applies, the questions are about the writing: whose hand it is, which provisions are in that hand, and where the signature sits. A transcript answers none of them.

The general preference for originals also appears in evidence rules. Federal Rule of Evidence 1002 states that "an original writing, recording, or photograph is required in order to prove its content unless these rules or a federal statute provides otherwise." That is a federal rule. Your probate court's own evidence rules govern, so read them, but the idea carries over: the content of a document is proved by the document.

A typed transcript also drops information that may matter later:

So label every transcript for what it is. Put "Working transcript. Not the will." at the top of the page, name the image it was made from, and never circulate it without that image attached. Scan or photograph the original once, store the paper safely, and do your reading from the image.

Converting typed probate records and trust documents

Typed probate records and trust documents are ordinary printed text, and they convert as well as any other clean scan. This is where OCR pays for itself in an estates practice: wills prepared by a law office, codicils, trust instruments, court orders, inventories and accountings.

An estate file that has been made searchable lets you find every mention of a beneficiary, an account number or a parcel of land across decades of paper. That matters when a trust was amended four times, or when an accounting has to be reconciled with an inventory filed years earlier.

Inventories and accountings are tables, and a table is only useful if its rows and columns survive. After export, add up each column in the spreadsheet and compare the total with the total printed on the page. A mismatch tells you a figure was misread or a row was dropped. The post on scanned PDF to Excel explains how tables are detected and where they break.

Expect trouble at the end of each instrument. Signature blocks, witness attestations, notary acknowledgments and embossed or inked seals sit close together, and text under a seal may be unreadable. The post on whether OCR can read stamps over text describes what to expect there.

In Schema Law the upload can be a PDF, PNG or JPEG of up to 20 MB. Word, searchable PDF, plain text and Excel are available on every plan, and the pricing page lists the page allowance of each.

Old court records, cursive and faded ink

Old probate records are the hardest material in this area. A file from several generations back may hold a clerk's cursive ledger entries, typewritten carbon copies, and prints made from microfilm. Expect OCR to help with the typewritten pages and to give you little on the rest.

Each kind of page fails for its own reason:

• Clerk's cursive. Letterforms and abbreviations from an earlier period differ from modern handwriting. Software trained mostly on recent writing has less to go on.
• Faded ink. Ink that has turned pale brown on yellowed paper leaves little contrast. A black-and-white scan may erase the strokes entirely.
• Bleed-through. Writing from the back of a thin sheet shows on the front, and recognition tries to read both sides at once.
• Carbon copies. The letters are soft and thick, and closed shapes such as "e", "a" and "o" fill in.
• Microfilm prints. These may be negative images, white on black, with scratches and uneven focus across the frame.

Before you give up on a page, look at how it was captured. A color or grayscale scan keeps faint strokes that a pure black-and-white scan throws away, and the post on digitizing law firm paper files covers scanner settings. If the copy came from a courthouse or archive, ask whether a fresh image can be made from the original volume instead of from film. The post on OCR for low quality PDF scans gives a checklist for deciding whether a page is worth converting as it stands.

When the page still will not convert, transcribe it by hand and be honest about doubt. Put a word you are unsure of in square brackets with a question mark, and write [illegible] where you cannot read it at all. A transcript that admits three uncertain words is more useful than one that guesses silently.

Preparing a working transcript of a holographic will

A working transcript of a holographic will is a line-for-line typed copy that a person has corrected against the image, with every doubt and every alteration marked. It exists so that lawyers, family members and the court's staff can read the will easily. It is never offered as the will.

1. Capture the whole page once. Lay the will flat and scan or photograph it in color, with all four edges visible, including the back of each sheet. If you are using a phone, the post on how to OCR a photo of a document explains how to avoid glare and distortion.
2. Get a draft. Run the image through OCR and export plain text or Word. If the draft is mostly wrong, discard it and type from the image.
3. Match the lines. Break the transcript where the handwriting breaks, and number the lines. A reader can then find line 9 of the transcript on the page at once.
4. Correct one line at a time. Zoom the image in, read a line, and fix the typed line beneath it. Do not read ahead and fill in what the sentence "must" say.
5. Leave the writer's words alone. Keep the spelling, capitals, punctuation and abbreviations as written. A transcript that tidies them is no longer a transcript.
6. Mark every alteration. Use a fixed set of notes, such as [struck: ...], [inserted above line: ...], [illegible] and [Daniel?] for an uncertain reading.
7. Have a second person read the doubtful words without seeing your reading first. Record both readings where you disagree.
8. Add a header. State who prepared the transcript, when, and from which image file, and that it is a working transcript only.

A fictional example of the result:

WORKING TRANSCRIPT. NOT THE WILL.
Source image: will-2019-03-11-p1.jpg. Prepared by R.T., checked by M.O.

 1  March 11 2019
 2  This is my last will. I leave my house
 3  on Alder Street to my daughter [Carol? Carole?]
 4  and [struck: one half] [inserted above line: all] of my
 5  savings to my son Daniel.
 6  [illegible, about three words] my brother.
 7  (signed) Edith M. Varga

Lines 3, 4 and 6 are exactly the places a reader needs to go back to the image, and the transcript says so.

Trust documents to Word for restatements

Converting a scanned trust to Word gives you editable text to draft a restatement or amendment from, so nobody retypes forty pages. The trust and its amendments are typed, which suits OCR well. The risk moves to the details: a trustee's name, a percentage, a defined term.

Convert the original instrument and every amendment, each as its own file. In Schema Law's Word export each source page begins a new page, so page 12 of the Word file is page 12 of the scan while you are checking. Do that checking before you start to edit, because once paragraphs move the page match is gone.

Then build the consolidated text in a set order:

1. Start with the converted original instrument, corrected against its image.
2. Apply the first amendment, and add a comment at each change naming the amendment and section that made it.
3. Repeat for each later amendment, in date order.
4. Read the result against all the source images one more time.

Check these points in particular:

• Names and roles. Trustees, successor trustees and beneficiaries, with every spelling compared with the signed page.
• Shares. Fractions and percentages, and whether the shares in each article still add up to the whole.
• Defined terms. A capital letter lost in conversion can turn "the Trust Estate" into ordinary words.
• Numbering. Articles and sections may arrive as plain typed numbers. The post on whether OCR keeps formatting explains what survives, and the general method is in the guide to converting a scanned PDF to Word.
• Schedules. A schedule of assets is a table and deserves the same total check as an accounting.

Keep the scans of the signed instruments with the file, clearly separate from the working draft. The converted text is raw material for drafting. What the trust provided on any given date is answered by the signed documents. If the trust holds land, the deeds that funded it are covered in the post on OCR for real estate closing documents.

Estate files are private family papers. Before uploading them anywhere, read what the service does with documents. Schema Law's commitments are listed under security on the homepage.