Conversion of Handwriting to Text: Choosing a Route

Conversion of handwriting to text happens by one of four routes: you write on a screen and the device converts the ink, you scan paper and software reads the image, someone in your office types the page, or an outside service types it. The right route depends on where the handwriting is now and what the text is for.

What are the routes for conversion of handwriting to text?

There are four routes for conversion of handwriting to text. Two rely on software: converting ink written on a screen, and recognizing a scan or photo of paper. Two rely on people: a typist inside the firm and a transcription service outside it. Every handwritten page in a file fits one of the four, or needs none.

Each route starts from a different thing and ends with a different kind of result.

People searching for the best way to convert handwriting to text may be hoping for a single winner. There is none, because the four routes are not competing for the same pages. Live ink is only available for writing that has not happened yet. A typist is slow and can say where a word was unclear. Software is quick and varies from one hand to the next. A service costs money and adds a confidentiality question.

So the useful question is narrower: which route for this page? Four things answer it: where the handwriting sits, what kind of document it is, who does the reading, and whether the page deserves converting in the first place.

Routes can be combined. A scanned-paper draft that a paralegal then corrects is route two followed by a light version of route three. A service transcript that an attorney proofreads against the image is route four followed by a check. What matters is that you choose the combination on purpose and can say afterwards how a given typed page was made.

Handwriting written on a screen versus handwriting already on paper

Handwriting written on a screen can be converted by the device that captured it, because the device recorded the pen's movement. Handwriting already on paper has to become an image first, and software must work from the picture alone. Which of the two you have in hand is the first fork in choosing a route.

The fork is simple to state. If the writing does not exist yet, you can choose to make it on a screen. If it already exists on paper, that choice is gone.

Writing that has not happened yet

Your own notes of next week's client meeting fall here. You could take them with a stylus on a tablet, in an app that turns ink into typed text. Before you rely on that, find out three things from the app's own documentation:

• whether it converts handwriting into editable text or only makes the ink searchable
• which languages it recognizes in handwriting
• where the notebook is stored and whether conversion happens on the device

The different kinds of handwriting to text app are compared in their own post, including pen-input keyboards and note apps.

Writing that is already on paper

Everything a client, a witness or an opposing party hands you falls here, along with your own old pads. Nobody recorded the pen's movement, so a tablet's ink conversion is of no use. The page must be scanned or photographed, and from then on your options are routes two, three and four.

Why the two kinds of input behave differently is a technical subject. The post on handwritten text recognition explains how reading pen strokes differs from reading a page image. For choosing a route, the consequence is enough: do not judge how software will handle a scanned letter by how well your tablet converts your own stylus writing. They are separate tests on separate inputs.

A page that is both

Some pages mix the two. You might annotate a scanned PDF on a tablet, or print a typed draft and mark it up in pen. Treat each layer by its own origin. Stylus annotations made in an app follow the live-ink route inside that app. Pen marks on a printout are paper handwriting once the page is scanned.

Which handwriting conversion route fits which legal document?

The handwriting conversion route that fits a legal document depends on three things: how much rides on each word, how many pages there are, and whether the writer is available to ask. High stakes and few pages point to a person. Many pages and a search-only purpose point to software.

Run each document through those three questions before you pick.

How much rides on each word? A handwritten statement you will quote in a motion needs every word right. A box of a decedent's notebooks that you only need to search for a name does not. For the first, plan on a person reading every word, whatever produced the first draft. For the second, an uncorrected software draft may be enough as a finding aid, so long as nobody quotes from it.

How many pages? Three pages can simply be typed. Three hundred pages are a project, and that is where a recognition draft or a service earns consideration.

Can you ask the writer? If the writer is your client or a colleague, unclear words can be settled by a phone call. If the writer is dead, adverse or unknown, unclear words stay unclear and the transcript has to say so.

Here is how some familiar documents line up. Treat the route column as a starting point to test, not a rule.

Two of these rows connect to posts that go further. Handwritten wills carry their own cautions, covered in the post on OCR for a handwritten will. Pages of figures are risky under any route, for reasons set out in the post on why OCR gets numbers wrong.

Before you quote a handwritten last will and testament, compare the typed line with the page for the decedent's name, the date, each bequest, and every signature. Give a codicil clipped to it the same check. A signed statement you may quote in a motion needs the witness's name, the date, any amount, and the signature read against the image in the same way.

Once a paper page has been assigned to the scanned-paper route, the method is the same each time. The guide to converting handwriting has a five-step procedure for turning a handwritten page into typed text, from capture through export.

Recognition software, a note app, a typist or a service: who does the reading?

In every route somebody or something reads the handwriting, and that reader's strengths set the limits of the result. Recognition software and note apps read fast and cannot be questioned. A typist and a service read slowly and can tell you where they were unsure. Choose the reader, and you have chosen the kind of error to expect.

Recognition software

You give it a scan or a photo and it returns text. It does not tire, and it treats the first page and the five-hundredth alike. Its weakness is that a misread word arrives looking the same as a correct one. Schema Law's homepage, in its questions section, says recognition is strongest on clean print and weaker on handwriting, and that nothing marks a passage as uncertain. Assume the same of any tool you have not tested, and budget time for a person to compare the draft with the image.

A note app

A note app reads ink that was written inside it. It is a reader only for the live-ink route. Its documentation is the place to learn what it converts and in which languages. It cannot help with a letter that arrived in the post, unless the same app also has a separate feature for reading photographed pages, which is a different function to test separately.

A typist in the office

A person who knows the matter reads with context. A paralegal who has worked the file knows the parties' names, the street addresses and the dates in dispute, and will catch a word that software would get plausibly wrong. A human reader can also mark doubt, writing "[illegible]" or "[Castell? Castillo?]" where the page is unclear. The cost is time, and the risk is the reverse of software's: a reader who knows the story may see the word they expect. A second reader reduces that risk.

An outside service

A transcription service supplies the typist when the firm cannot. Before sending pages out, settle the points that have nothing to do with handwriting: what the service does with the images, who there can see them, and whether your duties to the client allow the disclosure. The post on whether it is safe to upload client documents to an OCR tool lists questions that apply equally to a transcription vendor. The comparison of handwriting tools also covers when a transcription service beats any handwriting software.

Whoever reads, record it. A line at the top of the typed file saying "software draft, corrected by M.V." or "typed from the image by D.K., checked by R.O." tells the next person how far to trust the text.

Which handwritten pages are not worth converting at all?

Handwritten pages are not worth converting when nobody will search, quote or edit their words. A page that only needs to be seen, counted or produced can stay an image with a good file name. Converting it costs time and adds a second version that someone might later mistake for the original.

Ask what the typed text would be used for. If you cannot name a use, skip the conversion. These are candidates for leaving alone:

• Signatures and initials. The mark matters as a mark. Typed text of a signature tells nobody anything the image does not.
• Sketches, maps and diagrams. A hand-drawn floor plan with three labels is a picture. Note in the file name what it shows.
• Duplicates. If the same handwritten page appears four times in a production, convert one copy at most and record where the others are.
• Envelopes, sticky notes and routing slips with a word or two on them. A descriptive file name does the job of a transcript.
• Pages where the handwriting is the issue. When authorship or alteration is disputed, the image and the paper are what count. A typed version can help you discuss the content but settles nothing about who wrote it.
• Pages nobody can read. If two careful readers cannot make out a page, a typed version would be mostly brackets. Record that it is illegible and move on, or ask the writer.

Leaving a page as an image is still a decision, and it should be visible. Three small steps make it so:

1. Name the file so it can be found without opening it: date, writer if known, and a few words on the content.
2. Add the page to whatever index the matter has, marked "image only, not transcribed".
3. Revisit the decision if the page becomes important. Converting one page later is cheap.

There is a middle option between converting and ignoring. You can run a batch of low-value handwritten pages through recognition and keep the uncorrected draft purely as a search aid, labeled as uncorrected. A search hit sends you to the image, and you read the image. The danger is a false sense of coverage: a name that recognition misread will not be found, so a search with no hits does not prove the name is absent.

A decision table for the handwritten pages in one matter

A decision table lists every handwritten item in a matter with its purpose, its volume, whether the writer can be asked, and the route chosen. Filling one in replaces a string of page-by-page choices with one plan that others on the file can follow.

Here is a worked example with invented facts. The firm acts for a homeowner, Ms. Pruitt, in a dispute with a builder, Castellano Renovations, over extra charges. The file holds six kinds of handwriting.

Several things in this table are worth copying.

The same matter uses every route but one. No single tool was chosen for the file. The outside service was considered for the notebook and set aside, because the firm preferred to keep an opposing party's production in house. Another firm might decide differently.

Purpose drove the choice more than legibility. The foreman's notebook and Ms. Pruitt's timeline might be equally hard to read. One is searched and the other is quoted, so they get different treatment.

Every row names a check. A route without a check is half a plan. Where the check is "none", that is written down too.

The table is a file document. Save it in the matter. If a typed page is later questioned, the table shows how it was made and who looked at it.

If you would like to see what a recognition draft of your own pages looks like before settling the software rows of such a table, Schema Law's free plan takes one document a day of up to three pages and, like every plan, exports all eight formats, Word, searchable PDF and plain text among them. The pricing page has the page allowances for larger files. Use a page you are permitted to upload, compare the result with the image line by line, and let that comparison decide whether the scanned-paper route belongs in your table. If your own notes are the pages in question, the post on getting handwritten notes converted to text covers turning them into a file memo.