“To Write a Good Hand Is a fine Accomplishment“

I don’t think Boston 1775 would exist without Google Books, which launched the year before I began blogging.

Being able to search for phrases and names in the vast corpus of nineteenth-century books that Google scanned has let me identify early sources, track local lore, and find rare sources.

That search depends on optical character recognition (O.C.R.) translating the printed marks into digital text.

I’ve learned how to unearth that digital text from Google Books pages for long quotations. At first this involved some tricks to maneuver to the pages I wanted. Then Google reprogrammed that site, so it became easier to capture particular text—but without any punctuation marks.

Either way, I learned to go over the O.C.R. transcription carefully to make sure it was complete and accurate. There are still a few bugs in the system. But it’s good to stay on my toes with historical sources.

Handwritten Text Recognition (H.T.R.) is the next step in this process, converting cursive writing into digital text. The University of Innsbruck’s Transkribus platform was the early leader in this field. The complexity of our writing means people started applying artificial intelligence (A.I.) software to the problem.

Indeed, some historians have said that H.T.R. is now the most advanced and reliable product of A.I. for the field, especially as it’s improved in the last couple of years. 

In September 2025, Civil War History published a study by Prof. Julie Mujic of Denison University on how ChatGPT transcribed a handwritten poem from 1864, finding many discrepancies and problems. Then in January 2026, Prof. Cameron Blevins of Colorado University Denver revisited the same text with the most recent version of Google’s Gemini 3.0, finding it produced a complete and reliable transcription.

Likewise, in November 2025 Dan Cohen, dean and professor at Northeastern, found that the Gemini Pro 3 program did a very good job transcribing some difficult letters. Even letters with cross-writing, that bane of nineteenth-century correspondence. Cohen concluded:

It allows human beings to focus their time on the important, profound work of understanding another human being, rather than staring at a curlicue to grasp if it’s an L or an I.

There’s still room for improvement, as Prof. Mark Humphries of Wilfrid Laurier University discussed in May 2026, but this application of A.I. will probably continue to get better. As with a lot of technology, one big question will be whether proprietors will be able to resist adding features that actually reduce its practicality for many users.

The transcriptions of handwritten texts that I’ve worked with the most come from the Thomas Gage Papers at the Clements Library. That site states: “The plain text we provide for this collection was created using Transkribus, a tool that uses machine learning and artificial intelligence to transcribe handwritten documents.” How do the results hold up?

TOMORROW: Reading the general’s mail.