This is a summary of my research project to digitize and OCR a historical document written in Italian in cursive handwriting in the style of 1900.
Background
My uncle Aldo Bechis (1880-1978) was a career naval officer in the Italian Navy and retired as an Admiral.
My cousin, Toni Mattarucco (1940-2009), orphaned at birth (father died in the war in Yugoslavia in 1944), was raised by Uncle Aldo as a son. He had his own short naval officer career at the Morosini College Naval Military School in 1961-63, and then the Naval Academy, and as a Lieutenant until 1969.
I followed Toni’s steps at the Naval College (1967-1970), but a broken neck in 1969 ended my naval experience in June 1970, and I emigrated to the US shortly thereafter.
Uncle Aldo, as a Lieutenant, circumnavigated the globe by steamship (Royal Navy Ship Calabria) from Venice to Japan in 1909-1911 to deliver the first diplomatic respects to the Emperor of Japan on the occasion of Japan’s opening the kingdom to the West.
During his trip, Uncle Aldo maintained a detailed handwritten log of his travels. It is a unique view of the world by a naval officer of the Italian Royal Navy, as he discovered the world and reported on details of military importance and human curiosity.
The original Log Book was lost over the years, but Toni had made a photocopy of it (before Xerox) for my education and memories.
The OCR Project
Over the intervening 60 years I never found it convenient to read the Log Book, written in Italian cursive writing in the style of 1900, quite different from the cursive I learned to read and write in Italy in my elementary school years. I repeatedly tried digitizing and OCR some pages without success.
In 2025 I digitized pages of the photocopied manuscript and used ChatGPT to OCR . This was yet another attempt following repeated failures with various software and services during 2020-2024. The particular style of handwriting had been an immovable rock.
ChatGPT was amazing. But only intermittently. Some days, some pages would yield fast, spotless OCR transcriptions; other days, unpredictably, could not even try for lack of a software module called ita.traineddata by Tesseract (open software on GitHub). Repeated attempts to get ChatGPT Admins to load the module, even temporarily and on a schedule of their choosing, went nowhere. Only about 200 pages were OCRd out of about 800, then the module was never found again loaded in the environment. ChatGPT, too, was a dud despite its amazing capabilities when enabled.
I installed the module in my laptop and attempted to run locally with Python scripts. That failed because ChatGPT also included extensive image filtering and improvements in its processing pipeline that I never managed to match. Strike out.
The search for a new tool started again, and I came across handwritingocr.com thanks to a Reddit referral. Awesome.
The transcription quality was on par with ChatGPT (although not free), but required fewer steps to get it done. The UI was well designed, and the founder, Matt, was incredibly responsive to my suggestions for UI improvements. Really amazing customer support.
HandwritingOCR.com was a godsend and worth every dime I paid for it. BRAVO Matt.
Conclusion:
I got the document completely OCRd in record time
I downloaded images and transcribed text as a PDF and a MS Word docx
The Word doc is nicely visible 2-page-up with the image on the left page and the OCR text on the right. HOWEVER, Word’s inability to stop the continuous word counting makes editing a single large document hopeless. Strike out.
Moving to Google Docs allows multiple editors to collaborate to verify and clean up minor errors, but it loses the 2-page-up convenience.
The project is continuing
A first edition of a few pages is below.
In due course, I will use ChatGPT to translate and post the whole document in English. I already tested this, and it does an amazing job. Despite my native Italian knowledge, it is the way to go in the age of AI. It translates as well me but it writes at light speed.
https://drive.google.com/file/d/1TeZkqDRks-fztAxnQmCdHHKx2Kjec4V6/view?usp=sharing
