Can you read this cursive handwriting? The National Archives wants your help
The U.S. National Archives is seeking online volunteers to transcribe handwritten historical documents, many in old-style cursive that younger people increasingly struggle to read as schools phase out cursive instruction. Commenters debate whether modern OCR and large language models are already capable of handling such material, with examples showing that while AI can do impressively well on clean, modern cursive and typeset text, it still makes subtle but important errors on harder manuscripts that require human review. Others focus on the joy and historical intimacy of transcription work itself, while noting that the project’s onboarding and tooling could be smoother if it’s to attract and retain more volunteers.
Enjoyment and value of transcription
- Several commenters say transcribing old letters and journals is deeply satisfying, giving a strong sense of closeness to the writers and their moment in time.
- People note how small details (crossed-out words, mistakes, changes of mind) make historical figures feel very immediate.
- Some see this as a great “semi-productive” hobby and mention similar projects (e.g., war memorials, genealogy, family journals).
AI/OCR vs humans: can machines do it?
- One major thread debates whether modern OCR/LLMs can handle cursive as well or better than “random humans.”
- Some argue OCR is now “very good,” demonstrate GPT‑4o accurately transcribing the sample document, and claim this is close to a solved problem for modern English cursive.
- Others counter with hard examples: medieval scripts, Old French, highly degraded or idiosyncratic handwriting, and complex archival pages where current models misread key details (names, dates, place names, “Teapot” vs “Tenorio,” etc.).
- The article’s own note that AI/OCR are used but “don’t always work” is cited both as support for skepticism and as possibly understated PR.
How to combine AI and human effort
- Several suggest a hybrid workflow: run OCR/LLMs first, then have humans verify, correct, or reconcile multiple machine outputs.
- Critics worry humans will rubber‑stamp 95%‑correct AI results and miss subtle but important errors.
- Others propose multiple independent transcriptions (human and/or machine) plus comparison, or public version control with ongoing corrections and known error rates.
Cursive literacy and education
- Many younger commenters admit they struggle to read the sample cursive; older ones often find it easy and are surprised it’s now a “rare skill.”
- There is debate over whether schools should teach cursive: some call it obsolete; others cite motor-skill benefits, Montessori practice (cursive before print), and accessibility for some dysgraphic students.
- Several note that many US schools stopped teaching cursive, and some have reintroduced it; typing is often poorly taught as well.
Handwriting, aesthetics, and difficulty
- People admire the beauty and straightness of historical handwriting, but note huge variation: some hands are elegant, others “chicken scratch.”
- Cross-writing (writing perpendicular layers on the same page) and inconsistent or phonetic spelling are highlighted as especially hard for machines and sometimes even for humans.
Project design and practicalities
- Some users find the National Archives signup and login flow frustrating (redirect loops, 2FA, hard-to-find missions).
- Others share direct links and note once in, it’s easy and fun to start transcribing.