Hacker News Reader

Comment by clasicallyabbyy | Hacker News Reader

Would be interesting to know if they’ve gotten into OCR solutions at all on their collection of documents. Worked for a company back in the day that sourced historical, public domain books from JSTOR, Google Books, etc. We used Abbyy to run OCR across all these old texts to provide highlighting in-app, text search, etc.

Took some tuning but we had good quality output across a variety of (mostly European) languages.

View on HN

They do OCR when it is feasible.

An impressive number of typewritten documents from World War 2 were OCR'ed.

Volunteers do manual data entry and verification for documents that do not OCR well (called "indexing"): https://www.familysearch.org/indexing/projects

It's probably worth noting that some of those indexing efforts make more harm than good, there's a large chunk of Polish records which were "indexed" by people who clearly have absolutely zero knowledge of Polish, and the results are SIGNIFICANTLY worse than any OCR could have been. Always compare to source material to avoid surprises like that.

That's good to know. Even Google's OCR in English is pretty spotty, so I can just imagine what it does with other languages.

These 1950 census records were written in cursive. Quite readable to a human; not so much to a computer.

I should mention that the 1950 census DOES have a text search facility. It did not work for my family.

FamilySearch has a mobile app called "Get Involved" that does OCR and just has you confirm that it did it correctly. I used to do a lot of manual indexing for them when it was mostly based on sharing physical copies of documents, but I really struggled to read anything once it switched to mobile. Now the OCR is pretty good and just once in a while needs a human to recognize some weirder handwriting using cultural clues.

Reply on news.ycombinator.com