Handwriting recognition and indexing on phone coming soon to FamilySearch. Woohoo!

LegacyUser

Michael W. McCormick, AG® said: This morning at RootsTech I went to presentation called FamilySearch record innovations. The most exciting parts for me when they said they’re going to be doing more and more with handwriting recognition and we will see more records actually pop up online that will have the tag saying they’ve been indexed by a computer. And equally exciting is the new cell phone app that’s going to help with Indexing they said will be released next year in 2021. I made a post a month ago on this forum that I was trying to talk about how excited I was for a features like this to come but I worded it in a way that kind of created some argument so I apologize for that but mostly I just want to say I’m really excited for this technology that’s coming. It sounds like we might be a couple years out from being able to index images when we’re browsing images. They actually said that that’s not going to be ready at the same time but it’s on the road map.

LegacyUser

Jeff Wiseman said: Agreed! I think it's cool that indexes for digital images will potentially be generated at a much higher rate than today, making a lot of unindexed images easily searchable.

I'm also thinking of a kind of a "side benefit" of doing this as well. Many people currently put far too much trust in the contents of index files. Since handwriting recognition is potentially less accurate than having an actual person doing the indexing, perhaps this awareness will get more folks involved in checking the accuracy of the index data instead of just plopping it down into a person record directly from the index.

So improved search ability to more images and (potentially) more scrutiny of the data coming from those images. Sounds good to me!

LegacyUser

Paul said: I still believe this will be of very limited use. If I struggle to read names - with my 30+
years experience of dealing with 17th century documents, etc. - how will a computer manage to read such handwriting?

LegacyUser

ATP said: And, even, 19th and early 20th century docs before typewriters proliferated!

LegacyUser

Juli said: It'd take some highly sophisticated linguistic AI to get a computer to read even relatively neat handwriting. I agree with Paul and ATP: I'm not holding my breath.

LegacyUser

David Newton said: Ditto. If they've really made such a breakthrough itnwill be wonderful. However given the difficulty and intractability of the problem I'm extremely skeptical.

LegacyUser

A van Helsdingen said: I believe I mentioned this on a previous Michael McCormick post. There is a program called Transkribus that at least for Dutch records appears promising. The Dutch indexing site VeleHanden currently has a project where indexers are given the program's reading of a document, and the document, and make a transcription using the computer's transcription as a guide. They are reporting high levels of accuracy. I wouldn't be so skeptical about the future of handwriting recognition.

LegacyUser

joe martel said: Thanks Michael for keeping up with the current technology events and keeping the community informed.

It's amazing what technology can bring to us. One of its strong points is a computer can sample and recall billions of pieces of information, like images, and compare, and extrapolate, and interpolate. You as a human do the same thing but on a much limited set. You learn what a cursive 'a' looks like for a certain writer and learn to distinguish that from a cursive 'o'.

I didn't think this would be possible for along time a couple years ago. But I've seen the progress made in the last year and look forward to having handwriting/pattern recognition give us more transcribed records to start with. I'm always wanting them to "test" their most recent version with some of my handwritten artifacts in other languages.

Sure there are always edge cases so a human is always the final say in FamilySearch products. With ability to browse images and user corrections becoming more available think how much more available  records are now versus a decade ago. Who would have thought...

LegacyUser

MaureenE said: September 17, 2019 21:36 FS Employee Jason Pierson made some comments about OCR in respect of Digital Books in the topic
"Better OCR needed for digital library - and / or the ability to correct the text "
https://getsatisfaction.com/familysea... including the wording

"With the volume of books going through our current system, we use a set of parameters that seems to do best across a large variety of formats, but it is not optimized for everything so results can vary from book to book".

With handwriting recognition, I image the same circumstances would apply, but even more so, as there is an immense variety in handwriting styles.

I am sure that there is technology available for handwriting recognition, but it sounds very expensive. Perhaps it is a question of how much money FamilySearch is willing to put into research.