Best way to make corrections to transcriptions
Hi,
is it fair to assume that part of the transcription (extraction of names from images to convert it to a data in a digital structure) is done by Artificial Intelligence? If so, when I make corrections, do they help to retrain the Artificial Intelligence? are there ways to make those corrections in the most helpful way?
As an example, I am fixing some wrong dates.
Artificial Intelligence did catch the yellow boxes as relevant to dates, but wrongly assigned 1967 when it should be 1945. When I edit the highlight, should I highlight every word or can I create large boxes?
This volume and similar ones are in "Nuestra Señora de la Consolación, Toro, Valle del Cauca, Colombia records, Aug 4, 2018," images, FamilySearch (https://www.familysearch.org/ark:/61903/3:1:3QS7-L9C2-HVPC?view=explore : Jan 7, 2024), image 82 of 407; Archivo Diocesano de Cartago (Colombia).
(in the next volume, very similar in format, many other images were wrongly transcribed because the word "mil" in "mil novecientos cuarenta y ocho" was not read due to the crease. This resulted in records like these:
"Nuestra Señora de la Consolación, Toro, Valle del Cauca, Colombia records, Aug 4, 2018," images, FamilySearch (https://www.familysearch.org/ark:/61903/3:1:3QS7-89C2-HNM4?view=explore : Jan 7, 2024), image 270 of 417; Archivo Diocesano de Cartago (Colombia).
I wonder how can I give useful feedback to the artificial intelligence algorithms in charge of this volume.
Answers
-
The majority of indexing (the process of extracting names to make a document searchable by name) is done by human volunteers.
Those record sets that have been indexed by OCR are generally noted. For example - https://www.familysearch.org/ark:/61903/1:1:Q5SX-8RQM
1 -
The records seen in the Name Review activities in Get Involved are indexed by a computer. Users are invited to check the computer's work for errors. I don't have details about if our corrections are used to "teach" the computer, bur I would think so.
1
