A pipeline combining Google Vision OCR, fastText language ID, masking, and neural post-correction reduces character error rate on Kwak'wala legacy books from 0.43 to 0.18 and from 0.33 to 0.15.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Developing a Mixed-Methods Pipeline for Community-Oriented Digitization of Kwak'wala Legacy Texts
A pipeline combining Google Vision OCR, fastText language ID, masking, and neural post-correction reduces character error rate on Kwak'wala legacy books from 0.43 to 0.18 and from 0.33 to 0.15.