Synthetic and real data improve Manchu OCR accuracy
Combining Synthetic and Real Data for Low-Resource Historical OCR: A Manchu Case Study
This research provides a practical method to improve OCR accuracy for endangered languages like Manchu, which is relevant for anyone working with historical documents or low-resource languages.
This study found that combining synthetic and real historical training data can significantly improve the accuracy of optical character recognition (OCR) for low-resource languages like Manchu.
Using 60,000 synthetic and 20,306 real historical word images, the research evaluated three vision-language models (VLMs) and a compact CRNN under four training regimes.
Introducing real training images raised the leading configurations' word accuracy to between 95.09% and 96.
28%, while no synthetic-only configuration exceeded 87.92%.
A compact CRNN also reached the leading performance range once real images were available, showing that model scale alone does not determine recognition accuracy.
Combining Synthetic and Real Data for Low-Resource Historical OCR: A Manchu Case Study
Manchu, now critically endangered, was one of the principal languages of the Qing empire (1636-1912), and its extensive archival record is increasingly digitized but remains difficult to search and analyze at scale. Previous work showed that vision-language models (VLMs) trained only on synthetic Manchu word images can reach 87.4% word accuracy on real Qing manuscripts and prints, leaving a substantial synthetic-to-real gap. This study examines how synthetic and real historical training data should be combined for low-resource OCR. Using 60,000 synthetic and 20,306 real historical word images, we evaluate three pretrained VLMs and a compact convolutional recurrent neural network (CRNN) under four regimes: synthetic-only, real-only, joint synthetic-real, and sequential synthetic-to-real training, following a common checkpoint-selection and archival evaluation protocol. Introducing real training images raises the leading configurations to between 95.09% and 96.28% word accuracy, while no synthetic-only configuration exceeds 87.92%. Synthetic supplementation substantially improves all three VLMs, whereas its marginal effect for the CRNN is sensitive to the training objective. Joint and sequential training yield broadly similar archival accuracy under the tested practical pipelines. A compact CRNN also reaches the leading performance range once real images are available, showing that model scale alone does not determine recognition accuracy. Finally, complementary errors among strong recognizers allow voting to raise accuracy to 98.27% without additional training, while an eighteenth-century Manchu dictionary provides a principled rule for adjudicating disagreements.
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include OCR, Manchu, low-resource, synthetic data.
Source attribution: curated from arXiv cs.LG (@Yan Hon Michael Chung, Hanlin Wang).
