Researchers have developed a method to improve transcription accuracy in complex historical Sanskrit manuscripts. They use an iterative fine-tuning process on the text layout and appearance, which reduces human annotation effort required for digitization. The team tested their approach on three manuscripts and introduced a dataset with granular annotations. Their results show improved performance compared to leading multi-modal language models. --- Why it matters: This matters because it enables more accurate and efficient digitization of historical texts, making them easier to study and preserve. By reducing the need for human annotation, this method can save time and resources for researchers and scholars working with these manuscripts. Source: https://arxiv.org/abs/2608.18696