Few outside Maharashtra may have heard of the Modi script, which was used in administrative and legal documents between the 13th and early-20th centuries. About 40 million handwritten documents are scattered across places, including government archives, the Institute of Oriental Studies, and the Bharat Itihas Sansodhan Mandal, Pune.
On first glance, the writings in them appear to be gibberish. But who knows what knowledge resides within them?
Three researchers from the College of Engineering, Pune — Harshal Kausadikar, Tanvi Kale and Onkar Susladkar — and one from IIT-Roorkee — Sparsh Mittal — decided to find a way to decipher them. They built a computer program, which they call MoScNet, a novel vision-language model (VLM) framework for transliterating Modi script images into Devanagari text. “MoScNet leverages ‘knowledge distillation’, where a ‘student model’ learns from a ‘teacher model’ to enhance transliteration performance,” the authors say in their yet-to-be-peer-reviewed paper.
With this, the researchers have built a collection of 2,043 images of Modi script documents with their transliterations in Devanagari. “Our work is the first to perform direct transliteration from the handwritten Modi script to the Devanagari script,” their paper says.
This was not easy because the Modi script changed six times during its lifetime. There is an era-wise classification of the script, such as Aadyakalin, Shivakalin, and so on. Further, the script’s cursive nature, differing writing styles, and issues like angular strokes, broken lines and blurring make the task of deciphering them even more difficult. However, it would be interesting to learn what the documents say because the texts deal with medieval sciences, medicine, land records and Indian history. Now, we have a tool to bring to life the 40 million documents that speak to us from the distant past.