Text Recognition for Nepalese Manuscripts in Pracalit Script

Abstract

This dataset is a model for handwritten text recognition (HTR) of Sanskrit and Newar Nepalese manuscripts in Pracalit script. This paper introduces the state of the field in Newar literature, Newar manuscripts, and HTR engines. It explains our methodology for developing the requisite ground truth consisting of manuscript images and corresponding transcriptions, training our model with a PyLAia engine, and this model’s limitations. This dataset shared on Zenodo can be used by anyone working with manuscripts in Pracalit script, which will benefit the fields of Indology and Newar studies, as well as historical and linguistic analysis.