MUMBAI, India, July 30 -- Intellectual Property India has published a patent application (202641088088 A) filed by Vellore Institute Of Technology on July 19, 2026, for System And Method For Morphology-Driven Sanskrit Language Processing Using Hybrid Sandhi Resolution.
Inventors include Sanjiban Sekhar Roy; and Abhishek Kumar Pandey.
The application for the patent was published on July 24, 2026, under issue no. 30/2026.
Abstract: A morphology-driven Sanskrit language processing system and method are disclosed. The system comprises a multi-script normalization module, a morphology-aware tokenizer, a hybrid Sandhi resolution module combining a finite-state transducer with a neural disambiguator, a recursive Samasa decomposition module, a transformer language model, a retrieval-augmented generation engine, and a Vedic knowledge base. Sanskrit text received in multiple scripts or transliteration formats is normalized, segmented at morphological boundaries, resolved for Sandhi, and recursively decomposed into constituent compounds before transformer-based encoding. Retrieved knowledge passages are incorporated to generate contextually grounded responses with citation attribution. The system further supports question answering, part-of-speech tagging, named entity recognition, and transliteration through shared task-specific output heads. The language model is fine-tuned using QLoRA with 4- bit NF4 quantization, enabling offline execution on resource-constrained mobile devices with reduced memory consumption while improving linguistic accuracy and semantic understanding
Disclaimer: Curated by HT Syndication.
