Language transition on temple inscriptions using Tesseract OCR and T5 model

Translating ancient languages like Sanskrit poses significant challenges in Natural Language Processing (NLP) due to its complex morphology, low-resource status, and script variations. This study proposes a hybrid Optical Character Recognition (OCR) and NLP-based translation system to enhance Sanskrit-to-English translation accuracy. The system integrates Tesseract OCR for high-precision text extraction and SpaCy for linguistic preprocessing, including tokenization, part-of-speech tagging, and named entity recognition (NER). To address Sanskrit’s grammatical complexity and contextual dependencies, a fine-tuned T5 Transformer model is employed for context-aware translation, text correction, and summarization. The framework is evaluated on a diverse Sanskrit corpus , demonstrating significant improvements in word recognition accuracy, BLEU scores, and overall processing efficiency compared to existing OCR-NLP models. By combining OCR with deep learning-based NLP techniques, this approach enhances translation quality while contributing to the digital preservation and accessibility of Sanskrit literature, supporting further research in computational linguistics and historical text analysis.