Comparative Study of LSTM-Based and Generalized Neural Machine Translation Systems for English-To-Sanskrit

Main Article Content

Neha Vaswani
Krupa Mehta

Abstract

The study of this research paper designs and also develops and then evaluates an LSTM-based English-to-Sanskrit translation system with Google Translation System. Automated translation faces meaningful challenges since Sanskrit is a highly inflected classical language with rich grammatical structures. Due to their ability to model longterm dependencies, Long Short-Term Memory (LSTM) networks are used in the proposed system which implements a character-level encoder-decoder architecture with optional attention mechanisms. For the deployment of the model that was trained on 2,000 English-Sanskrit word pairs a Tkinter-based offline user interface was used. Concerning linguistic precision and grammatical accuracy, the model beats Google Translate. Comparative analysis also shows better handling of Sanskrit morphological features. The system offers advantages with customization, interpretability, and offline usability. Because of this situation, the system is quite suitable for educational plus research contexts. The results do validate that LSTMbased architectures can effectively translate low-resource languages and do advance Sanskrit digital processing. The goal of this research is to build an LSTM-based translation system to scan existing tools and frame English to Sanskrit translation. The research aims to create translations that are contextually accurate, grammatically correct, and linguistically respectful to Sanskrit by utilizing the strengths of LSTM networks in handling long term dependencies and complex sequences.

Article Details

Section
Articles