Bi-LSTM-Based Machine Translation from Dogri to English: An In-Depth Study of Neural Approach for Low-Resource Language

Translating low-resource languages like Dogri presents significant challenges, primarily due to the limited availability of parallel corpora and the complex linguistic structure of the language. Dogri, an Indo-Aryan language spoken mainly in northern India, is one of the twenty-two scheduled languages in Indian Constitution but remains underrepresented in computational linguistics. This study presents the first experimental investigation into the application of Bidirectional Long Short-Term Memory (Bi-LSTM) networks for translating text from Dogri to English. A manually generated parallel corpus consisting of 85,000 Dogri-English sentence pairs was developed and pre-processed to train the Bi-LSTM-based neural machine translation model. The model leverages the bidirectional context of input sequences to improve translation accuracy, particularly for a low-resource language like Dogri. Experimental results demonstrate the effectiveness of the proposed approach, achieving a BLEU score of 46.03, and highlighting its potential for integrating underrepresented languages like Dogri into modern computational and technological frameworks.