Language Translation using deep learning

Language Translation using deep learning

ABSTRACT

Machine translation has advanced significantly with deep learning, enabling accurate and context-aware translations. This project, "Language Translation using Deep Learning," explores different approaches for English to Hindi, English to Chinese, and English to French translations using LSTM-based Seq2Seq models, Transformer models, and MarianMT models. Additionally, a MarianMT + BERT model is implemented for enhanced context-aware translation. The project utilizes datasets from TED Talks, HindEncorp, and French-English corpora, with extensive preprocessing, including text normalization, tokenization, and sequence padding. The LSTM models use an encoder-decoder structure, Transformer models leverage self-attention mechanisms, and MarianMT models provide efficient and high-quality translations with fine-tuning. The MarianMT + BERT model integrates contextual embeddings, improving fluency and meaning retention. Evaluation metrics such as BLEU scores and accuracy demonstrate that Transformer and models outperform LSTMs, with the MarianMT + BERT model achieving 92% accuracy. The project highlights the benefits of self-attention, fine-tuning, and context-aware learning, paving the way for further advancements in AI-driven multilingual communication. Future work includes expanding datasets, optimizing for low-resource languages, and integrating reinforcement learning for adaptive translation.