yumiko-55/lost-in-translation-nmt

Improving Neural Machine Translation for low-resource languages using Attention Mechanisms. # Lost in Translation ## Overview This project investigates how different Attention Mechanisms affect Neural Machine Translation performance on low-resource datasets. ## Technologies - Python - Linux (Ubuntu) - OpenNMT - Hugging Face - PyTorch ## Methodology - Fine-tuned transformer-based translation models - Compared: - Scaled Dot Product Attention - Multi-Head Attention - Cross Attention - Evaluated models using: - BLEU Score - Perplexity (PLP) ## Results Cross Attention achieved the best translation performance and improved translation accuracy by 93% compared to baseline methods. ## Files - Final Poster - Figures - Project Summary ## Note The original source code is no longer available, but the project methodology and results are included in this repository.