About this document
2023 Emnlp-Main 303 by sylasree K Sheshadri is a document available to read on EtoBox.
The paper presents MMNMT, a modularized framework for multilingual neural machine translation (MNMT) that combines dense and Mixture-of-Experts (MoE) models to enhance translation quality across various language resources. It introduces a three-stage training strategy, including pre-training, module initialization, and fine-tuning, to effectively leverage the strengths of both model types while mitigating overfitting issues seen in low-resource translations. Experimental results demonstrate that MMNMT signi
- Author
- sylasree K Sheshadri
- Language
- EN