README

Abstract

This project aims at achieving English to French translation with a focus on deep learning methods, beginning with Recurrent Neural Network (RNN) and advancing to Sequence-to-Sequence (Seq2Seq) models, encompassing transformers. By capitalizing on the extensive bilingual content from the YouTube Video Subtitles API, the initiative employs a thorough multi-stage preprocessing regimen post-data extraction, which includes data cleansing, tokenization, and semantic enrichment through Word2Vec, coupled with VecMap for semantic space alignment across languages. This careful, iterative approach involves rolling out diverse models, finetuning pre-existing models, and conducting comparative assessments to enhance translation precision. By melding cutting-edge neural network frameworks with complex natural language processing tactics, the project aspires to craft highly accurate translation models.