Through this report we are generating an English to French translation using the seq2seq model without and with attention mechanism – a breakthrough concept that allows the model to dynamically focus on different parts of the input sequence while generating the output. Unlike a transformer layer, our goal is to implement attention as an augmentation to the existing Seq2Seq architecture, enhancing its ability to capture dependencies and nuances in the source language.

  python 3.10.9 
      tensor flow 2.13.0 

numpy
tensorflow
keras - pad_sequences
keras -Tokenizer
keras - LSTM, GRU, Dense, Embedding
Pillow
tqdm
matplotlib
scikit-learn

Initially, we choose a target language for translation, ensuring the availability of a suitable parallel dataset for training and evaluation. We have chosen the language French for this assignment. We have received the sentence pairs from “manythings.org” portal. We evaluated the performance using the seq2seq model with and without attention. We have used Glove 50 embeddings which are not included in the submission folder. The notebook is well formatted and best viewed in Google Colab

Project submitted towards the partial procurement of credits for CS747 Deep Learning at GMU under Professor Daniel Barbara