wxy829475 / Neural-Machine-Translation

In this project we build neural machine translation systems for two parallel corpus, Chinese-English and Vietnamese-English. There are four different model architectures we evaluated in this project.

NLP_final_project

DS1011-NLP Final Project Neural Machine Translation

1. Recurrent neural network based encoder-decoder without attention
   • vi-en: model_1_simple_vi_en.ipynb
   • zh-en: model_1_simple_zh_en.ipynb
2. Recurrent neural network based encoder-decoder with attention

   • (a). dot_product based attention

   • vi-en: dotpro_atten_vi_en.ipynb

   • zh-en: dotpro_atten_zh_en.ipynb

   • (b). neural net based attention

   • vi-en: neural_attent_vi_en.ipynb

   • zh-en: neural_attent_zh_en.ipynb

3. Replace the recurrent encoder with convolutional based encoder
   • Convolution based dot product attention: conv_dot_product_attention.ipynb
4. Transformer, fully self-attention translation system
   • Transformer.ipynb
5. Preprocessing:
   • Preprocessing.ipynb: for dot_product based attention model, Convolution based dot product attention model and Transformer