Neural Machine Translation - Coverage and Attention

Models

Using implementations of following paper make 4 translation models and check their performance on test data

• Sequence to Sequence Learning with Neural Networks
• Neural Machine Translation By Jointly Learning To Align And Translate Basic attention
• Effective Approaches to Attention-based Neural Machine Translation Try the three global attention options, ignore local attention
• Modeling Coverage for Neural Machine Translation Does this help for repeated words ?

Dataset

Enclosed are parallel corpus

• English Bangla Gujurati Hindi Konkani Malayalam Marathi Punjabi Tamil Telugu
• Kindly choose a pair (En-IL; where IL is one of (Bangla Gujurati Hindi Konkani Malayalam Marathi Punjabi Tamil Telugu)).
• Keep in mind while choosing a pair that you have to also evaluate the performance by going through test data.
• Each pair has 3 sets of data- train, dev and test.

Output

• Keep your code and model (since model’s size will be big) on drive and share the link.
• Test input and output (can also be shared via drive)

• Analyse of 50 interesting sentences.

• The trained models for the concepts covered here can be found
  here.