Machine Translation with Recurrent Neural Networks

Project Type: Master/Diploma Thesis
Project Status: Open


Short Description  

Machine translation is a hard machine learning task due to additional difficulties compared to simple classification (such as image classification) or speech recognition. Long-range dependencies between words in source and target sentences and different sentence lengths to mention some. Furthermore, short- and long-range re-orderings of words or blocks of words makes the task even more challenging. Finally, the vocabulary size (the input and output dimensions) can be huge. 
Fortunately,  recent advances in GPU computing enable to train even complex neural networks for machine translation tasks [1,2]. These 
neural translation models are extremely successful in terms of performance. Your task will be to reproduce the state-of-the-art work, extend already existing software to perform machine translation and to compare to the mentioned reference. Neural machine translation currently a hot and interesting topic and holds the capability for great publications.

Your Tasks 

  • Redroduce and compare to state-of-the-art neural translation models
  • Extend existing models in Python and Theano (libary for GPU computing)
  • Analyze the implemented systems in terms of accuracy and computational performance

Your Profile 

  • Very good theoretical and mathmatical background (mandatory) 
  • Good knowledge in machine learning
  • Very good knowledge and experience in Python programming (mandatory)

Additional Information

As this work combines theoretical and experimental aspects of non-standard methods, a very good mathmatical and programming background is mandatory. This thesis project is planned for a duration of 6 months starting immediately. It has a good chance for publications.


Martin Ratajczak ( or +43 (316) 873 - 4379)


[1] D. Bahdanau, K. Cho, andY. Bengio. Neuralmachine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473, 2014

[2] I. Sutskever, O. Vinyals, Q. V. Le. Sequence to Sequence Learning with Neural Networks, 2014