Reproduction studies in machine translation and NLP

Supervisor: Prof. Dr. Rico Sennrich

In this project, we will together pick a recently published paper that shows great potential, but also leaves open research questions that you can answer in this project. A starting point will be confirming the results of the research paper in a reproduction study (this can be a small or larger part of the project, depending on the resources (code/data) provided, but your goal is to gain further insights into the method described in the paper: will the method generalize to other settings (e.g. languages with varying amounts of resources)? Can the method be refined to make it more efficient or effective? Does an error analysis uncover interesting new types of errors that a baseline system does not produce? The exact research questions depend on the paper chosen and your interests and skills.

• Background in machine learning
• Python