Low-Resource Machine Translation with Large Language Models

Supervisor: Prof. Dr. Rico Sennrich

Large Language Models have been shown to be competitive with dedicated translation systems for some high-resource languages. For low-resource translation directions, they typically lag behind pure machine translation systems. In this project, you will first analyze the translation performance of LLMs on selected low-resource translation directions, and aim to gain insights into the typical errors that stand in the way of better translation quality. You will then explore if these errors can be addressed with existing or new decoding techniques, ideally without requiring costly training of LLMs.

• Background in machine learning
• Python