Projects
Fine-Tuning Generative LLMs for Cantonese-English Translation

In this work, we will explore the usage of generative AI models for machine translation and how to further improve its performance by fine-tuning. The generative model chosen for fine-tuning is OpenAI’s gpt-4o. A parallel corpus of Cantonese-English is created and used in the fine-tuning of gpt-4o. This fine-tuned model is compared with its own baseline model and another popular commercial translator, Google Translate, to determine its performance quality. By using sacreBLEU – an automatic evaluation metric – and human evaluation, models can be analysed, and patterns and behaviour can be recognised.
http://www.linkedin.com/in/denise-poon-721148255
Type of employment sought
computer science / software engineering
Areas available to work
London / UK / Hong Kong
Contact Yin Ching Denise