Xiaojing Yang

My name is Xiaojing Yang. I am a master's student in Language Technology at Uppsala University with a background in Business Analytics and four years of professional experience as a data analyst. My previous work gave me a strong foundation in data analysis, machine learning, statistical modelling, and experimental design.

My research focuses on multilingual NLP, model adaptation, and reliable language technology. It began with low-resource machine translation in the petroleum domain, where I explored parameter-efficient adaptation using LoRA. I later extended this work in my master's thesis to multilingual adaptation using Mixture-of-Experts and language-specific adapters. Through these projects, I became increasingly interested not only in how models can be adapted to improve their performance, but also in how training data shape model behaviour.

My recent work investigates this question through group-level training-data attribution using Shapley values. I also work on evidence-grounded retrieval, including GraphRAG for long documents, and bias evaluation in language and vision-language models. Across these areas, my broader goal is to develop NLP systems that are adaptable, interpretable, and reliable, together with methods for systematically auditing and understanding model behaviour.