Zhou, Ej

Spelt Yijie

About

I am doing my third year of a PhD at the Language Technology Lab, University of Cambridge, advised by Prof. Anna Korhonen. I am working with the Qwen Pretrain/Multilingual team as a research intern at the moment.

I work on multilingual NLP: getting LLMs to work well, and fairly, across languages. My interests span interpretability, alignment, generalisation, tokenisation, bias and calibration.

Language et al.

Wu is and Mandarin are my muttersprache. Apart from that I speak some French, Japanese, Russian, German (B1, Aug 2026) to varying degrees. I am learning Polish and Portuguese this year. I am passionate about languages and it has always been a motivation for me to work on NLP. I have ~1000 days on Duolingo.

Research Interests

Language Modelling Across Languages
Representation
The Strong Platonic HypothesisUniversal Convergence of Representations Across Languages Without Joint Training(EMNLP ’26)
Multilinguality is MultimodalityLanguages are modalities on a spectrum.(on arXiv soon)
Behaviour
Bias Beyond EnglishDebiasing Transfer(NLPCC ’25)
Low-Resource LeniencyUnreliable LLM Judges across languages(COLM ’26)
Beyond the Final LayerLanguage-agnostic representations with better calibration(EMNLP ’26)↖ builds on Cross-Lingual Alignment
Data
ConvSumXCross-lingual summarization(ACL ’23)
MSRSMulti-source RAG(COLM ’25)
HealthDialMultilingual spoken dialogue(ACL ’26 Findings)
गते गते पारगते पारसंगते बोधि स्वाहा