Three papers from the Health SciTech Group have been accepted at EMNLP 2026, one of the leading international conferences in natural language processing.

• PsPLUG: Do Implicit Personalization and Explicit Styles Conflict? A Lightweight Plug-in for Balancing Personalization and Style in Customized LLMs. Led by Yutong Song, this work addresses “personalization collapse” by helping language models follow style instructions without losing a user’s individual voice. Read the paper.
• AP-GRPO: Anchor-Gated Phonetic Alignment with Policy Optimization for Pathological Speech Reconstruction. Led by Pengfei Zhang, this approach improves faithful reconstruction of speech affected by neurological and neuromotor conditions. Read the paper.
• CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation. Led by Yutong Song with collaborators at TikTok, CARD offers scalable personalization through cluster-level adapters and compact user preference vectors. Read the paper.

Congratulations to our students and collaborators on this outstanding achievement.