Title: Improving Language Model Through Context

Speaker: Dr. Dong Ho Lee @ Microsoft AI

Time : 10:30 - 11:30, Sep 24th, 2025

Location: Online

https://hli.skku.edu/InvitedTalk250924 

Language: English speech & English slides

Abstract: 

Contextual cues are important in recent LM research, enabling models to reason effectively, handle complex tasks, and exhibit social intelligence through context-aware interactions.

My research proposes foundational groundwork for the systematic study and practical incorporation of multiple contextual information into LM.

Specifically, I address three key research questions: (1) Can LMs effectively learn from context during inference?; (2) Does adding context during training enhance model behavior?; (3) Can LMs dynamically generate and refine context to improve its output quality?

To explore these questions, I explore a variety of contextual cues including (a) human-provided explanations [TriggerNER (ACL 2020), AutoTriggER (EACL 2023), LEAN-LIFE (ACL 2020), XMD (ACL 2023)]; (b) in-context examples [FewNER (ACL 2022), LLM-Data-Creation (EMNLP 2023), TKG-LLM (EMNLP 2023)]; (c) dialogue context [Normvio-RT (EMNLP 2023), LoCoMo (ACL 2024), REALTALK (2025)]; and (d) model-generated context [QUEST(2025)].

Bio:

Dong-Ho Lee is a member of technical staff at Microsoft AI, where he works on enhancing Copilot with more personalized, emotionally aware, and socially intelligent experiences.

He received his PhD in Computer Science at the University of Southern California (USC) and USC Information Sciences Institute (ISI).

His research focuses on enhancing contextual reasoning in LLMs and developing socially intelligent AI systems that drive real-world impact.

He has presented his work at AI and NLP conferences, including ACL, EMNLP, and NeurIPS, receiving distinctions such as the TrustNLP 2021 Best Paper Award.  

He has also served as an area chair and reviewer for major conferences such as ACL, EMNLP, ICML, ICLR, NeurIPS, WWW, and KDD.

Previously, he was a founding engineer at SoftlyAI (2022-2025), and research intern at Google Deepmind (2024), Snap Inc. (2023), and Microsoft Research (2022).