I am Ph.D. student at the NLP & AI Lab, Korea University, advised by Prof. Heuiseok Lim. My research centers on large language models — Cross-lingual Transfer, Policy Optimization, Information Retrieval, and Retrieval-Augmented Generation. My primary focus lies in scalable representation learning and developing new training paradigms for language models.
Seungyoon Lee
News
- Aug 2026: Two papers, CRPO and CultureConverse, have been accepted @ EMNLP 2026! 🎉 🇭🇺
- May 2026: Our work, CLEAR, has been accepted to the main conference @ ACL 2026! 🎉 🇺🇸
- May 2026: Two papers have been accepted @ ACL 2026 MeLLM Workshop, on cross-lingual information retrieval. 🇺🇸
- Feb 2026: VAETKI, Korea's sovereign AI foundation model, has been released. I served as the post-training team leader for the Korea University team. 🇰🇷
- Oct 2025: 🏆 Our work, KULLM-RAG, received the Best Paper Award @ HCLT 2025. 🇰🇷
- Aug 2025: Our work, TORSO, has been accepted @ EMNLP 2025 as an Oral presentation! 🇨🇳
- May 2025: Our work, SALT, has been accepted @ ACL 2025 Findings! 🇦🇹
- Jan 2025: Two papers, FindIntInst and FLEX, have been accepted @ NAACL 2025 Findings! 🇺🇸
- Dec 2024: Our work, MIGRATE, has been accepted @ COLING 2025. 🇦🇪
- Sep 2024: Our work, Translation of Multifaceted Data, has been accepted @ EMNLP 2024 Findings. 🇺🇸
- May 2024: Our work, LeVoC, has been accepted @ ACL 2024 Findings. 🇹🇭
- Mar 2024: Our work, Exploring Inherent Biases in LLMs within Korean Social Context, has been accepted @ NAACL 2024 SRW. 🇲🇽
- Feb 2024: Our work, KHSCP, has been accepted @ LREC-COLING 2024 as an Oral presentation. 🇮🇹
- Oct 2023: 🏆 Received two Best Paper Awards @ HCLT 2023, for KoCheckGPT and Analysis of Toxicity and Bias of ChatGPT. 🇰🇷
Publications
2026
Language Chain in Alignment: Cross-Lingual Ranking Preference Optimization
EMNLP 2026
CLEAR: Cross-Lingual Enhancement in Retrieval via Reverse-training
ACL 2026
CultureConverse: A Multilingual Multi-turn Simulation Harness for Culturally Grounded Assistance in East and Southeast Asia
EMNLP 2026
Query-Synergy: Leveraging High-Resource Languages for Improving Retrieval Performance Across Multiple Languages
ACL 2026 MeLLM Workshop
Improving Korean-English Cross-Lingual Retrieval: A Data-Centric Study of Language Composition and Model Merging
ACL 2026 MeLLM Workshop
LAMAR: An Open Language-Aware Multilingual Alignment Reranker
arXiv 2026
Answer-Conditioned Chains of Thought Degrade Verifiable-Reasoning Distillation in Large Language Models
arXiv 2026
Beyond Penalizing Mistakes: Stabilizing Efficiency Training in Large Reasoning Models via Adaptive Correct-Only Rewards
arXiv 2026
PRISM: Pruning via Interference-Signal Maximization for LLM Merging
Preprint
Skin-Deep: A Geometric Diagnostic for Alignment Fragility in Large Language Model Representations
arXiv 2026
2025
Semantic Aware Linear Transfer by Recycling Pre-trained Language Models for Cross-lingual Transfer
ACL 2025 Findings
TORSO: Template-Oriented Reasoning Towards General Tasks
EMNLP 2025Oral
Find the Intention of Instruction: Comprehensive Evaluation of Instruction Understanding for Large Language Models
NAACL 2025 Findings
FLEX: A Benchmark for Evaluating Robustness of Fairness in Large Language Models
NAACL 2025 Findings
MIGRATE: Cross-Lingual Adaptation of Domain-Specific LLMs through Code-Switching and Embedding Transfer
COLING 2025
2024
Translation of Multifaceted Data without Re-Training of Machine Translation Systems
EMNLP 2024 Findings
Length-aware Byte Pair Encoding for Mitigating Over-segmentation in Korean Machine Translation
ACL 2024 Findings
Exploring Inherent Biases in LLMs within Korean Social Context: A Comparative Analysis of ChatGPT and GPT-4
NAACL 2024 SRW
Leveraging Pre-existing Resources for Data-Efficient Counter-Narrative Generation in Korean
LREC-COLING 2024Oral
~2023
Alternative Speech: Complementary Method to Counter-Narrative for Better Discourse
ICDM 2023 DCAI Workshop
Focus on FoCus: Is FoCus focused on Context, Knowledge and Persona?
COLING 2022 CCGPK Workshop
Project
Jun 2025 – Jan 2026
VAETKI
— Korean Sovereign AI Foundation Model
Post-training Team LeaderNC AI (NCSOFT)-led consortium
- Led the full post-training pipeline — Supervised Fine-Tuning, Preference Optimization, and Multi-Node Distributed Training.
- Biggest National sovereign AI foundation model project organized by the Ministry of Science and ICT.
- VAETKI is a 100B+ parameter Mixture-of-Experts model (11B active at inference) with custom Multi-head Latent Attention.
Feb – Oct 2025
KULLM-R
— Efficient Korean Reasoning Model
Project ManagerNLP & AI Lab, Korea University
- Build an efficient Korean reasoning model based on Qwen3 via reinforcement learning.
- Superior reasoning efficiency and mitigation of the language confusion phenomenon in Korean.
Jan – Jul 2024
KULLM3
— Language Transfer in Large Language Model
Lead ResearcherNLP & AI Lab, Korea University
- Developed a new embedding transfer method for smooth language adaptation.
- An advanced instruction-following and fluent chat model specialized in Korean, closely following gpt-3.5-turbo.
Honors & Awards
Best Paper Award
Annual Conference on Human and Cognitive Language Technology (HCLT 2025)
Oct 2025
Best Paper Award
Annual Conference on Human and Cognitive Language Technology (HCLT 2023)
Oct 2023
Best Paper Award
Annual Conference on Human and Cognitive Language Technology (HCLT 2023)
Oct 2023
2022
Statistics Korea Award
Artificial Intelligence Competition
2022