Proposed PA-Tool, a training-free method that adapts tool schemas to align with models’ pretrained knowledge, improving tool-use performance by up to 17% and reducing schema misalignment errors by 80%.
@inproceedings{lee2026patool,title={Don't Adapt Small Language Models for Tools; Adapt Tool Schemas to the Models},author={Lee, Jonggeun and Song, Woojung and Han, Jongwook and Pyun, Haesung and Jo, Yohan},booktitle={Annual Meeting of the Association for Computational Linguistics (ACL)},year={2026},}
Introduced a benchmark evaluating whether Large Audio-Language Models can reliably judge speaker consistency across multi-turn conversations, revealing significant biases in prioritizing text over acoustics.
@inproceedings{lee2026speakersleuth,title={SpeakerSleuth: Can LALMs Judge Speaker Consistency across Multi-turn Dialogues?},author={Lee, Jonggeun and Pyo, Junseong and Seo, Gyuhyeon and Jo, Yohan},booktitle={Annual Meeting of the Association for Computational Linguistics (ACL)},year={2026},}
Developed a spoken user simulator that jointly generates text and speech tokens, modeling realistic spoken behaviors (cross-turn slots, barge-in, disfluency, emotion-aware speech) for task-oriented dialogue systems.
@article{lee2026spokenus,title={SpokenUS: A Spoken User Simulator for Task-Oriented Dialogue},author={Lee, Jonggeun and Pyo, Junseong and Park, Jeongmin and Jo, Yohan},journal={Under review @ EMNLP 2026; Machine Learning for Audio Workshop @ ICML 2026},year={2026},}
Developed a time-accelerated smart home simulation environment with 600 benchmark episodes, revealing that even top models struggle with temporal scheduling and state verification.
@inproceedings{seo2026simuhome,title={SimuHome: A Temporal- and Environment-Aware Benchmark for Smart Home LLM Agents},author={Seo, Gyuhyeon and Yang, Jungwoo and Pyo, Junseong and Kim, Nalim and Lee, Jonggeun and Jo, Yohan},booktitle={International Conference on Learning Representations (ICLR)},year={2026},}
Proposed a framework to systematically measure data contamination in psychometric evaluations of LLMs, providing evidence of strong contamination in popular inventories.
@inproceedings{han2026contamination,title={Quantifying Data Contamination in Psychometric Evaluations of LLMs},author={Han, Jongwook and Song, Woojung and Lee, Jonggeun and Jo, Yohan},booktitle={Findings of the Association for Computational Linguistics: EACL},year={2026},}
Comprehensive survey examining the evolution of tool-augmented agents, focusing on the shift from autonomous capabilities to interactive paradigms in human-centered interaction.
@article{jo2025toolagents,title={Tool-Augmented Agents: Evolution from Autonomy to Interaction},author={Jo, Yohan and Lee, Jonggeun},journal={Communications of the Korean Institute of Information Scientists and Engineers},volume={43},number={11},pages={14--25},year={2025}}