ACL and EMNLP are two flagship conferences in computational linguistics, natural language processing, and AI in general. Yu Wang, a former member of the A02 and B07 projects, presented project A02’s paper at ACL 2026 and will present a newly accepted one at EMNLP 2026.
The paper presented at ACL titled "Investigating the Representation of Backchannels and Fillers in Fine-Tuned Language Models," focuses on linguistic units such as backchannels and fillers (e.g., "uh-huh"), which are often considered signals of understanding during interaction. The paper addresses the existing issue that backchannels and fillers are generally underrepresented in language models and develops methods to improve their representation. This research aligns with the theme of the A02 project, as it aims to build conversational agents that can adaptively interact with users and signal understanding during conversation.
The new paper, "Syntactic Complexity in Dialogue: Metrics, Convergence Phenomenon, and LLM Detection Application," which was recently accepted by EMNLP 2026, builds on A02's dialogue data and discusses the intrinsic differences between human-human and human-LLM dialogues by examining syntactic complexity convergence as a form of alignment phenomenon. It concludes that syntactic complexity convergence in human-human dialogues differs from that in human-LLM dialogues. Based on this, the authors built an application to detect whether an LLM is involved in a dialogue by analyzing syntactic-level features.
Yu Wang submitted his PhD dissertation in June and is currently a postdoctoral researcher at Utrecht University, working on an ERC project about memory retrieval in language processing.