Evaluating Zero-Shot and One-Shot Adaptation of Small Language Models in Leader-Follower Interaction

Rafael R. Baptista, André de Lima Salgado, Ricardo V. Godoy, Marcelo Becker, Thiago Boaventura et al.|February 26, 2026arXiv

Key Takeaway

Small language models can handle real-time role classification in robotics with fine-tuning, but adding more context in conversations breaks their ...

Summary

This paper tests whether small language models can quickly learn to identify leader and follower roles in human-robot conversations without needing large models. Researchers fine-tuned a tiny 0.5B model on robot interaction data and found it achieved 86% accuracy while running fast enough for robots to use locally, but struggled when conversations got longer.

efficiency evaluation applications

Key Terms

fine-tuning zero-shot-learning prompt-engineering small-language-models