Computational and Neuronal Mechanisms of Goal Signaling in Novel Human Interactions

Abstract

This thesis investigates how humans communicate goals in novel interactions where shared linguistic conventions are absent. Traditional models of communication rely on predictability within established languages to facilitate understanding. However, in situations lacking a common language, communicators must develop alternative strategies. The thesis proposes a unified framework centered on expectation violations as a key mechanism for signaling intentions in such contexts. This approach emphasizes two mechanisms: first, using universal knowledge as shared expectations, and second, intentionally deviating from these established expectations to create moments of surprise, thereby drawing attention to critical aspects of a message.

The proposed framework is validated through the Tacit Communication Game (TCG), an experimental platform designed to study non-verbal communication strategies in novel interactions. Using a computational model called the Surprise model, the thesis demonstrates that effective communication is achieved through expectation violations. Furthermore, it shows that these surprise-based signals correlate with physiological and neural responses of the Receiver, such as pupil dilation and EEG activity, indicating the brain's sensitivity to unexpected communicative cues.

The thesis also compares the effectiveness of the Surprise model with Theory of Mind (ToM)-based models, which require mental state reasoning, in communicating the goal. Results show that the Surprise model performs as effectively as more cognitively demanding ToM strategies in guiding communication, offering a robust alternative approach to goal signaling in novel interactions.

By integrating behavioral, computational, and neurophysiological evidence, the thesis provides a comprehensive framework for understanding how humans adapt their communication strategies to achieve mutual understanding in an unfamiliar context.