The field of Human–Robot Interaction (HRI) explores the mechanisms that enable robots to engage meaningfully with humans across diverse contexts. A central focus within this domain is robotic learning, where Learning from Demonstration (LfD) has emerged as a promising approach for enabling robots to acquire behaviors by observing human actions. However, the implementation of LfD poses fundamental challenges, particularly in parsing human demonstrations and translating them into executable robotic actions. These challenges are often framed within a Theory of Mind (ToM)-based paradigm, which assumes that social understanding relies on inferring human intentions through internal state modeling. Yet, this reliance on inferential models not only imposes significant computational demands but also reflects deeper epistemological assumptions about cognition and interaction. This chapter critically examines these assumptions and explores an alternative approach rooted in cognitive semiotics and enactive theories of social cognition. By drawing on insights from developmental robotics and motionese, I consider whether a more interaction-sensitive perspective—one that foregrounds embodied, relational meaning-making—might offer new pathways for robotic learning. This shift not only challenges established perspectives in LfD but also calls for a reconsideration of how robotic learning algorithms engage with the dynamics of social interaction.