Teacher–student interactions are crucial in language learning, often modulated by students’ emotional status. Although the interaction quality is known to be associated with improved learning outcomes, its effect in anxious contexts remains unclear. This study objectively captured students’ classroom anxiety using heart rate variability, and innovatively assessed teacher–student interaction quality with functional near-infrared spectroscopy (fNIRS) hyperscanning. We specifically explored how students’ anxiety levels and teacher–student interaction quality contribute to language learning. Results showed that higher anxiety levels were associated with higher teacher–student brain synchrony, which in turn negatively impacted students’ improvement in sentence complexity. This suggests that heightened anxiety may promote passive learning behaviors, leading students to rely more on their teacher and potentially hindering independent thought. Such dependence could limit language development by reducing active engagement and knowledge application. By integrating real-time physiological and neuroimaging methods, this study advances our understanding of the neural mechanisms underlying effective language learning.