Systematic comparison of computational measures of linguistic synchrony in online educational environments

Linguistic synchrony, the alignment of linguistic features between conversational partners, is a key indicator for understanding students’ learning outcomes in computer-mediated learning environments, where communication quality directly influences success. In online tutoring, synchrony fosters better comprehension and engagement. Despite numerous methods to measure synchrony—such as lexical, syntactic, and semantic alignment—systematic comparisons across these approaches remain limited, impeding a full understanding of how synchrony operates across different educational contexts. This study addresses this gap by analyzing tutoring conversations from diverse contexts, particularly English as a Second Language (ESL) tutoring and algebra instruction, using seven computational models. These models yielded 29 synchrony indices, ranging from traditional lexical overlap to advanced natural language processing-based embedding models, capturing synchrony across multiple dimensions. Our results show that while these methods generally align with theoretical dimensions of synchrony, variations in embedding methods and distance measures can significantly impact their ability to capture meaningful interactions. Additionally, different subject domains influenced the finer-grained dimensions of synchrony observed, particularly in temporal aspects.