Answers are grounded in this paper and cite their sources.
Cognitive load is a foundational construct in learning. Yet, most adaptive educational systems continue to depend on behavioral traces or physiological instrumentation to infer learners’ mental effort, limiting their scalability and ecological validity in open-ended learning contexts. This paper introduces a language-based framework for turn-level, real-time, and non-invasive cognitive load regulation in interactive learning environments. We present the Cognitive Load Estimate from Language (CLEL). This metric synthesizes eight linguistically grounded indicators into a unified estimate of cognitive load derived directly from learners’ written explanations. Building on this metric, we develop the Cognitive Load-Adaptive Multi-Agent Tutor (CLAMAT), a multi-agent instructional architecture in which specialized LLM-based agents coordinate adaptively through a cognitive-load-aware orchestration mechanism. Empirical evaluation with undergraduate learners performing explanatory tasks shows a strong association between CLEL and self-reported mental effort. Teacher evaluations favour CLAMAT feedback in clarity, helpfulness, and perceived adaptivity. Students also show an overall pre–post score improvement after receiving feedback from both systems, although the design does not isolate their respective contributions to learning. These findings demonstrate the promise of language-based cognitive modeling as a foundation for scalable, theoretically grounded, and highly personalized interactive learning environments.
No takes yet. Share an insight, caveat, or question.
Sachete et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: