How Do Learners Read the Content in a Multi-source Reading-to-Write Task? – A Multimodal Study

Text comprehension is an important skill that a student can develop, and reading behaviours like skimming and scanning can provide deep insights into how comprehension unfolds in realtime. These behaviours can be examined through deliberate eye movements, which offer a unique lens for investigating reading processes. Yet, most studies rely on low-level eye gaze metrics, such as fixations, as proxies for reading that do not distinguish reading from other tasks like video watching. More precise operationalisations of reading behaviours can provide deeper insights into learning when examined in real-world educational activities. In this study, we assessed the feasibility of a reading-skimming detection algorithm in a two-hour reading-to-write task. Our analysis identified three distinct reading behaviours—persistent reading, scanning, and skimming, demonstrating the potential of this algorithm for multimodal analysis. Our findings indicate that learners engage in thorough reading when encountering new content for the first time and tend to re-read only specific sections while drafting their essays. By integrating log data with reading behaviours, we identified several theoretical learning strategies and reconstructed a temporal narrative of reading-writing that aligns with contemporary models of reading comprehension.