Theses Doctoral

Using Eye Tracking to Investigate Reading Task Complexity Effects on L2 Learners’ Content Learning and Language Use

Sun, Haimei

Task-based language teaching (TBLT), a research-informed pedagogy for fostering second language (L2) learning through functional language use, advocates the use of tasks for organizing instructional content and the sequencing of tasks based on task complexity. While the focus of much research has been on the complexity of speaking and writing tasks, to date, scant research has been directed at the impact of reading task complexity, especially when aimed at the learning of subject matter (i.e., content learning). With increasing numbers of multinational learner classrooms, the effectiveness of such instruction constitutes an ever more indispensable factor in all levels of education, exerting a profound impact on the lives of millions of L2 learners as well as on the cultivation of skilled bilingual and multilingual citizens capable of applying content area knowledge to tackle society’s wider challenges such as pandemics.

Adopting a within-subject design, this dissertation zeroed in on a specific type of reading task—read to summarize—examining the degree to which the manipulation of reading task complexity affected L2 learners’ reading processes (i.e., attention allocation and depth of processing) and reading outcomes (i.e., content learning and language use). 30 international students enrolled in graduate programs in the U.S. were recruited to complete three read-to-summarize tasks online while their eye and mouse movements were recorded. Follow-up stimulated recall interviews based on the eye-tracking heatmaps and mouse-tracking recordings were conducted to probe depth of processing. Written summaries served as measures of content learning and language use; additionally, familiarity ratings and short-answer responses were included to gauge learning of main ideas and specific details, respectively. Screening and exit surveys were also administered to collect participants’ demographic information and task perception ratings. Data analyses were performed in Python 3.9 and R Studio 2021.9.1.

Findings from the language use measures show that the most complex task, in general, elicited greater phrasal complexity and the least complex task engendered greater amounts of subordination and coordination. As for content learning, the task of medium complexity yielded more correct major and minor idea units. These findings collectively suggest that while the most complex task was more facilitative of advanced language use, the task of medium complexity was more conducive to content learning. Regarding the results of the process measures, more complex tasks generally led to longer dwell time and more fixation counts than less complex ones.

However, when disaggregating the results, the high-performing group had shorter dwell time and produced more main ideas in the most complex task than its low-performing counterpart. Results from the interview data further reveal that the high-performing group strategically engaged in efficient higher- and lower-level processing, whereas the low-performing group tended to demonstrate inefficient lower-level processing. Furthermore, focused analyses of four participants uncover a great deal of individual variability both in online processing and in the resulting learning outcomes. These findings are discussed in relation to the comprehension and production processes as encapsulated within one pedagogic task; theoretical, methodological, and pedagogical implications are expounded.

Geographic Areas


This item is currently under embargo. It will be available starting 2026-10-18.

More About This Work

Academic Units
Arts and Humanities
Thesis Advisors
Gordon, Peter
Ed.D., Teachers College, Columbia University
Published Here
November 9, 2022