Diagnostic assessment of adults' reading deficiencies in an intelligent tutoring system
In this paper, we investigate whether a version of AutoTutor that teaches comprehension strategies can be used to diagnose reading deficiencies in adults with low literacy. We hypothesized that the speed and accuracy with which participants answered questions during the AutoTutor conversation could be diagnostic of their mastery of reading comprehension components: Words, the explicit textbase, the situation model, and rhetorical structure. We used linear mixed effect models to compare the accuracy and response times of 52 low literacy adults who worked on 29 AutoTutor lessons during a four-month intervention period. Our results show that adults' response accuracy for questions addressing more basic reading components (e.g., meaning of words) was higher than for those pertaining to deeper discourse levels. In contrast, question response time did not vary significantly among the theoretical levels. A correlation analysis between theoretical levels and performance (accuracy and time) supported this trend. These results affirm that adults with low literacy tend to have more proficiency for basic reading levels than for deeper discourse levels. In addition, the results of exact binomial test showed that hints or prompts were effective in scaffolding learning reading. Furthermore, we describe how response accuracy on the four comprehension components can provide a more nuanced diagnosis of reading problems than a single overall performance score. More fine-grained diagnoses can assist both educators wanting more detailed insight into learner difficulties, and ITS developers looking to improve the personalization and adaptivity of learning environments.
Shi, G., Lippert, A., Hampton, A., Chen, S., Fang, Y., & Graesser, A. (2019). Diagnostic assessment of adults' reading deficiencies in an intelligent tutoring system. Retrieved from https://digitalcommons.pvamu.edu/psychology-facpubs/52