Designing AutoTutor Agents to Help Adults With Reading Comprehension Difficulties

• In Event: Symposium: Future Directions for Pedagogical Agent Research

Abstract

The National Research Council (2011) reported that over 50 million adults do not read at a level for them to land a job to land a job to keep pace with US economy. Whereas there is an abundance of technologies designed to improve basic reading skills (such as word decoding, letter sound correspondence, vocabulary), there are fewer technologies to help adults improve reading comprehension (such as sentence meaning, inferences, text cohesion, discourse structure) and use digital media. Conversational agents (talking heads, avatars) are a promising technology to provide comprehension instruction for adults with low literacy because they provide oral communication that applies comprehension strategies to texts and that navigates digital technologies. AutoTutor agents (e.g., Graesser, A.C., Forsyth, C., & Lehman, in press; Graesser, A.C., Forsyth, C.M., & Foltz, P. 2016; Li, H. & Graesser, A. C. (2017) were designed to provide this training on the web. We developed 35 lessons that focused on different comprehension strategies, with each lesson taking approximately 30 minutes to complete by an adult with low literacy. Each lesson has two agents (a tutor agent and a peer agent) that engaged with the adult in conversational trialogues. The technologies were designed systematically by aligning: multilevel theoretical frameworks of comprehension, interventions to improve comprehension, reading curriculum standards, assessments of comprehension, and text characteristics. The strategies are aligned with an intervention that teachers have successfully implemented (face-to-face) to improve reading comprehension in high school students with reading difficulties (Lovett, Lacerenza, De Palma, & Frijters, 2012).

A feasibility study was conducted on 100 hours of an intervention on 52 adult learners in Toronto and Atlanta. The intervention was a blended between teacher-led sessions and AutoTutor that took 4 months to complete, with 2-3 sessions each week. Self-report data from the adult learners indicated they were very engaged with AutoTutor, with 90% having positive ratings. The behavioral performance data were also encouraging. The adults in the feasibility study attended 81% of the sessions, completed 79% of the lessons, answered 71% of the questions, and answered them correctly 70% of the time (33% is chance). This is an outstanding retention statistic compared with norms of attrition rates in adult literacy centers (see NRC, 2011). There were significant learning gains from pretest to posttests on three standardized comprehension tests. This conversational scaffolding with agents is very different than the traditional computer-based training with multiple choice questions and little or no scaffolding. In conclusion, conversational agents are very suitable for adults with low literacy. One could imagine a portal free to the public to teach comprehension skills and also provide professional development to teachers and tutors in adult literacy centers, most of whom have had very little training on comprehension strategies.

Author

• Arthur C. Graesser, The University of Memphis