A staple of college-level writing instruction is individualized feedback that students receive about the texts that they write. This feedback is intended to help students understand what they could have done better in the current text, but also to help them learn how to be a more effective writer in the future. However, this feedback focuses on properties of written products (i.e., what a good text is supposed to look like) rather than on the process of writing (i.e., what to do while writing in order to produce a good text). This is because texts that students submit to instructors for feedback bear little (if any) evidence of the moment-by-moment actions taken by the writer in the process of composition. This Cyberlearning project will develop an intelligent tutoring system for writing, ProWrite, that will automatically capture such moment-by-moment actions using unobtrusive biometric technology, and then provide data-driven, personalized, actionable feedback about the composition process to the student. This feedback will take the form of focused strategy guidance: Instead of simply telling the student to try out a particular strategy, ProWrite will provide a coached writing session where the student will receive real-time, automatic scaffolding for the target strategy. This type of intelligent tutoring system in the context of writing instruction, if effective, has the potential for massive application and, therefore, economic and educational gain.

Specifically, this project will utilize deployable, combined, time-aligned keystroke logging and eye tracking to (1) precisely diagnose issues with a student's writing process that may be preventing them from producing high-quality texts, (2) provide individualized writing-strategy advice for remediation, and (3) scaffold, automatically and in real time, the student applying these new strategies to their own writing. To this end, the researchers will develop an automated end-to-end pipeline for writing-process analysis that will include (1) pause detection based on statistical modeling of inter-key intervals, (2) classification of eye movement patterns, and (3) classification of within-draft revisions learned from an annotated corpus of writing processes. Research activities will be organized into four phases. In the first phase, the researchers will collect and annotate a dataset of writing-process data (keystroke logs, eye movement logs, and final texts), which will then be made publicly available. The second phase, focusing on the development of a first prototype of the full system, will follow the design-based research approach and consist of approximately six small-scale iterations of system development and evaluation. In Phase 3, system functionality will be expanded to include a larger set of diagnosable writing-process issues, with a new series of iterations involving more participants. Finally, Phase 4 will evaluate system efficacy in a randomized controlled study designed to provide a robust test of the benefits of process-focused feedback over currently-used instructional approaches.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Project Start
Project End
Budget Start
2020-10-01
Budget End
2023-09-30
Support Year
Fiscal Year
2020
Total Cost
$750,000
Indirect Cost
Name
Iowa State University
Department
Type
DUNS #
City
Ames
State
IA
Country
United States
Zip Code
50011