Dynamic Support of Contextual Vocabulary Acquisition for Reading (DSCoVAR): An Intelligent Tutoring System
The primary goal of this project is develop an intelligent tutoring system (ITS) to address challenges in acquiring literary ("Tier 2") words from written contexts. The benefits of the resulting intervention will be tested among 6th-8th grade children enrolled in an after-school program for economically disadvantaged students. We will address questions about optimal conditions for contextual word learning, including factors supporting robust, incremental gains in word knowledge and strategies for using specific context cues to guess the meaning of an unknown word in context.
Setting: Participating middle schools will be located in Atlanta, Georgia and in Pittsburgh, Pennsylvania. Falk School (Pittsburgh site) will participate in initial feasibility testing and iterative development. Atlanta schools will serve as the testbed for initial (pilot) evaluation of outcomes for low-SES students, including a large number of students at risk of academic failure.
Sample: Initial studies will be carried out in Years 1 and 2 at Falk School in Pittsburgh, and in Year 2 at schools in the Atlanta Public School system. In year 3, an evaluation (pilot) study will be carried out in five Title 1 schools within the Atlanta Public School system. For this Year 3 study, we will recruit a total of 300 students (grades 6-8). One-hundred percent of the participants will qualify for free or reduced lunch. The sample will comprise ~97% African-American students and will be approximately evenly divided into girls and boys.
Intervention: The proposed research will develop and test a web-based intelligent tutoring system (ITS) that exposes learners to new words in a variety of well-controlled contexts. It will use computational methods to estimate changes in partial word knowledge and will provide strategy training to promote effective use of context cues to meaning. Study findings will inform a follow-on Goal 3 proposal to test the efficacy and effectiveness of the intervention.
Research Design and Methods: In Year 1, the research team will develop a fully functioning version of the DSCoVAR system, which entails development and testing of stimuli, tasks, and algorithms that are key components of the system. In Year 2, the research team will conduct iterative usability studies to see if intended end users can understand and use the system. Participants in these studies will fill out a pre-survey, interact with the ITS, and then fill out a post-survey. Additionally, the research team will conduct feasibility studies to see if end users can feasibly implement the intervention in authentic education delivery settings. The feasibility studies will have the same design and procedures as the pilot study. Year 3 will be devoted to a pilot study evaluating the intervention using a randomized controlled trial design. Participants within each grade will be randomly assigned to the control or to one of two intervention groups - the ITS with enhanced, explicit training on the use of context cues to acquire meaning, or the ITS with no enhanced training - to see if explicit training will lead to better word learning outcomes. All participants will be given a pre-test, an immediate post-test, a 1-week delayed post-test, and a 1-month delayed post-test. Key Measures: During iterative development, the research team will collect information about the usability of the system with surveys that ask participants to rate their experiences with various aspects of the system. For the pilot study, the research team will collect both proximal and distal measures of vocabulary knowledge. The proximal measure will be a researcher-designed test that captures students' knowledge of the Tier 2 words targeted in the intervention. The distal measures will be the vocabulary and passage-level comprehension subscales of the Group Reading Assessment and Diagnostic Evaluation (GRADE) battery. In addition, the research team will develop measures to collect data on two aspects of fidelity of implementation: percent of completed trials and adherence.
Data Analytic Strategy: For all data analyses, the research team will apply one- or two-level generalized linear models (depending on nesting scenario) to assess the relationships between a given outcome measure and covariates. This will include the experimental manipulation and individual difference factors (pre-test vocabulary and comprehension scores and demographic variables such as age, gender, grade level, and socio-economic status).
|Faculty||Grad Students||Undergrad Students||Staff and Interns|
K. Collins-Thompson (PI, Univ of Michigan)
G. Frishkoff (co-PI, Georgia State Univ)
C. Perfetti (co-I, Univ of Pittsburgh)
G. Kuperminc (co-I, Georgia State Univ)
M. Eskenazi (co-I, Carnegie Mellon Univ)
S. Nam (Michigan)
J.S. Lee (Michigan)
A. Bhide (Pitt)
E. Schumacher (CMU)
R. Syed (Michigan)
A. Marcus (Pitt)
G. Tarcijonas (Pitt)
P. Hamilton (Michigan)
S. DeRoss (Michigan)
C. Maier (GSU)
K. Muth (Pitt)