Using Automatic Item Generation to Address Item Demands for CAT

Item writing is a craft that is both resource intensive and time consuming. In order to minimize item exposure, computer adaptive testing requires a large number of test items for administration. As test development become more complex, demand for the quantity and quality of items far exceeds the capacity of item writers writing in the current fashion. To address the demand for items to be created in a precise and high-volume fashion, this study suggests the use of automatic item generation (AIG) as a solution to current test development practices. The purpose of this paper is to explore development preferences of item models under AIG, using an item model taxonomy. After adequate training, 34 item models were created and categorized under the taxonomy, and generated 64,280 items. Implications from the study and potential challenges for applying AIG are discussed.