Developments in Syntax-Based Prescoring and Autoscoring on Short Written Answers

• In Event: 35.074 - Digital Literacies and Technology

Abstract

Pre- or autoscoring large datasets would tremendously increase accessibility of large written data for both research and practice. Two sets of automatically derived models based on large previously run studies with 5th-7th graders (n1 = 238; n2 = 4482) are compared for their predictive power to autoscore two complex scores for writing and recall. The models predict higher order and complex quality scores on the basis of easily computable syntactic chains of word forms (e.g., Adjective – Determiner – Noun). With about 200 scores by trained raters, models can be estimated and implemented to securely create autoscores. This paper investigates the boundaries of this approach for short written answers.

Authors

• Pablo Nicolai Pirnay-Dummer, Martin-Luther-University Halle-Wittenberg

• Kausalai K. Wijekumar, Texas A&M University - College Station

• Bonnie J. F. Meyer, The Pennsylvania State University