Advancing Cognitive Assessment Through Natural Language Processing and Machine Learning in Technology-Rich Environments

• In Event: AERA Poster Session 14
  In Poster Session: Advanced Technology and Learning Poster Session

Abstract

Natural language processing-based machine learning (ML) applications allow for the automated assessment of students' oral and written performance. Such technological advances contribute to strengthening test score interpretations beyond limited selected-response test formats. This technological development is especially significant for assessing young learners with learning-related difficulties. The purpose of the present study was to evaluate the effectiveness of ML classifiers for predicting young students’ reading difficulties and further to examine the potential of acoustic and prosodic features in predicting reading difficulty profile membership. The study results demonstrate that fine-grained acoustic markers can accurately classify students with different learning and reading profiles. Prosodic features, such as jitter, are shown to be a potentially valuable feature for classifying students with different learning profiles.

Authors

• Eunice Eunhee Jang, University of Toronto

• Hyunah Kim, Education Quality and Accountability Office

• Jeanne Sinclair, University of Toronto

• Clarissa Hin-Hei Lau, University of Toronto

• Christine Marie Barron, OISE/University of Toronto

• Melissa R. Hunte, University of Toronto

• Samantha Dawn McCormick, University of Toronto

• Elizabeth Jean Larson, Ontario Institute for Studies in Education/ University of Toronto

• Liam Hannah, University of Toronto

• Lois Maplethorpe, University of Toronto