Loading...
Citations
Altmetric:
Soloist
Composer
Publisher
Date
2025-05-19
Additional date(s)
Abstract
Oral Reading Accuracy (ORA) measures how accurately students read aloud and plays
a key role in assessing reading proficiency. This study investigates the use of speech
recognition systems for ORA scoring, focusing on the statistical estimation of misclas-
sification rates when both human and AI scores may contain errors. Using data from
507 elementary school students across ten passages of varying lengths and difficulties,
we evaluate classification accuracy in terms of true positive (correct words identified as
correct) and true negative (incorrect words misclassified as correct) rates. We develop
estimation procedures for these misclassification rates using the Method of Moments
(MOM) and the Generalized Method of Moments (GMM), accounting for scenarios
with two contaminated data sources. This work considers both the scenario where true
counts are observed and the more realistic case where only contaminated scores are
available, demonstrating that reliable performance metrics can still be recovered and
supporting the scalability of automated ORA assessments.
Contents
Subject
Subject(s)
Files
Research Projects
Organizational Units
Journal Issue
Genre
Description
Format
Department
Mathematics