Based on the interactionalist speaking construct,this study validates the human-machine collaboration in rating open-ended speaking tasks with an automated scoring system for the College English Test-Spoken English Test,in which test takers'performances are scored both analytically and holistically on rating scales at the task level.The findings show that the automated scoring system gives scores in relation to task features and test purposes,demonstrating a high level of rating accuracy.In the mode of human-machine collaboration,quite a large portion of the score variance could be attributed to the speaking abilities related factors deemed essential for task completion.When applying automated scoring to the large-scale rating of speaking tests,it is suggested that the task-based analytic scoring be used in setting gold standards for machine learning,and the task-based holistic scoring be adopted for human-machine collaboration in large-scale rating sessions,in order to facilitate score interpretation and ensure rating efficiency.
College English Testspeaking testautomated scoringinteractionalist construct theory
万方数据
维普
