Session Summary
Share...

Direct link:

Recent Innovations in Automated Scoring

Sun, April 14, 7:45 to 9:15am, Convention Center, Floor: First, 120C

Session Type: Coordinated Paper Session

Abstract

This coordinated session illustrates five recent developments in automated scoring. Some of the developments are due to the incorporation of transformer models or their variants (Devlin et al., 2019; Vaswani et al., 2017), the refinement of topic identification strategies (Xiong et al., 2021), or technological breakthroughs (Saxton et al., 2019). The first paper shows the challenges and complications of trying to operationalize the scoring of math items. The second uses an mBERT transformer model to evaluate writing in Persian, but can easily be applied to a number of different languages. The third paper highlights how Latent Dirichlet Allocation has improved to help with identifying the topic of writing. The fourth explains the advances in the automated scoring of spoken responses. The last paper is an example of how a traditional operational scoring engine was improved by adopting variations of BERT transformer models. Taken as a whole, these developments highlight that inroads in one area of automated scoring often impact other areas in a positive way. Ultimately, these recent innovations in automated scoring contribute to the broader goal of more equitable and effective educational measurement.

Sub Unit

Session Organizer

Moderator

Individual Presentations

Discussant