Instance-Based Learning in Content Scoring: Neural-Henry Engine in Practice

• In Event: Saturday Roundtable Session 9:45 am
  In Roundtable Session: Using Advanced Statistical Methods to Inform Education Policy (Table 32)

Abstract

Content scoring in educational assessments and language proficiency tests is crucial, but traditional methods may prioritize grammar over meaningful communication. Recent advancements in natural language processing offer solutions to this concern. The essay evaluation system by ETS introduces Neural Henry, an advanced essay scoring engine using transformer models and instance-based learning. Unlike conventional algorithms, Neural Henry mimics human comprehension by assimilating vast knowledge and experiences to understand context and semantic coherence. It accesses linguistic patterns and thematic associations to discern the true essence of a response. With the aid of this transformative approach, Neural Henry is empowered to generate precise and impartial evaluations of constructed responses.

Authors

• Xinhui Xiong, ExamRoom AI

• Kenneth Steimel, Educational Testing Service