A Differential N-gram Use Measure for Automated Essay Scoring

EasyChair Preprint 6196

Abstract

This study implements and expands a differential word use (DWU) measure for automatically assessing the quality of independent and integrated student essays. For independent writing samples, the baseline unigram DWU measure successfully predicted human scores in a training corpus but was less generalizable to a test corpus. For integrated writing, tri-gram measures yielded the best performance, and using n-grams as a unit of feature development improved the performance on both training and testing corpora. This research contributes to our understanding of discourse by validating and expanding current DWU measure and by showing to which extent vocabulary use is related to the quality of written discourse in varied tasks.

Keyphrases: Automated Essay Scoring, Automated essay evaluation, vocabulary use, written discourse

@booklet{EasyChair:6196,
  author    = {Qian Wan and Scott Crossley and Laura Allen and Danielle McNamara},
  title     = {A Differential N-gram Use Measure for Automated Essay Scoring},
  howpublished = {EasyChair Preprint 6196},
  year      = {EasyChair, 2021}}