Semester Award Granted

Summer 2025

Submission Date

August 2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Thesis/Dissertation Advisor [Chair]

Xingquan Zhu

Abstract

Writing is an essential skill that affects success in nearly every academic subject and professional field. For undergraduate students, strong writing helps them organize ideas, communicate clearly, and perform better in both written assignments and overall coursework. Good writing also supports critical thinking, which is key to problem-solving and academic growth. Beyond school, writing continues to be important in the workplace, where it is used for emails, reports, presentations, and formal documents. Yet, despite its importance, many students and graduates do not have strong writing skills, and this gap is noticed by employers. A recent survey by Ashley Finley [1] found that while 90% of employers value written communication, only 44% believe graduates are prepared.

At the same time, recent progress in artificial intelligence has made tools like neural language models useful for supporting writing instruction and grading. These models offer faster and more objective ways to assess student writing. In this study, we explore how automated writing assessment can work at the sentence level, focusing on Writing Across the Curriculum (WAC) categories used to assess student writing in College Writing 1 and 2 at Florida Atlantic University. We collected final argumentative essays written by students and analyzed them using a neural language model to assess writing quality, at both sentence by sentence level and across the whole essay. Our findings show that the model can recognize patterns in writing and provide useful evaluations, but there are still challenges with scoring consistency. This research shows possible improvements to address these issues and highlight key takeaways from the case study that support using sentence-level assessment in writing instruction.

Download

Share

COinS