Prompt-Specificity in Integrated Writing Assessment: Cross-Prompt Generalizability of Linguistic Indicators in Source-Based Writing Ref.No.SSTCRC2691
1. Introduction
This study asks whether the linguistic features used as indicators of writing quality stay stable when the writing prompt changes. It focuses on phraseological, lexical, and syntactic features in source-based (integrated) writing. Instead of only identifying which features correlate with scores, it asks whether those relationships reflect a fairly stable component of writing quality or depend partly on the prompt that produced the text. The answer matters for assessment. A feature that works well for one prompt and poorly for another can give misleading evidence about a student's underlying writing ability. The study therefore examines prompt-specific feature–score relationships and tests whether models trained on some prompts can predict scores on unseen prompts. Earlier research has looked at task effects on linguistic complexity and on how well features differentiate proficiency levels, and it shows that task characteristics can strongly shape learners' linguistic production.
2. Research Progress
The theoretical foundation and the research questions are developed. The data is also ready.
3. Required Cooperation
Research funds as well as an expert collaborator.
4. Benefits
The project will show whether linguistic features give stable evidence of writing quality across integrated-writing prompts, or whether their relationships with scores depend on the task. It should identify features whose associations with writing quality stay consistent across prompts and features that vary substantially from prompt to prompt. It will also show whether models built on linguistic features from existing prompts generalize to essays written for unseen prompts. The findings can inform work on construct representation in automated writing assessment by separating features that reflect general writing quality from features that mainly reflect the task or prompt.
5. Outputs
One or two research publications in top journals