Exploring the Relationship of Cohesion and Coherence in Test-takers’ Written Texts with their Writing Scores in a Continuation Task

Abstract

The purpose of writing assessment is to make inferences about test-takers’ language ability in real-world language use (Weigle, 2002). To serve this purpose, integrated writing tasks have been increasingly used in high-stakes tests due to their authentic nature. The continuation task is an integrated reading-writing task that has been used in the National Matriculation English Test in China since 2016. This task requires learners to read an incomplete story with its ending removed and then complete it as a coherent text (Wang & Wang, 2015). Thus, a key criterion for this task is connections within and between written text and source text. Given the unique role of connections as underlying constructs for this task, research is essential on its impact on writing quality (Ye et al., 2021). To address this issue, this thesis investigated writing quality in the continuation task regarding connections, which are explicitly and implicitly featured by cohesion and coherence in writing (Plakans & Gebril, 2017). Cohesion refers to the existence of recognizable cohesive ties in texts, and coherence indicates the logical flow of ideas beneath texts’ structure. A discourse analysis using TAACO 2.0, a tool for analyzing cohesion, and a corpus of 183 test-takers’ written texts in the testing context was conducted. All written texts were scored and analyzed for cohesion and coherence in relation to writing quality. Correlation and regression analyses were conducted to examine their relationships. The findings suggested that, firstly, there were significant correlations between 17 cohesion indices in TAACO 2.0 and writing scores. Additionally, human ratings of coherence were significantly and positively correlated with writing scores. Secondly, there were significant but moderately small correlations between 15 cohesion indices and coherence. Thirdly, coherence was positively predicted by three cohesion indices, explaining 20% of the variance. Fourthly, the predictive model showed that coherence and five cohesion indices were positive predictors of writing scores, which explained 57% of the variance. These results help understand the relationship of cohesion and coherence with writing scores and provide supporting evidence of the construct validity of this task. Furthermore, this research has implications for explicit instructions for connections in writing.