Mapping the Relationships among the Cognitive Complexity of Independent Writing Tasks, L2 Writing Quality, and Complexity, Accuracy and Fluency of L2 Writing

Drawing upon the writing literature and the task-based language teaching literature, the study examined two cognitive complexity dimensions of L2 writing tasks: rhetorical task varying in reasoning demand and topic familiarity varying in the amount of direct knowledge of topics. Four rhetorical tasks were studied: narrative, expository, expo-argumentative, and argumentative tasks. Three topic familiarity tasks were investigated: personal-familiar, impersonal-familiar, and impersonal-less familiar tasks. Specifically, the study looked into the effects of these two cognitive complexity dimensions on L2 writing quality scores, their effects on complexity, accuracy, and fluency (CAF) of L2 production, and the predictive power of the CAF features on L2 writing scores for each task. Three hundred and seventy five Chinese university EFL students participated in the study, and each student wrote on one of the six writing tasks used to study the cognitive complexity dimensions. The essays were rated by trained raters using a holistic scale. Thirteen CAF measures were used, and the measures were all automated through computer tools. One-way ANOVA tests revealed that neither rhetorical task nor topic familiarity had an effect on the L2 writing scores. One-way MANOVA tests showed that neither rhetorical task nor topic familiarity had an effect on accuracy and fluency of the L2 writing, but that the argumentative essays were significantly more complex in global syntactic complexity features than the essays on the other rhetorical tasks, and the essays on the less familiar topic were significantly less complex in lexical features than the essays on the more familiar topics. All-possible subsets regression analyses revealed that the CAF features explained approximately half of the variance in the writing scores across the tasks and that writing fluency was the most important CAF predictor for five tasks. Lexical sophistication was however the most important CAF predictor for the argumentative task. The regression analyses further showed that the best regression models for the narrative task were distinct from the ones for the expository and argumentative types of tasks, and the best models for the personal-familiar task were distinct from the ones for the impersonal tasks