A Corpus-Based Analysis of Narrativity Features and Their Impact on Writing Performance among Pakistani ESL Students

Abstract

This corpus-based study examines the influence of discourse-level narrativity features specifically verb cohesion (PCVERBz), narrative cohesion (PCNARz), and causal reasoning (SMCAUSv, SMCAUSlsa) on the academic writing performance of Pakistani English as a Second Language (ESL) undergraduate students. A localized corpus of 100 argumentative essays was collected from universities across Punjab, Pakistan, and 26 multi-level linguistic indices were extracted using the computational tool Coh-Metrix. The essays were evaluated using local automated scoring software calibrated to the official IELTS Task 2 band descriptors across four core domains: coherence and cohesion (CC), lexical resource (LR), grammatical range and accuracy (GR), and task response (TR). Due to severe violations of normality among the data distributions, Spearman's rank correlation was applied to establish bivariate associations. Concurrently, ridge regression analysis was executed to model the multi-feature predictive relationships while controlling for extreme multicollinearity and overfitting among the cohesive metrics. Bivariate and regression analyses demonstrate that an elevated frequency of repetitive verbs (PCVERBz) consistently predicts lower writing quality across all evaluation descriptors. Conversely, narrative cohesion (PCNARz) and forward causal reasoning via latent semantic analysis (SMCAUSlsa) exhibit strong positive effects on text coherence (CC) and task fulfillment (TR). These findings suggest that English for Academic Purposes (EAP) pedagogy must transition from traditional structural grammar tracking to explicit discourse-level interventions. Pedagogical applications emphasize the implementation of targeted vocabulary substitution drills, cause-and-effect cognitive mapping, and explicit exercises guiding the genre shift from sequential narrative structures to analytical expository prose.