The present study developed a model of L2 speaking proficiency investigating how lexical sophistication, lexical diversity, syntactic sophistication, syntactic complexity, and cohesion are associated with holistic scores of L2 speaking proficiency employing structural equation modeling (SEM). A corpus of 419 monologues delivered by Iranian EFL learners was compiled and rated to develop the model. Based on the overall scores, the corpus was divided into independent (B1 and B2) and proficient (C1 and C2) users. The results of SEM analysis revealed that the developed L2 speaking proficiency model had an acceptable fit, with partial generalizability across independent and proficient users. Structural regression analysis showed that lexical diversity, lexical sophistication, syntactic sophistication, cohesion, and the indirect effect of syntactic complexity through lexical sophistication explained 34% of the variance in L2 speaking proficiency in descending order of importance. However, their relative importance changed depending on proficiency level. Based on the results, while lexical, syntactic, and cohesive features are sound predictors of L2 speaking proficiency, they function differently across proficiency groups. These findings offer valuable insights for improving speaking proficiency assessment by showing that lexical sophistication, lexical diversity, syntactic sophistication, syntactic complexity, and cohesion do not contribute equally to overall L2 speaking proficiency, and their order of importance varies across proficiency levels. Therefore, prioritizing indicators of L2 speaking proficiency in assessment frameworks based on their importance in each proficiency level can add to the validity and reliability of speaking assessments.

Keywords