A readability model for lexical complexity indicators in reading texts for learners of English as a foreign language in Iraq

Abstract

In the existing literature, researchers have constructed a large number of lexical complexity (LC) indices in order to predict the readability of texts in English as a foreign language (EFL) contexts. However, in the Iraqi context, there still exists the problem of redundancy of indices and their use. Attempting to answer the research question of which lexical complexity indices are most predictive of the readability of Iraqi EFL university-level reading texts, the present study aims to develop a model of LC indices that can reflect the readability of Iraqi reading texts. A total of 14 reading texts from the Select Readings: Intermediate textbook were selected. The scores of eight LC and two EFL-specific readability indices were computed using the computational software Coh-Metrix, TAALED, and LCA.  Based upon the multicollinearity of variables, partial least squares regression (PLSR) and variable importance for projection (VIP) were utilized for each LC index in accordance with readability scores. To determine the most predictive LC indices in the final model, two criteria were applied including model accuracy and VIP scores across the resulting models. The results of PLSR and VIP indicated three indices as the more readability sensitive: lexical_density_types, P_Lex, and MATTR. This readability-sensitive model addresses the issue of inconsistency in previous studies as it provides a statistically more accurate model for calculating LC.