Cohesion Features of Material for Young Readers of English : A Computational Investigation

Abstract

  This study aims to investigate the cohesion features of material marked for young readers of English and to examine whether such features predict their readability. A total of 129 texts were collected from the material written for young readers of English. These texts varied in the levels of difficulty (readability) assigned by experts. The texts were analyzed using a computational tool, the Tool for the Automatic Analysis of Cohesion (Crossley et al., 2019), that produces over 150 indices related to text cohesion. Some indices were selected after controlling for correlations, normality, and multicollinearity, and were then used as predictors in the multiple regression analysis to estimate the readability level of the texts. The results demonstrated that the combination of indices in (a) verb overlap between adjacent two sentences, (b) semantic overlap between adjacent sentences, (c) the use of temporal connectives, and (d) the number of repeated content words accounted for 37% of the variance in readability. The difficulty level of the texts increased as they contained less verb overlap and semantic overlap between sentences, and showed less repetition of content words, but included more temporal connectives. These results have been discussed in the paper in light of the theories of discourse comprehension and characteristics of texts written for young readers. The study's findings provide theoretical implications for developing reading materials for young learners of English. The findings also highlight the importance of considering the multidimensional features of text cohesion as well as the developmental stages of targeted readers when evaluating text readability.