A computational approach to assessing second language reading texts.

摘要

Within the field of second language material development, there is a divide over the value of text types (authentic and simplified) and disagreement on the value of traditional readability formulas to measure text difficulty. In both cases, researchers support their positions based on the linguistic features found in the actual text; however, they do so with little practical evidence. To fill this research gap, this dissertation examined second language reading texts empirically using the computational tool Coh-Metrix. The use of a computational tool to measure linguistic features allowed for an investigation into the differences between linguistic structures in sampled simplified and authentic reading texts at both the beginning and intermediate levels. It also allowed for an examination of how texts develop sequentially between levels and a comparison between how accurately a computational tool can measure text qualities as compared to human participants.; Using a large corpus of simplified and authentic texts, this dissertation demonstrates that many of the perceived linguistic differences between the two were valid, but that many were also invalid. This dissertation also analyzed how authentic and simplified texts developed sequentially from beginner to intermediate level. This analysis found that simplified texts built upon the linguistic features of previous levels to a greater degree than authentic texts. Lastly, this dissertation compared the ability of a computational tool to distinguish between text types as compared to human participants. This analysis demonstrated that human participants were motivated to distinguish between text types using linguistic features in a manner that significantly correlated to the degree in which a computational tool distinguished between text types. Based on these analyses, this dissertation concluded that computational tools can be used to better inform discussions about the linguistic difficulty of texts, measure how texts develop sequentially between levels, and, as an evaluative device, complement human intuition about the linguistic qualities of text. All these findings have important implications for the future of material development, second language acquisition theories, and second language pedagogy.