Assessing the Readability of Narrative Texts in ESL Settings: A Comparison of Subjective Ratings and Vocabulary Distributions

This vocabulary-based corpus study explores how the varying distributiom of high-frequency words, academic words, and low-frequency words in narrative texts relate to the subjective readability level that each text has been assigned to by ESL teachers. Based on previous research findings that vocabulary is one of the strongest predictors of text difficulty, the researcher explored the relationships between the subjective readability levels given the texts and their vocabulary distributions in the three base lists of the corpus analysis program *Range*. Additionally, the researcher investigated the effects of passage length (i.e. 5000-, 3000-, or 1000-words) on those relationships.A corpus of 30 narrative texts that had already been subjectively rated by a committee ESL teachers into five readability levels were used for the vocabulary analysis. The researcher scanned 5000-, 3000-, and 1000-word passages from each of the narrative texts and compared the vocabulary content of each text with the three base lists in *Range*: the first and second 1000 word families from the *General Service List (GSL)* and approximately 570 word families from the *Academic Word List (AWL).* A fourth list of words (*Other*) that did not fall in the three base lists of *Range* was generated for each text. The vocabulary distributions in each of the four categories were analyzed using descriptive and inferential statistics.The results of this study illustrate how narrative texts intended for ESL students at a specific readability level vary in vocabulary distributions of high- and low-frequency words. Statistically significant vocabulary distribution trends were discovered at each readability level in the first 1000 word families of the *GSL* and in the list of text-specific, low-frequency *Other* words generated by *Range* for each text. Distributions of types (unique words) were more telling than token distributions. Additionally, a relationship between type-token ratio and subjective readability level was found. Some of the pedagogical implications of these findings are discussed.

Thesis Author: Harnois, Rebeka Diane

Year Completed: 2003

Committee Members: Cynthia L. Hallen, C. Ray Graham

Thesis Chair: Dee I. Gardner