Reconstructing Readability: Recent Developments and Recommendations in the Analysis of Text Difficulty

Abstract

Largely due to technological advances, methods for analyzing readability have increased significantly in recent years. While past researchers designed hundreds of formulas to estimate the difficulty of texts for readers, controversy has surrounded their use for decades, with criticism stemming largely from their application in creating new texts as well as their utilization of surface-level indicators as proxies for complex cognitive processes that take place when reading a text. This review focuses on examining developments in the field of readability during the past two decades with the goal of informing both current and future research and providing recommendations for present use. The fields of education, linguistics, cognitive science, psychology, discourse processing, and computer science have all made recent strides in developing new methods for predicting the difficulty of texts for various populations. However, there is a need for further development of these methods if they are to become widely available.

The corpus of Basque simplified texts (CBST)

“…Usually, an analysis of several linguistic and statistical features such as word types, dependencies or n-grams is performed and then machine learning techniques are applied in order to determine the complexity grade of the text. Surveys about readability assessment techniques can be found at DuBay (2004), Benjamin (2012) and Zamanian and Heydari (2012).…”

Beauty and the Beast: on the readability of object-oriented example programs

“…Recent research, though, shows that including such factors does not necessarily lead to better predictors for readability (Benjamin, 2012). "The [readability] formulas have survived 80 years of intensive application, investigation, and controversy, with both their credentials and limitations remaining intact" (DuBay, 2004).…”

Automating readers' advisory to make book recommendations for K-12 readers

“…To accomplish this task, Rabbit first determines the readability level of a reader R by analyzing the grade levels of books in R's profile. The readability level of R is determined by averaging the grade level of each book PB in R's profile, computed using TRoLL [12], a tool for regression analysis of literacy levels, which captures the central tendency of the grade levels of books that have been read by R. Unlike popular prediction formulas/tools (such as Flesch-Kincaid, Lexile, and ATOS [1]), which rely on text of a book to compute its grade level (a severe constraint, since text is not always freely accessible due to copyright laws), TRoLL computes the grade level of any book using book metadata publicly accessible from reputable online sources, even in the absence of sample text.…”