Conceptual and empirical bases of readability formulas

The question of what features make a text difficult or easy for a reader is examined in this paper, which looks at the implications from the perspective of readability formulas. This is related to the larger question of text comprehensibility. Problems arise when difficult words and long sentences are treated as the direct cause of difficulty in comprehension and are used in readability formulas to predict the readers' comprehension. Readability formulas are not the most appropriate measure and cannot reliably predict how well individual readers will comprehend particular texts. Far more important are text and reader properties which formulas cannot measure. Neither can any formula be a reliable guide for editing a text to reduce its difficulty. Conceptual and Empirical Bases of Readability Formulas The question of what features of a text make it easy for a reader is interesting from many different perspectives. In this paper we will examine this question and its implications from the specific perspective of readability formulas, pointing out the basic choices and assumptions made in their development and use. These assumptions will be discussed in relation to the larger question of text comprehensibility in which the use of formulas is embedded. We question to what degree readability formulas actually do what they were intended to do: to gauge whether particular texts can be read and understood by particular readers or groups of readers, on some particular use or occasion of reading. We will argue that readability formulas are not the most appropriate measures for this purpose, for the reasons which follow. Summarizing the arguments, we note that the aggregate statistical model which readability formulas are based on is inappropriate. As a consequence, formulas do not reliably predict comprehension for individual readers. Formulas are also misleading guides for editing a text to reduce its difficulty. They measure features of a text which are at best correlated with difficulty, without being a more specific causal model. A causal model would define what features of language actually contribute directly to difficulty in comprehension, whereas formulas, being based only on statistical correlations, cannot be used to diagnose what is difficult about the language in a text. Readability Formulas 4 Formulas are applied by calculating the average sentence length and word difficulty in short samples of texts. Features of a text not among the features of sentence and word difficulty almost certainly make a much greater difference to comprehension than the features which are measured by applying a formula. The criteria of comprehension associated with formulas are comprehension measures which are generally the least sensitive to specific features of language, of the experimental measures currently in use. Finally, to the extent that formulas do capture some plausible intuitions about the working memory capacity of a reader, this notion needs to be made more explicit in the context of basic research using on-line measures of attention and