Readability is a topic of importance to both practitioners and scholars across a wide range of fields and interests, including education, applied and text linguistics, library science, and business, medical, and technical communications. On a practical level, readability criteria are needed for a wide variety of tasks, including selecting appropriate reading materials, effectively communicating technical, medical, and business information to both specialists and non-specialists, creating standardized tests, and teaching writing and communication skills. On a theoretical level, readability is relevant to areas such as applied linguistics, text and discourse theory, and natural language processing.

Although there are many articles concerned with readability from various perspectives, there have been no recent attempts to consider the field more generally, as an area of scholarly research, albeit one of practical import. This book is an attempt at such a study. It brings together the relevant literature and theories, and situates them within a unified account. We hope that it will serve as a one-stop resource for both scholars and practitioners who seek a single source that offers a comprehensive, principled discussion of the issues. While written from a linguistic perspective, the book also makes practical suggestions based on the wide range of research it examines.

However, as we will see when we review various attempts at creating readability formulas, examining evidence with the intent of building a theory based strictly on the degree of correlations is highly problematic. A standard maxim often taught to students when they are introduced to statistics is “correlation is not causation.” Efforts to create readability formulas on the basis of simple correlation provide substantial evidence to support the saying. One of the major problems with an extreme empirical approach is that it tends to lead researchers to focus on correlations that may or may not be present outside of the texts they are examining. This is not to say that there may not be more general underlying processes responsible for a particular correlation in a specific corpus. Still, an examination of readability formulas clearly shows a tendency not to look for underlying processes but to focus on the surface co-occurrence of elements—a tendency that has led different researchers to tweak one factor or another in the hopes of creating a better formula. What tends to be lost is the search for the underlying processes.

To identify underlying processes, theories are necessary. However, theories are only as good as the evidence which they can account for, and claims made on the basis of a theory without evidence are not much more than idle speculation. On the other hand, frequently, evidence that supports one theory can also support many others as well, so in examining the plausibility of a theoretical claim it is important to look for more than one type of evidence to support it.

For example, consider the sentence “the horse raced past the barn died”. In reading this sentence devoid of a particular context, a reader may well think that raced is the main verb until he or she reaches the end of the sentence and can identify died as the word that actually functions in that role. At that point the reader has no choice but to go back and to reanalyze raced past the barn as a participial phrase (or truncated subordinate clause, depending on the syntactic theory being used to understand the process). Some of the evidence that sentences with “garden path” interpretations are difficult to follow can be easily found in computational linguistic discussions. These computational discussions present evidence that relates to the operations of computational parsing mechanisms. However, the way a human being makes sense of a sentence is not necessarily the same as the way a formal parser does. Other kinds of evidence are necessary both to support a hypothesis and to gain a better understanding of the way in which such sentences are difficult for humans to understand. Fortunately, evidence about how humans approach garden path sentences is available from psycholinguistic research.

Again, it is important to emphasize that since every kind of evidence has inherent limitations, robust theoretical claims can only be based on a wide range of different kinds of data. One may argue that evidence used to support linguistic theories is less viable because it is frequently based on intuitions which may have been influenced by the theories they are intended to support. Each kind of evidence has limitations, but a theory that can account for many kinds of data, we would argue, is less likely to be affected by the limitations of a particular type of evidence.

This is not to say that readability research is often able to find more than one kind of evidence to support a contention. However, to the extent that the claims made here and elsewhere fall short of this standard, this should be understood as an invitation for more and different kinds of studies to confirm or refute the claims.

We have hoped and assumed that the readers of this book will come from a diverse range of disciplines and that the book will be accessible to anyone with a professional interest in the principles of effective written communication. Consequently, although we use evidence from a number of technical disciplines, including linguistics and psycholinguistics, we have attempted not to presuppose that our readers are equipped with any specialized knowledge prior to reading this book, or that their interest is necessarily in any way theoretical. Nevertheless, in order to make our arguments in a sufficiently rigorous manner, technical details were necessary. Readers for whom this amounts to too much information, however, can focus exclusively on the general discussion: we have tried to ensure throughout that the arguments we make will be clear even for those who cannot or do not wish to follow the more technical details. While we believe these technical aspects enrich the discussion substantially, the work is intended to be comprehensible to the less technically inclined.

By the word written we mean any visually comprehensible encoding of language, including alphabetic, ideogrammatic, or hieroglyphic systems. The material form in which the writing is encoded—for example, print, digital, or handwriting—is of no concern to us here. We are not suggesting that the material presentation can have no effect on communication. However, it is not the focus of this book. Our concern here is only with the language that has been written down for the purpose of communicating something to somebody—and not with the physical form in which it happens to be presented.

A variety of linguistic studies have identified properties of texts and contexts that can affect comprehension. In order to move the study of readability beyond correlations, we propose examining these properties with an eye to constructing a theory of readability. More specifically, we propose that readability can best be understood through three basic concepts related to textual comprehension: (i) linking of units of information, (ii) ambiguity, and (iii) background knowledge. Let us look a little more closely at each of these concepts.

What is true of sentences is also true of other aspects of texts. In particular, we will show that linking can play an important role in our comprehension of texts. So, for example, narratives can contain stories embedded within stories that are in turn embedded within stories. This can make it difficult for readers to link together units of information so that they can understand the text (see Chapter 5, “Domains”).

No matter what the source of the ambiguity, it is resolved by context in most cases. However, as we will see, when context does not resolve ambiguity (and sometimes even when it does), a text may be more difficult to read as a result.

Contextual knowledge is not static from our perspective. It includes not only the prior knowledge and assumptions that readers bring to a text, but also the inferences from the text that readers use in interpreting subsequent parts of the text. In Dostoyevsky’s (1978) novel Crime and Punishment, for example, readers use the information that the protagonist commits a murder in understanding subsequent parts of the novel. In many scholarly and scientific works, the meanings of words and terms are assigned specific definitions for use in the text. Readers use these technical definitions in understanding the rest of text. Although in some cases the kinds of inferences you can make depend on where the background knowledge comes from (see the discussion of garden path sentences in Chapter 3), for the most part we will not differentiate between contextual knowledge readers bring independently to the text and contextual knowledge they infer from elsewhere in the text. What you have read previously in a text and what you know independently of the text are both part of what you use to understand other parts of the text.

We will argue that if readers lack the contextual knowledge to understand the text, either because they did not begin with sufficient information or because they failed to understand parts of the text, the text may be more difficult for them to read. This is rather clear in the case of vocabulary. If readers do not know the meanings of many of the words in a text, that text will be difficult for them to read. In subsequent chapters, we will see, however, that this is by no means the only way contextual knowledge may impact readability.