Previous Next Up Comments
SEHR, volume 4, issue 1: Bridging the Gap
Updated 8 April 1995

beyond cognitivism: studying readers

David S. Miall

Begin with the meanings, Professor Simon seems to say, and just build out from there. Meanings are held in human heads, and heads contain neurons, so why not start to work out the structures and patterns that produce literary meanings? Simon's paper appears to deal with the fundamental components of meaning, but the specific models of meaning he develops in the paper are top-down, and for this reason they offer an inadequate map of the literary problems he is attempting to understand. As I will suggest in a moment, a bottom-up approach is likely to be more effective, but this points to a different role for artificial intelligence.

Much of Simon's paper is devoted to an overview of the kinds of knowledge relevant to the literary domain, from author's intention to the reader's world knowledge. Perhaps the most surprising feature of Simon's discussion is what he doesn't say: there are severe limitations on our present computing power to model the world knowledge required to understand a literary text-let alone the knowledge inherent in the various possible alternative critical views of it. This suggests that Simon is offering what Karl Popper termed a metaphysical research theory, that is, one that cannot be carried out in practice, although it may point to potential areas for actual research.

Thus there are two questions raised by Simon's contribution: first, whether the areas he nominates will be the most fruitful place to start (should we be concerned with author's intention? is the reader's imagery likely to be significant?); second, whether the models of analysis available from cognitive science are likely to be the right ones.

Two models commonly nominated for the task are schema theory (that is, one version of it or another) and story grammars (depending on which of the many alternative versions you prefer). Both involve presuppositions about what the significant elements of data will be: a necessary first step, in order for computational analysis to begin. Simon's account of schema theory well illustrates the limitations of applying a top-down approach of this kind to literature.

Reading, in this view, involves an interaction between two types of schemas: the prior knowledge of the reader or writer, upon which preliminary understanding of the text is based (the "context" schema), and the "problem schema. . . .that grows out of the information found in the text itself." In the case of fiction this would involve situation, characters, motives, and the like. "Together," Simon notes, "these schemas constitute the context in which the meaning of the text is interpreted"(11). This is what I would call a schema elaboration model of literary comprehension. It is inadequate for two reasons.

First, it prescribes the information that will be considered in literary understanding, limiting it to a narrow range of "cognitive" contents. Adding "emotion" as another kind of content, as Lenhert and Dyer have done, does not overcome this limitation. Second, while the schema or story grammar approach offers some purchase on a simple story such as Little Red Riding Hood, it only describes an aspect of the first phase of literary response. What is missing is what I might call a schema creation model. A literary text is distinguished from other kinds of texts, because it calls into question the adequacy of the schemas that the reader applies to understanding it. We know what meaning to attribute to the figure of Claudius, who has murdered his brother and seized the throne of Denmark; we understand the problem faced by Hamlet, who is procrastinating over bringing justice to Elsinore. What we do not know is what the action of the play as a whole means; for that, we have to create our own understanding (and the proliferation of critical interpretations of Hamlet suggests that we are each likely do this in our own terms). As we do so, we go beyond any of the schemas that we first called up to locate the "context" of the play and understand the "problem" it presents.

The cognitive mechanisms involved in instantiating and elaborating schemas are relatively well understood, but we have very little understanding of how a schema is created, as Vosniadou and Brewer (1987) have pointed out. In fact, the outcome of the process of responding to a literary text is probably only poorly represented by such a notion as a schema, if this concept is still tied to the cognitive science paradigm. The outcome of an effective response to Hamlet probably involves a range of phenomena, from the physiological to issues of identity and culture. What is required, if progress is to be made, is attention to all the phenomena that may be implicated in the process of reading literary texts and an ability to perceive the transformation processes at work in going beyond the initial schemas of the reader.[1]

Given the little attention paid to the literary response process by psychologists (more attention, however, than from the AI community), we have the proliferation of theories about literature of which Simon complains but almost no working knowledge of how the response process functions or even what tools we might bring to bear on studying it. In this situation a bottom-up approach is likely to be more effective: let us set ourselves to collect all the information that we can from as many different kinds of readers as we can about what is actually taking place during literary reading. No instrument of data collection is free of theoretical presuppositions, as we all know, but by attending as directly as possible to the phenomena of reading we may begin to perceive patterns that go beyond our existing preconceptions.

And it is here, once we have collected our data, that we should call upon the expertise of our colleagues in AI. Whether we employ the algorithms of machine learning, Bayesian logic, or some other heuristic, what AI also offers in various intelligent systems is the ability to perceive patterns within a complex, multivariate field of data. By taking a given text and contributing to the analysis all that we know about the style and structure of the text, as well as all that a range of readers can tell us under various conditions about their responses to that text, we may discover some of the underlying determinants of literary reading. I would invite Professor Simon to lend his expertise and his interest in literary issues to this undertaking.[2]

Previous Next Up Comments


1. This view is developed in (Miall, 1989) and (Miall and Kuiken, in press).

2. For a fuller account of this possibility, see (Miall, 1993: 333-339).