In common with many other accounts of electronic-driven change, Johnson-Eilola's account suggests that a certain technological imperative is at work: the introduction of the computer will force teachers to rethink their practices. The computer obliges us to examine our definitions of reading and writing: "it opens those definitions up to debate and change" (Johnson-Eilola, p. 204). While this perspective is common, it may also be misleading. The issue should perhaps be put in a different way: given what we know about reading and writing, and the mental processes that underlie them, how effectively does hypertext embody those processes in electronic terms? To what extent does hypertext change the processes, or promote some component process to a more prominent role? Put this way, the issue moves away from the implications of facilitating a technological development that has come to seem inevitable, however desirable that development may turn out to be. Instead the question is how well do we currently understand those underlying processes. Until we have some convincing answers to this question the impact of hypertext on reading or writing must be unpredictable: we cannot be sure whether we are supercharging the process or throwing a monkey-wrench into it.

The overview of Johnson-Eilola and the chapters by Charney and Smith provide only a tentative grasp on the problem, principally because the theoretical frameworks to which they appeal have their own limitations and cannot with confidence be generalized to situations in which we read or produce hypertext. Thus Kintsch's work, discussed by Charney and Smith, has considerable empirical validity in accounting for readers' encounters with expository prose or simple narratives, but its relevance to other kinds of text, including literary, is dubious. In the light of the theory, Charney shows why one might have strong reservations about the effectiveness of hypertext, but some hypertexts almost certainly invoke modes of response which Kintsch's theory is not designed to explain. Kintsch and his colleagues deal with "thin" cognition, in Smith's term. On the other hand Langer's work, to which Smith appeals, is based on an aesthetic theory whose principles are violated by almost every claim made by creators of the more inventive hypertexts. Smith's appeal to Langer and to "thick" cognition is no more than an interesting speculation: while Langer provides an intelligible and suggestive philosophy of the body and feelings as well as thought and reason, we know rather little, as yet, about the psychology of such processes (for a discussion, see Miall, 1995).

Is hypertext, as Dobrin puts it, only another form of text (p. 308)? It has often been pointed out that hypertext users may be duplicating in a different medium the non-linear reading that we accomplish in consulting an encyclopedia or following footnotes (Johnson-Eilola, p. 201). Cross-referenced or annotated structures of information can be built in a computer environment, but it is far from clear whether readers perceive them or can use them in the same way as a printed text, as several of the studies cited by Charney show: readers find it difficult to tell where they are within a group of nodes; they cannot judge whether they have read something essential and tend to give up too early; they find it hard to decide on an appropriate sequence through material (Charney, p. 249). The physical structure of the printed book may give readers in search of information a better framework for a search or learning strategy. On the other hand, giving readers more familiarity with a particular hypertext allows them to use it more effectively over time.

Charney points to some of the basic problems of reading hypertext: it imposes a greater demand on short-term or working memory; readers may find navigation becomes arbitrary through a lack of cues to the meaning of links between nodes; and hypertext may disable the reader's existing knowledge about how texts are structured and about different text genres. And if we are to depend on hypertext presentations for learning, hypertext information cannot simply be mapped onto long-term memory structures, as Charney points out (p. 243), even if they do have comparable network structures (but this in itself is a contentious proposal).

Charney's account is limited to an information processing model of reading and learning, the type of reading which has been most extensively studied and theorized. "I focus," she says, on the application of hypertext for readers "who read to learn, to understand and evaluate the ideas and argument of others, to come to realizations about the subject matter," and to integrate their learning with what they already know. This, according to Charney, is already a process which will "push hypertext to its logical extreme" (p. 241). In this view the prospect of hypertext models for literary reading seems quite out of reach, a consideration which is not addressed by Charney (and hardly mentioned by any other author). But Charney's central and most compelling point is that hypertext designers currently work largely in ignorance of what is known about the reading process. To support this contention, she appeals to the most influential work in text theory by Kintsch and Van Dijk, Meyer, and others -- what is often known as the schema theory of reading -- showing that the reading and memory processes engaged during learning or information seeking "are both strongly conservative forces" (p. 259). Hypertext systems that are premised on creative and imaginative linking thus conflict with the familiar, systematic processes of reading, and would appear to be inherently ineffective (p. 259).

Here, however, the limitations of the text theories described by Charney come into view. The strictly cognitive account of Kintsch et al. gives no purchase on other modes of response such as feeling, imagery, or kinaesthetic and bodily forms of response (see Miall and Kuiken, 1994), all of which are probably implicated in literary reading. The "shadow life of reading" described by Birkerts, for instance, cannot be explained by a theory at the level of propositions and macro-structures. It is possible that hypertext of the right kind might evoke the self-referential and feeling-imbued responses that persist in the unconscious, as Birkerts suggests; or that hypertext could initiate the self-focused agenda of issues that sustains reading through interruptions and across long periods of time. This would correspond to the "thick" cognition that Smith envisages.

In the text theory approach, Smith points out, the reader is conceived of as "an external operator performing operations on free-standing objects of thought" (Smith, p. 279), a model that Smith describes as Cartesian. She accuses hypertext designers of confinement to the same paradigm, hence her appeal to the alternative framework of Langer in which subject and object, thinker and thought, are merged. Interestingly, the branch of neuropsychological research that has just begun to provide firm evidence for this kind of thinking has also declared itself anti-Cartesian: a recent book in the area is titled Descartes' Error (Damasio, 1994). Ironically, however, the convergence of contemporary postmodern theory with hypertext (to use Landow's term) is antipathetic to precisely the construal of the reading self to which Smith points, since it explicitly deconstructs the Kantian idealism that lies behind it. Nor does it find text theory, as developed by Kintsch, any more congenial, since in this approach mind and text are treated as separate entities. This violates a cardinal feature of postmodern theory, the attribution of agency to language or related cultural forces (e.g., the semiotics of power, gender, or race).

Smith argues for a kind of deep reading through hypertext, postulating "thick" cognition as the appropriate matrix (p. 265). But she then appeals to the more recent discourse model of Kintsch (first published in 1988). Her application of this to hypertext, however, is problematic. In his model of reading, Kintsch proposes two phases: a construction phase, during which each proposition we encounter will spark a network of associations both relevant and irrelevant; this is followed by an integration phase in which associations that are mutually supported and reinforced are selected, while weak, unsupported associations are pruned. She proposes seeing hypertext as a "Kintschian knowledge net." How would a reader's approach to it facilitate the integrative phase of response? The reader poses a query:

Imagine an activation process, stimulated by the new query or by the history of previous interests, rippling through the net, revealing local 'hot spots' of relevance, showing in high relief the portions that pertain. The overall result . . . shows large-grain patterns of relations elicited by a query . . . In this way, the user's quest drives the integration . . . (p. 275)

The logical error in this is perhaps obvious. Whereas Kintsch provides a model to explain the reader's first encounter with a text, Smith's description presupposes a reader who has already read all the nodes in the hypertext and needs only a specific query to mobilize the integration phase. This assumes, then, that readers will read a whole hypertext in an unmotivated fashion in order to set up the weak associative network on which Kintsch bases the construction phase -- a highly improbable scenario. This is her model of how future hypertext systems will work: examples already exist, she says, pointing to the medical domain, and "a dynamic medical therapeutics handbook for physicians' use" (p. 276). Unfortunately, this example subverts her claim: as a system used by physicians who are already highly expert, it represents a domain that is well-formed and delimited (as expert-system developers insist is essential for any automated advice tool). It is remote from the first reading of an associatively structured hypertext by readers with no prior knowledge.

Dryden's account of hypertext is vulnerable to some rather similar misconceptions: he asserts that "In its structure of branching links and nodes, hypertext simulates the mind's associative processes, thereby providing an electronic platform for constructing and recording the reader's literate thinking" (p. 285). While hypertext might play such a role in the writing process, which is one of Dryden's main concerns, as an account of the reading of hypertext (which Dryden seems to include: he refers to users interacting with texts), hypertext can of course do no such thing. The author's associations are not those of the reader. As Dobrin puts it, in terms of hypertext linking, "The author's conception of the connection's relevance is not the reader's, and the reader gets lost" (p. 310). Association, moreover, will probably not turn out to be a particularly helpful way of construing the mental processes at play during reading or writing. We have, after all, been this way before: eighteenth-century associationism was challenged by Kant and Coleridge; its twentieth-century version in behaviorism has been challenged by several versions of cognitive theory. It hardly seems a promising basis on which to build a theory of hypertext.