Literary Processing and Response as Transaction: Evidence for the Contribution of Readers, Texts, and Situations

[ as published in Comprehension of Literary Discourse: Results and Problems of Interdisciplinary Approaches, ed. Dietrich Meutsch and Reinhold Viehoff. Berlin: DeGruyter, 1988. [155]-174.]

The first step in building a bridge is to know where the two sides are. Working in the interdisciplinary span between the literary theorist's and the psychologist's ideas about literary experience -- the most productive place, we believe, to construct a model of how literary texts are read by engaged human beings -- it is important to begin with a clear map of the territory on either side of the divide.

The literary side, for most of this century, has been characterized by preoccupation with the literary text. New Critics tended to make an idol of the perfect text, "the verbal icon," in effect denying the reader's contribution to literary experience. More recently, many poststructuralist and reader-response critics have turned the tables, in effect idolizing a small group of brilliant and committed readers (often, themselves), but so minimizing the role of the text that works of literature become almost incidental to the needs and interests of individual readers. Regardless of whether it is texts or readers that are the object of study, however, scholars on the literary side have tended to spin their discourse at will, unconstrained by "hard evidence" or considerations of how most people actually do read.

On the psychological side of the divide, in contrast, there has been concern both with hard evidence and with the "ordinary" reader. The problem rather is that the texts used in traditional psychological studies have not been literary texts at all, but have been instead the sort of pragmatically truncated fragments that Hunt (in press) terms "textoids." It may be that such materials are used for legitimate control reasons, but they have the effect of ruling out genuine literary experience from the beginning.

In short, on neither the literary nor the psychological side has literary experience been treated in a way that accords a significant role both to normal readers and to naturally-occurring literary texts. Neither approach, that is, has succeeded in studying the reading of literature without slighting either the reading or the literature. To do so is a challenging and complex task, and an important one. It is, arguable, for instance (cf. Rosenblatt, 1981), that this richest and most engaged form of reading must be understood if we are to fully understand any reading.

Transaction

In attempting to build a bridge across the divide separating literary theory and psychology, we have found the concept of "transaction" to be a useful tool. The idea of the literary transaction was proposed most influentially, of course, by Louise Rosenblatt (1969, 1978, 1985), who in turn drew upon the pragmativist tradition in American philosophy and psychology, especially the thought of John Dewey (Dewey & Bentley, 1949; see also Ratner, Altman, & Wheeler, 1964). "Transaction" must be differentiated from "interaction," a distinction first made by Dewey and Bentley. At the risk of oversimplifying, it may be said that in an interactional approach constituents of a phenomenon are "analyzed out" and considered in isolation, whereas in a transactional approach the aim is to study "whole" phenomena, inextricably engaged with their contexts.

Applying the concept of transaction to literary experience, then, Rosenblatt insists that reader and text each contribute, in a specific situation, to a new entity -- "the poem." It follows that each reading; event is "a unique coming-together of a particular personality and a particular text at a particular time and place under particular circumstances" (Rosenblatt, 1985, p.104), and all of these -- reader, text, and situation -- need to be kept simultaneously in mind if one is to fully understand the phenomenon of literary reading.

Although the concept of transaction has achieved some degree of acceptance in modern literary theory (and, incidentally, in modern science; see Weaver, 1985 for a review), it has not been adopted widely among empirically-oriented reading researchers, perhaps because it seems to resist clear, operational definition and quantification. For many people, it seems, a transaction between two active, changing, social beings (between a mother and her infant, for example; Kaye, 1982;1 is easily conceived of, but a transaction between a reader and a collection of inkmarks on a page can seem merely a convenient fiction: no more, perhaps, than a fancy way of describing something that really, after all, is accomplished by the reader alone.

A growing body of thought, however, suggests that the notion of transaction is not so easily dismissed. "Ecological" theories of perception, such as those of Ulric Neisser (1976) and James J. Gibson (1979), recognize the contribution of both object and perceiver to the constructed perception. If one's concern is with the process by which that perception is constructed by an engaged reader and a literary text- in Rosenblatt's terms, how "the poem" is made- the transactional metaphor becomes, at the very least, a powerful reminder that all three participants in any instance of reading must be taken into consideration.

To study reading transactions without leaving any of these elements out of account, we have begun to develop a conceptual model of literary experience, and to derive from it a number of empirically-testable hypotheses. In the remainder of this chapter, we sketch the model, and then describe in some detail an experiment that tested some of its implications.

Three Modes of Reading

In brief, a reading event can best be understood as a transaction between a reader and a text, where the nature of this transaction is profoundly influenced by the situation in which the reading occurs ("situation" here includes at least the reader's understanding of the concrete situation, of the social context, and of any task demands that may be relevant). 'We have argued (Vipond & Hunt, 1984) that three main types or "modes" of reader-text transaction can be identified: In information-driven reading, the reader's goal is to acquire information from text, to carry something away from it. In story-driven reading, the reader becomes immersed in the storyworld of plot, events, characters, and settings. In point-driven reading, the reader sees the text as an intentionally shaped artifact, rooted in a social-pragmatic context; the reader's goal is to construct a version of what the narrator might be "getting at." (Of course, these modes should be seen as theoretical categories rather than as clearly differentiated practical realities. In practice, the different types seldom or never exist in pure form; they often overlap or succeed one another during a given instance of reading.)

The third, "point-driven" mode of reading is especially appropriate to what might be called the literary reading event, in which a competent reader reads a "literary" text in a situation that affords genuine engagement. This may be made clearer by considering an analogy between that situation and the sort of situation in which stories are told as part of conversation.

Evaluation Structure in Storytelling

In conversational storytelling, narrator and audience are socially engaged, situated in a specific, pragmatic context. In such a situation, the teller and the listeners do not exchange information so much as negotiate, construct, and share values, aptitudes, and beliefs -- in a word, "points." How do they do this? Sociolinguists such as Labov (1972) and Polanyi (1985) have shown that the negotiation of points is achieved mainly through the evaluation structure of the discourse, where evaluation is defined as any element of the narrative that calls attention to itself because it departs from the local norm of the text.

In literary reading, we suggest, the reader and the (implied) author are also -at least imaginatively -- socially engaged. That is, literary reading, like conversation, is not centrally a matter of transferring; information, but of negotiating and sharing beliefs, values, and attitudes. As we argue at length elsewhere (Hunt & Vipond, 1986), this is accomplished in much the same way it is in face-to-face conversation: through attention to the evaluation structure of the text. Three major types of evaluations may be distinguished in the literary reading situation: discourse evaluations, story evaluations, and telling evaluations. All of these are connected with the idea of incongruity, distinctiveness, or surprisingness with respect to the local norm of the text. They, differ, though, in what it is that is surprising: discourse evaluations can occur when the way something is expressed is surprising; story evaluations when some event in the storyworld is surprising; and telling evaluations when the fact that something is mentioned at all (or is mentioned at a particular moment) is surprising.

To summarize, then: A reading event is a transaction among reader and text, where the nature of that transaction is shaped by the situation in which the reading takes place. There are three main modes of reader-text transactions: information-, story-, and point-driven. Point-driven reading, like conversational story-listening, is characterized by attention to the evaluation structure of text.

Hypotheses

As noted above, an advantage of such a conception of the reading process is that it generates empirically-testable hypotheses. In the following pages, we report an experiment whose purpose was to test some of these. In general, the model suggests that whether information-, story-, or point-driven reading will predominate depends on the particular conjunction of reader, text, and situation that is operative for that reading event.

More specifically, if situation affects the likelihood that a given mode will predominate, then point-driven reading should be more likely when the readers' task encourages them to perceive the text as consistent with a concrete pragmatic context, whereas story-driven reading should be more likely when the task encourages them to attend to the events and characters of the storyworld. To test this hypothesis, in the present experiment half the readers were given a context or "frame" and asked to think about the story in relation to the frame, whereas the other half were given a task that promoted attention to the storyworld.

Second, if the likelihood of point- vs. story-driven reading is also a function of the reader., then qualitatively different groups of readers may tend to read a literary text in different modes. This assumption was tested by comparing the readings of more experienced readers (faculty members) with those of less sophisticated "novices" (undergraduates)..

Finally, if the predominance of the two modes is a function of the text, then presenting a literary text "whole", as compared to presenting it in a truncated form (as a textoid), should affect the way It is read. We tested this by giving half the readers a short story in its original, heavily "evaluated' form, whereas the other half read a version of the story in which many of its discourse evaluations were replaced by nonevaluative paraphrases.

Measures

It is all very well to generate hypotheses, but where and how does one find evidence for "point-driven" or "story-driven" reading? As Ernst Rothkopf (1972) wryly observed, "Anyone who has ever conducted an experiment on learning from written text is struck by the mute and unrevealing posture of the reader" (p. 331). Finding out about literary reading is particularly difficult, not only because readers of literary texts are equally "mute and unrevealing", but when they are asked to say something, what they say is shaped by knowledge about the "appropriate" rhetoric for responding to literature.

In an effort to overcome these difficulties, we could try to find "the" measure of literary reading. It seems a wiser research strategy, however, to use a variety of tasks and measures, and for each one to define a pattern of response that, in principle, implies "point-" or "story-driven" reading. Then, by looking for patterns of convergence among the different measures, we should be able to assess the reliability of any given phenomenon.

The measures; actually used are explained in greater detail below. For now it is sufficient to say that they range from those that are "mediated" by the reader's language (e.g., open-ended responses), to "nonmediated" ones (e.g., reading time). The measures can further be classified according to whether they are taken during or after reading; that is, whether they are "in-process" or "retrospective."

Description of the Experiment

The purpose of the experiment was to examine literary experience as a function of texts, readers, and situations. More specifically, we wondered whether variations in texts, readers, and situations would result in different proportions of "point-" and "story-driven" reading, as measured by a variety of tasks. The experiment consisted of a main study, using; undergraduate ("novice") readers, in which the independent variables included text and situation; and a subsidiary study, using more experienced ("skilled") readers, which allowed comparisons to be made across reader groups.

In the main study, the subjects were 96 male and female introductory psychology students who were fulfilling a course requirement. The subjects were randomly assigned to one of the 8 conditions defined by the factorial combination of Text (evaluated vs. paraphrased)-. Task (frame task vs. plot task), and Modality (silent vs. oral reading). All variables were between subjects, with 12 subjects per cell. (Note that in this experiment we varied one aspect of the situation only - the reader's "task." Furthermore, to limit the complexity of the experiment we did not use a task intended to promote the information-driven mode.)

In the subsidiary study, the subjects were 12 skilled readers who were all tested in the Evaluated text/frame task/silent modality condition. These readers were faculty members with an average of 13 years of university teaching experience. It is worth emphasizing that we label the faculty members "skilled" because of their academic training and experience; however, they were not necessarily experts in reading literature. In fact, each represented a different academic discipline: sociology, French, mathematics, religious studies, philosophy, and so on. Only one taught English literature.

Procedure

Overview. All the subjects were tested individually by a 25-year old male experimenter. The session began with a short structured interview concerning reading preferences. Then the subject read a short story, "The Day We Got Our Own Back," a short, realistic work by the Irish-American writer, Maeve Brennan (1969). Participants who were to read silently were asked to read as normally as possible, and specifically told that they could re-read any part of the story they, wished. Participants who were to read orally were given no special instructions. After every page of reading, all readers were given the Phrase Selection Task (described below), and then asked a "task question" pertaining to their assigned task, either frame or plot. After the final page of reading, the readers were asked open-ended questions concerning their immediate reaction to the story. They were then given the Probe, the Sentence Ranking, and two Highlighting tasks. More open-ended questions completed the session. All sessions were tape-recorded; selected portions were later transcribed.

Text manipulation. Half the participants read the story in its original, evaluated version. For the purposes of this study, we identified 20 sentences containing "discourse evaluations" -- places where the narrator conveys her attitudes or values by expressing something in surprising or incongruous language. For the other half of the participants, we prepared an alternate version in which the discourse evaluations were either replaced by nonevaluative paraphrases or deleted. The sentences containing discourse evaluations, and their paraphrases, are shown in Table 1. For example, in the fifth sentence listed, the evaluated phrase they camped around the room was paraphrased as they sat around the room. (It should be added that the paraphrases are "nonevaluative" only relative to the original evaluations; no linguistic item, of course, can be absolutely nonevaluative.) The two text versions were formatted so that page breaks were identical. The evaluated version was 4 pages (approximately 1,400 words) in length; the paraphrased version was 4 pages (approximately 1,375 words} in length.

Task manipulation: Frame group. Before being shown the story, half the readers were given a context or "frame" designed to increase the probability that they would read the story in a "point$:int-driven" way. The experimenter said: "Before reading, here's something else to read that your might find helpful in thinking about the story. It's part of a letter." The experimenter then handed to the subject the following "letter" (which in reality had been composed by theinvestigators). The experimenter read it aloud, while the subject read along silently.

I read a story the other day that really seemed to capture something important about the way families have to live when they have no power at all. Living here in East Germany, trying to get along under a military dictatorship, is so much worse than I could ever tell you -- for instance, the way you have to settle for tiny victories because you're too powerless even to imagine a large one. Anyway, the story is called `The Day We Got Our Own Back,' and it was written by the Irish writer, Maeve Brennan, about the Irish Troubles back in the 1920s, but it sure hit me hard. See if you can find a copy of it, and let me know what you think of it...

Task manipulation: Plot group. Before reading, readers in the plot group were told only the title of the story, its author, and her nationality. After every page, however, they were asked a task question designed to encourage attention to the story's events and characters; that is, to increase the probability that they would read it in a "story-driven" way. For instance, after page 1 they were asked, "The way this story is developing -- what kinds of things do you think might happen next?"

Dependent Measures and Results

We turn now to a more detailed descriptions of the tasks and dependent measures used in the experiment. Because each measure requires a substantial amount of description, it will be clearer to report the results for each one before going on to describe the next. The dependent measures can be grouped according to whether they were "in-process" (reading time, Phrase Selection Task) or "retrospective" (open-ended questions, Probe, Task, Sentence Ranking, and Highlighting). Another way of classifying the tasks is according to whether they were "mediated" by the reader's language (open-ended questions, Probe Task) or "nonmediated" (reading time, Phrase Selection Task, Sentence Ranking, Highlighting).

Measure l: Reading Time

Description. We considered only the reading times of those subjects who read silently (n = 48). Reading times were calculated by playing back the tape recordings of the sessions: Starting and stopping, times for each page could be inferred from the experimenter's comments. In a previous study (Hunt & Vipond, 1985), we found that students given a "frame" task had a different reading time pattern from those given a "read-for-details'" task. In the present study, therefore, we expected reading times would differ as a function of readers, text, and task.

Results. After being corrected for differences in page lengths, the reading times were subjected to a 4 (pages) x 2 (text) x 2 (task) mixed analysis of variance, with repeated measures on the first variable. (Data from one subject were lost.) The only significant effect was due to page, F (3,129) = 8.93. The mean speeds for the 4 pages were 242, 269, 277, and 269 wpm, respectively; page 1 was read more slowly than any of the other pages, which did not differ from one another. In a separate analysis, we found, surprisingly, that the reading times of the 12 skilled readers did not differ from those of the 12 novices who read the story in the same experimental condition. Also, in this analysis, there was no effect for page. In short, unlike the previous study, this experiment did not produce any reading time differences that could be ascribed to variations in readers, texts, or situations.

Measure 2: Phrase Selection

Description. Such was not the case with the second in-process measure, however. The purpose of this Phrase Selection Task was to determine whether, as predicted, readers are sensitive to the evaluation structure (more precisely, the discourse evaluation structure) of text. Readers were shown, after every page (immediately preceding the task question), a list of phrases drawn from that page, and asked. to say which ones they "noticed particularly" as they were reading.

Half the phrases were considered critical phrases, because they contained discourse evaluations (or, for the readers of the paraphrased text, their corresponding paraphrases). For example, readers of the evaluated version were shown they camped around the room; readers of the paraphrased version were shown they sat around the room. The other half of the phrases shown after each page were nonevaluated control phrases (e. g.., he pointed out a blue-glass bead), equated with the critical phrases for number of words. Critical and control phrases were always presented in the same random order, with the restriction that no more than two of either type could occur consecutively. Altogether 36 phrases were shown in this manner. Subjects could choose as many phrases as they wished, provided they did not choose them all.

Since the rationale underlying this task may not be immediately apparent, and since it is common to several of the tasks used in this experiment, it may be worth spelling it out in detail. If it is true that discourse evaluations are an especially salient aspect of text, readers should choose the evaluated phrases more often than the nonevaluated control phrases. For instance, they should select they camped around the room more often than he pointed out a blue-glass bead. Readers' "sensitivity" to evaluated phrases can thus be treated as a signal detection problem and computed as the d' statistic: every selection of an evaluated phrase is considered a "hit", and every selection of a control phrase is a "false alarm."

A problem is that sensitivity to evaluated phrases may be due to some other aspect of the phrase being salient. For instance, what if they camped around the room happens to be especially salient as a story event, quite apart from the language the narrator uses to express it in? A measure of its salience as a story event, however, is provided by subjects' choices of the paraphrase, they sat around the room. Consequently, the difference between d' scores for evaluated and paraphrased items is a measure of the salience of discourse evaluations, with story content controlled. We will use this logic repeatedly throughout this experiment, and will be referring to the difference between d' scores for evaluated and paraphrased items as the "evaluation effect". As well as predicting an overall evaluation effect, we thought the effect would be especially large for the more sophisticated readers, and for novices given the frame task.

Results. For the Phrase Selection Task, and each of the following dependent measures, two sets of analyses of variance were conducted. In the first set -which we will subsequently refer to as the "text-task-modality analysis" -- the experiment was considered a 2 (evaluated vs. paraphrased text) x 2 (frame vs. plot task) x 2 (silent vs. oral modality) factorial, with 96 novice readers as the subjects. In the second set of analyses -- which we will refer to as the "reader analysis" -- the 12 skilled readers were compared with the 12 novice readers who had participated in the same experimental condition (i. e., evaluated text/frame task/silent modality). These were therefore one-way analyses of variance, with ability (novice vs. skilled) as the single between-subjects factor. Unless noted otherwise, all the effects reported here are significant beyond the .01 level.

Text--task-modality analysis. Analysis of variance of the d' scores revealed a significant evaluation effect, F (1,88) = 28.13. Readers of evaluated text showed greater sensitivity (mean d' = .030) to the critical items than did the readers of paraphrased text (mean d' = -.019). In other words, readers of evaluated text tended to prefer phrases containing discourse evaluations to a greater extent than readers of paraphrased text preferred the corresponding paraphrases. This finding supports our contention that discourse evaluations are a salient aspect of literary texts.

The analysis also revealed a significant three-way interaction involving text, task, and modality, F (1,88) == 4.14, p < .05. Although the evaluation effect was weaker for the frame readers who read silently than it was for any of the other three combinations of task and modality, the generality of the main conclusion -- that discourse evaluations are salient -- is not seriously compromised by this result. No other main effects or interactions were significant.

Reader analysis. There was no significant difference between skilled and novice readers on this task.

After completing the Phrase Selection Task for page 4, the readers were asked three general, open-ended questions, to which they made oral responses: (1) What do you make of this stow? (2) Did you like it? What did you like (dislike) about it:? (3) Was the way you read it similar to how you usually read? The frame readers only were further asked if they were thinking at all about the connection between the letter and the story as they were reading. They were then invited to glance over the story again, after which they were asked if they now felt the story had something to do -with the letter-writer's concerns. Finally, they were asked if they saw a connection between the "tiny victories" mentioned in the letter and the story.

All responses were transcribed. Scoring procedures and results will be treated together with those of Measure 7, delayed open-ended questions.

Measure 3: Probes

Description: All readers were then shown and read a list of 9 statements "that people have made about this story." After each probe was read, the subjects were asked to agree or disagree, "strongly" or otherwise. Finally, they were asked to explain their choice. All responses were made orally. The probes were always given in the same order.

Scoring and results. Responses to this task were later converted to a 5-point scale, with 1 indicating "strongly agree" and 5 "strongly disagree." The statements were scored so that the more the person agreed (low scores), the closer they were to a hypothetical "story-driven" reader, and the more they disagreed (high scores), the closer they were to a hypothetical "point-driven" reader. We expected that evidence of a more point-driven type of response would be found for skilled readers, as well as for novices who were given either the frame task, the evaluated text, or both.

Text-task modality analysis. Here we mill overlook responses to the separate probes, and consider only the overall probe score, which is the average of the 9 responses. On this global score there was a significant main effect for task, F (1,88) = 4.78, p< .05. Readers given the frame task were in greater disagreement (i.e., were more "point-driven") than the readers given the plot task (the means were 3.56 and 3.36, respectively). Thus, as predicted, the situational manipulation of embedding the text in a pragmatic frame had the effect of producing more "point-driven" responses on this task. No other main effects or interactions were significant.

Reader analysis. Skilled readers disagreed with the probes to a significantly greater extent than did the novices, F (1,22) = 8.49. The means were 4.02 for the 12 skilled readers, 3.51 for the 12 novices.

Measure 4: Sentence Ranking

Description. The subjects were then shown a list of 10 sentences. Half the sentences were considered "critical", because they contained discourse evaluations (or, for readers of paraphrased text., their corresponding paraphrases); the other half ware nonevaluated control sentences, equated in length with critical sentences. (Portions of each of the 10 sentences had appeared on the earlier Phrase Selection Task, so mere exposure was controlled.) Critical and control sentences were presented in one of 4 random orders, with the restriction that neither type could appear more than twice in succession.

The subjects were instructed to glance over the entire list, and then to choose the one sentence "you feel most strongly about." Then they chose the one they felt next most strongly about, and so on, until 5 sentences had been chosen. Then they were asked, for each sentence selected, what it was that "struck them particularly" about it. Finally, the subjects were asked to indicate, of the remaining 5 sentences, which one they felt least strongly about, and so on, until all 5 had been chosen.

Scoring and results. A score of 10 was subsequently assigned to the sentence the subject felt most strongly about, 9 to the sentence the subject felt next most strongly about, and so on. Therefore the range of possible scores for the 5 critical sentences was 1 5-40, with higher numbers indicating a greater tendency to select critical sentences.

Text-task-modality analysis. Analysis of the critical sentence scores revealed only one significant effect: a text x task interaction, F (1,88) = 4.85, p < .05. The tendency of frame readers who had read evaluated text to prefer evaluative sentences (mean critical sentence score = 34.58) was stronger than the tendency of frame readers who had read paraphrased text to choose paraphrase sentences (mean = 32.17). This familiar evaluation effect, however, was present only for frame readers. For the plot readers, paraphrased items (M = 34.21) were preferred to their evaluative counterparts (M = 33.29). Thus, the effect of text differed depending on the situation in which the reading took place: readers who were invited to read the text "pragmatically" were more sensitive to discourse evaluations, whereas those invited to focus on the storyworld were less sensitive to them.

Reader analysis. There was no significant difference between novice and skilled readers on this task.

Description. Then the experimenter showed and read the subjects a statement, allegedly made by a person who had read the story:

Right from the first page the main feeling I get from this story is how powerless they all are. I think the language that the author uses gets this feeling across very effectively.

The experimenter then asked the subject to look back over the first page of the story, and to use a highlighter to mark out any, words, phrases, or sentences that conveyed this feeling of powerlessness. The rationale of this task is similar to that of the Phrase Selection Task: we expected evaluated phrases to be highlighted more than paraphrases, especially by skilled readers and by novices given the frame task. Unlike the Phrase Selection Task., however, the present one occurs after, not during., the initial reading, and the reader's attention is directed more forcefully to the language used.

Scoring and results. In order to score the highlighted selections, we divided the page at major syntactic boundaries into 45 units. Eight of these units were considered critical because they contained discourse evaluations (or their corresponding paraphrases). A given unit was considered selected if all or any of it was highlighted. Sensitivity (d') scores were calculated as before, by considering that selection of .a critical unit constituted a "hit," and selection of a noncritical unit constituted a "false alarm."

Text-task-modality analysis. Analysis of variance of the d' scores indicated a significant evaluation effect, F (1,88) = 12.23. Readers of evaluated text were more sensitive to the critical units (M = .046) than were readers of the paraphrased text (M = -.001). This finding is consistent with that of the Phrase Selection Task; it again indicates than readers found evaluated language particularly salient. No other main effects or interactions were significant.

Reader analysis. The difference between skilled and novice readers was not significant.

Description. After the Highlighting of Powerless Language Task had been completed (i.e., page 1), subjects were asked to carry on for the remainder of the story (pages 2-4), this time highlighting "anything at all that stripes you or catches your eye". They were told specifically that they were no longer being asked to highlight "powerless" language- "just whatever strikes you", although it is quite likely that some subjects would have been sensitized to "powerless" language by the previous task.

Results. The rationale and scoring procedures for this task were the same as for the previous; one. The total number of units was 101,12 of which were considered critical.

Text-task-modality analysis. Analysis of variance of the d' scores again indicated a significant evaluation effect, F (1,88) = 9.44. Readers of evaluated text were, as expected, more sensitive to the critical units (M = .046) than were readers of paraphrased text (M = 0.13). No other main effects or interactions were significant.

Reader analysis. The difference between skilled and novice readers was not significant.

Description. At this stage in the experiment the subjects were asked, "What do you make of the story now?" Their oral responses were transcribed into a microcomputer database, and, for purposes of this analysis, grouped together with their responses to two of the questions asked earlier in the session, immediately after reading: "What do you make of the story?" and "Did you like the story? What did you (dis)like about it?" The reason for grouping the three responses together was that we wished to obtain as large and reliable a database as possible. It should be noted that the responses of the skilled as well as the novice readers were included together in a single database.

Following suggestions made by Ericsson and Simon (1984), the responses were handled in a two-stage process, editing and coding.

Editing stage. One "editor" eliminated miscellaneous statements and all other comments that did not refer to the text. Second, he divided syntactically compound and complex statements into simpler syntactic units. For example, the response "I found the story interesting and amusing, but I don't like the way it ends," would be divided into three separate statements (see below). Third, the editor replaced pronouns with noun phrases, ;end in general restored context as necessary so that each statement could be understood on its own, out of context. For example, after editing, the above statement would consist of the following:

1. I found the story interesting
2. I found the story amusing
3. I don't like the way the story ends

The statements were printed in random order and given without identifying marks to a different person for coding. This coder was not aware that some of the statements had been made by skilled readers.

Coding stage. The next step was to categorize the statements according to a content analysis system; for this we used a taxonomy that we, in collaboration with Garry Hansen, have been developing recently. The taxonomy is based in part on that of Purves and Rippere (1968), the main difference being that ours is a two-way scheme that maintains a distinction between element types (i.e., which aspect of text is referred to), and response types (i.e., how it is referred to). The types are chosen to reflect a range of content that, theoretically, is associated with "point:-driven," "story-driven," arid "information-driven" modes of reading. Currently, the element types are: author; theme; style; character; plot; text; setting; anti ideology (the latter two were not needed in this experiment). The response types are: description; inference; comparison; positive, negative, and indeterminate reaction; positive, negative, and indeterminate assessment; affirmation; denial; unsure (the latter three were not needed in this experiment). To illustrate, the statements above would be coded as follows:
 

Notice that the difference between a "reaction" and an "assessment" is that reactions use past tense verbs (that is, they refer implicitly to the reading experience), whereas assessments use present tense verbs (suggesting that the reader is considering the text in a more analytic way).

Results. One person coded all 1704 statements. A second person coded 20 of them as a reliability check. Reliability wars found to be .731, indicating that the categories are not as well-defined as we would wish, and that more development work is needed. For this reason we will not attempt a thorough analysis of the findings, but limit the discussion to some overall differences between skilled and novice readers -- differences that are generally robust enough to override any uncertainties there might be about the exact boundaries of some of the categories.

In brief, skilled readers talked more, talked about different things, and talked about them differently. The more sophisticated readers were far more loquacious than the novices, making on the average about 35 statements as compared to about 14 for the younger readers. Moreover, the two groups tended to refer to different text elements, X² (5) = 105.2. As shown in Table 2, 32 % of the skilled readers' statements referred to author, theme, or style elements, as compared to less than 13% of the novices' statements. This is consistent with the expectation that the more sophisticated readers would, on the average, read in a more "point-driven" way than the students. The students, on the other hand, made more than twice as many plot statements as the faculty members, as would be expected if they read in a more "story-driven" way than the older readers. Contrary to expectation, however, novices did not make a greater proportion of references to character than the skilled readers.
 

Also shown in Table 2 are the percentages of response types made by skilled and novice readers. Again, the pattern is different for the two groups, X² (8) = 65.8. Perhaps the most notable difference concerns the total number of reactions and assessments. Whereas novices made only half as many (present-tense) assessments as (past-tense) reactions, skilled readers made about the same number of each. This suggests that the experienced readers, to a much greater extent than the novices, treated the text as an artefact, as "point-driven" readers would be expected to.

Summary and Discussion

Since the experiment used seven different measures and obtained a variety of results, it may be helpful at this stage to summarize them in the form of a table. As Table 3 shows, there were statistically significant interactions involving text and task on two of the measures (Nos. 2 and 4). Furthermore, text had a statistically significant effect on three measures (Nos. 2, 5, and 6), reader had a significant effect on two measures (Nos. 3 and 7), and task (our operational definition of situation) had a significant effect on one measure (No. 3). Thus, even though each independent variable did not have a significant effect on every dependent measure, we interpret the overall pattern of results as consistent with our original proposal: Namely, that reading should be looked at as a transaction among readers and texts and situations, rather than breaking it down into components and studying one component as if it could be understood in isolation from the others.

Concerning more specific aspects of the results, we note the relatively large number of reliable effects due to text. On both in-process and retrospective measures, readers demonstrated that they are sensitive to the language of the text; specifically, to what we have termed "discourse evaluations." In opposition to the recent theoretical movement among reader-response critics, then, our results indicate that it would be misguided to ignore or minimize what is clearly an important participant in the creation of literary experience.

On the other hand, it would be equally shortsighted to attribute all power to the text: We see in Table 3 that literary experience is also, depending on the measure used, a function of readers and situations.

Reader effects were found only for probes and open-ended questions, both of which are retrospective, mediated measures. From such results it would be possible to conclude (S. Straw, personal communication) that the process of reading is the same regardless of expertise, and that differences in skill or sophistication become important only at the tune of post-processing. It should be emphasized, however, that with only 12 subjects per group, the statistical tests for the "reader" analyses lacked power. Moreover, since there are strong intuitive reasons to believe that there are processing differences between skilled and novice readers, the question is best considered, we think, an open one.

As a main effect, "situation" was statistically significant only on the Probe measure. Of course, it should be recalled that only one aspect of the total situation was manipulated: the readers' task (that is, half the readers were asked to consider the story in light of a context, the letter, whereas the other half were asked to attend to the storyworld). Perhaps the reason task was not more influential is that the situation of the experiment itself had powerful effects. By "situation of the experiment" we mean the overarching social and physical contexts in which the study as a whole was embedded. The experiment was sponsored by the psychology department, it took place in a psychology laboratory, the subjects were reading an assigned text, and they received course credit for their participation. Such often-neglected situational factors may have overridden the task manipulations we tried to introduce.

Since five of the seven dependent measures were "retrospective" in nature, it is not surprising that most of the significant effects were found on these, rather than on the "in-process" measures. In any case, there can be no doubt that there are important post-processing differences due to readers, texts, and situations. The importance of processing differences, however, is unclear at the present time.
 

To decide this question will require a greater range of in-process measures than used here. We believe that finer-grained analyses of reading times, think-aloud protocols, and oral reading miscues (Goodman, 1969; Hunt, 1985) all show promise as in-process measures of literary reading. Regardless of which particular measures eventually prove most useful, it seems clear that the basic theoretical stance suggested here-viewing reading as a transaction among reader, text, and situation - will continue to be a productive idea, not least because it powerfully reminds us that to study texts in isolation from readers; and situations, or to study readers without considering texts and situations, is to study only a part of reading. It is no accident, then, that the idea of transaction helps build a bridge between the literary theorist's and the psychologist's ideas about literary experience. So far, at least, it seems to be a bridge that can carry both theoretical and empirical traffic.

Acknowledgements

This work was supported by grant 410-85-0612 from the Social Sciences and Humanities Research Council of Canada. We are grateful to the students and faculty members who participated in our experiments, and Bill Toner, who conducted them. Correspondence may be addressed to the authors at the Departments of Psychology and English, respectively, St. Thomas University, Fredericton, New Brunswick, Canada E3B 5G3.

References

Brennan, M. (1969). In and out of never-never land: Twenty-two stories. New York: Charles Scribner's Sons.

Dewey, J., & Bentley, A. F (1949). Knowing and the known. Boston: Beacon Press.

Ericsson, K. A., & Simon, H. A. (1984). Protocol analysis: Verbal reports as data. Cambridge, MA: MIT/Bradford.

Gibson, J. J. (1979). The ecological approach to visual perception. Boston: Houghton Mifflin.

Goodman, K. S. (1969). Analysis of oral reading miscues: Applied psycholinguistics. Reading Research Quarterly, 5, 9-30.

Hunt, R. A. (1985, June). Pragmatics and miscue analysis in literary reading. Paper presented at the annual meetings of CELT, Bloomington, Indiana.

Hunt, R. A. (in press). A boy named Shawn, a horse named Hans: The Herr von Osten model of responding to student writing. In C. Anson (Ed.), The discourse of commentary. Champaign-Urbana, NCTE.

Hunt, R. A., & Vipond, D. (1985). Crash-testing a transactional model of literary reading. Reader, 14, 23-39.

Hunt, R. A., & Vipond, D. (1986). Evaluations in literary reading. TEXT, 6, 53-71.

Kaye, K. (1982). The mental and social life of babies: Hoes parents create persons. Chicago: University of Chicago Press.

Labov, W. (1972). Language in the inner city: Studies in the Black English Vernacular. Philadelphia: University of Pennsylvania Press.

Neisser, U. (1976). Cognition and reality: Principles and implications of cognitive psychology. San Francisco: W. H. Freeman.

Polanyi, L. (1985). Telling the American story: A structural and cultural analysis of conversational storytelling. Norwood, NJ: Ablex.

Purves, A., & Rippere, V. (1968). Elements of writing about a literary work: A study of response to literature. Urbana: NCTE.

Ratner, S., Altman, J., & Wheeler, J. E. (Eds.) (I964). John Dewey and Arthur F. Bentley: A philosophical correspondence, 1932-1951. Near Brunswick, NJ: Rutgers University Press.

Rosenblatt, L. M. (1969). Towards a transactional theory of reading. Journal of Reading Behavior, 1 (1), 31-49.

Rosenblatt, L. M. (1978). The reader, the text, the poem: The transactional theory of the literary work. Carbondale: Southern Illinois University Press.

Rosenblatt, L. M. (1981). On the aesthetic as the basic model of the reading process. Bucknell Review, 26, 17-32.

Rosenblatt, L. M. (1985). Transaction versus interaction: A terminological rescue operation. Research in the Teaching of English, 19, 96-107.

Rothkopf, E. Z. (1972). Structural text features and the control of processes in learning from written materials. In J. B. Carroll & R. O. Freedle (Eds.), Language comprehension and the acquisition of knowledge (pp. 315-335). Washington, DC: V. H. Winston.

Vipond, D. & Hunt, R.A. (1984). Point-driven understanding: Pragmatic and cognitive dimensions of literary reading. Poetics, 13, 261-277.

Weaver, C. (1985). Parallels between new paradigms in science and in reading and literary theories: An essay review. Research in the Teaching of English, 19, 298-316.