In Weapons and Hope, Freeman Dyson describes a practical first test of any new piece of artillery. The test consists of taking the gun up on a tower, some hundred feet off the ground, and dropping it to the pavement below. Only a gun that could still be fired after this crash test would be considered sturdy enough for further development. Similarly, with ideas. Although sitting in one's armchair and generating ideas is an important activity, it is also important, we suggest, to ask how well these ideas stand up when they collide with hard evidence. In the first part of this paper we will present a model of literary reading, and in the second part we will describe some experiments that can be viewed as ways in which we have tried to drop our model onto a sidewalk. In other words, we will try to show how ideas evolving primarily out of literary and rhetorical theory can be tested by means of research strategies developed in cognitive psychology and education.[as published in Reader No. 14 (Fall 1985), 23-39]
First, some theoretical background. Our work represents a transaction between two disciplines literary scholarship and criticism on one side, and cognitive psychology and reading research on the other. In the past these disciplines have had very little to say to each other, and that little has often been antagonistic. From the point of view of cognitive psychology, literary theory seems preoccupied with airy, introspective speculation, with a kind of discourse-spinning which is essentially not constrained by considerations of how the real world is. On the other hand, from the point of view of literary theory, cognitive psychology seems preoccupied with number-crunching, with trivial experimentation, and (most damning of all) with inane texts.
In trying to steer a middle course between these two caricatures, we have found Louise Rosenblatt's ideas about "transaction" provide a basis on which to build a model of literary reading which is fairly detailed and realistic, but at the same time empirically testable. We hope this model moves closer to an understanding of what happens in the real world when readers sit down with works of literature than either literary theory or reading research alone has been able to do. One way of putting it is to say we're trying to find ways to study the reading of literature without throwing out either the reading on one hand or the literature on the other.
Let us sketch our model as it stands at the moment. Here, we describe it conceptually rather than in the order in which it actually developed; it has grown sporadically, along with the experimental work, not in this neat orderly fashion, and not in advance of the testing, however efficient that sort of development might have been.
Most fundamentally, we propose that any reading event is a transaction among three mutually transforming entities: The reader, the text, and the situation. This is illustrated in Figure 1.
1. The reader. To some extent, readers come pre-formed to a reading event. They have expectations (about how literature works, for example, or about testing and educational situations); they come with different kinds and levels of knowledge; they come with a history of previous encounters with texts. But none of these are impervious to the impact of text and situation, and none remain unchanged through a particular reading.
2. The text. We take a constructivist view that texts do not exist in any objective sense (except as ink-marks on a page), but instead are constructed in process by readers. On the other hand, there are, we think, some identifiable text features that play a crucial role in the transaction. We are trying to think about how such features can give the text a role in the reading transaction without (as the New Critics did) enthroning it.
3. The situation. By situation we mean to include a range of forces external to the text and the reader. Texts are read in immediate physical and environmental settings; in various social situations; in various intellectual contexts; with different task demands. All of these play roles in the transaction.
Figure 1. Three components of the reading process
In short, we believe it is useful to think of the reading process as resulting from a complex set of transactions among three mutually transforming entities reader, text, and situation. Like other students of language (e.g., McCormick, Verbrugge), we have been impressed with the work of the psychologist James J. Gibson. Gibson (127-43) coined the term, "affordance," which is helpful here, because it allows one to suggest the ways in which texts and situations invite and constrain but do not determine the reader's activities.
Although all this can sound pretty loose and soupy, there is some structure to be found in the dynamic relationships among these conceptual entities. We propose that specific instances of reading tend to fall into three general modes which can be labeled, "information-driven," "story-driven," and "point-driven." In most instances one mode tends to be dominant, though elements of the other two are present also; this can be pictured as a sort of Venn diagram (see Figure 2). The mode which becomes dominant in any instance is not merely a function of the reader's purpose, or the nature of the text, or the situation, but of a transaction among all three. These three modes are described in more detail elsewhere (Vipond and Hunt), so we will characterize them fairly briefly here.
Figure 2. Three modes of reading
1. Information-driven reading. This is the sort of reading which most traditional investigations of reading comprehension have studied. It is concerned with what, as Louise Rosenblatt says, you carry away from the reading; she calls it, in fact, "efferent" reading (Reader, 24). A reader who is primarily after "the facts" of the text, who wants to walk away with new information, will tend to use this mode. It, like the others, may be used with any text, but certain texts for instance, a telephone book afford it. It may be used in any situation, but certain situations for instance, a classroom or a psychology laboratory afford it. It may be used by any reader, but certain readers may be more likely to use it for instance, anyone trained by Mr. Gradgrind. (We are not, by the way, contemptuous of this mode; it is highly complex, highly useful, and often appropriate perhaps even appropriate most of the time. But it is not usually appropriate in "literary" reading situations.)
2. Story-driven reading. A "story-driven reading" is characterized by the reader's desire to enter imaginatively into a storyworld. It is concerned primarily with vicarious experience. Rosenblatt, in defining "aesthetic" reading, says that in contrast to "efferent" reading, it is concerned with "lived-through experience." Again, it may be used with almost any text (though clearly a phone book would block it very effectively, whereas the latest Danielle Steele novel affords it); and it may be used in any situation (though a Lazy-Boy in front of a fire affords it better than a fluorescent-lit, plastic school desk, and a voluntary reading affords it better than an assignment). Similarly, a reader may have a pre-existing disposition to use this mode, especially if he or she has often felt comfortable and successful with it in the past.
3. Point-driven reading. This is the kind of reading we think is most clearly afforded by texts of the kinds usually called literary. It is also, by far, the rarest, at least among our readers. The term "point" is adopted from work done by sociolinguists such as William Labov and Livia Polanyi on conversational storytelling. When people listen to stories in ordinary conversation, they expect the teller will be making a point, or a number of points. They expect the teller will be "getting at" something. As we use the term a point is not equivalent to a "moral" or a "theme" or the "gist." Usually points are difficult or impossible to put into words, and we feel their existence most strongly when they're absent we recognize it immediately when the narrative seems "pointless." In that case we are tempted to respond with what Labov says is the question every narrator is always fending off: "So what?" (366).
Point-driven reading is directly analogous to this kind of listening. It is the sort of reading you do when you see the text as an intentional, purposeful piece of discourse; in which you try to construct something the narrator might be "getting at"; and, perhaps most important, in which you attempt imaginatively to share a point (or points) with the narrator and a community of other readers.
This brings us to the last aspect of the model to be described here. What we are particularly interested in at the moment is the way in which someone reading in a point-driven way accepts the invitations of a text and a situation to construct a point, and most specifically, the role in that transaction of what Labov and Polanyi have taught us to think of as "evaluations." Evaluations, as they use the term, are the features of conversational narrative that signal how the storyteller herself feels about an event, character, utterance, or other story element. The narrator encodes these evaluations in a wide range of ways, but in general it seems that evaluations are fore-grounded by their unexpectability. Almost any text element which is so marked (e.g., figurative language, a striking word choice or sentence structure, an unexpected event or utterance, or even an unexpected omission) can serve as an occasion on which the reader can share an evaluation with the narrator.
In the second part of this paper we will have some examples of how this works when we describe some specific experimental tests we've conducted of our model. Right now it is enough to say that we believe that such acts of evaluation are centrally important to point-driven reading, just as Labov says evaluations are crucial to conversational stories.
Some Implications of the Model
An obvious question at this stage might well be, So what? What is this model for? What makes us think it's any different from speculative discourse anyone might idly spin about literary texts and readers? Our response which we try below to make a little more concrete is two-fold. First, we submit that this model affords the generation of a rich array of hypotheses about readers' actions, hypotheses that may or may not survive the drop to the pavement. Second, we would argue that the model offers us some conceptual tools for accounting for what readers actually have done with literary texts during the process of reading.
In general terms, our strategy can be characterized in this way: if this model really does describe anything important about the reading process, we ought to be able to cause changes in that process by varying the nature of the elements represented by the three arrows in Figures 1 and 2. In other words, it is useful to ask what might happen to that process if the arrows were pushed selectively: what if we introduced into it discernibly different readers? what if we varied the texts? what if we manipulated the situation?
What follows is an overview of three such "crash tests" we have performed recently. Our basic approach is easily described. The participants in our experiments are generally (but not always, as will become clear) first-year psychology students who are fulfilling a course requirement. They are taken to a small, comfortably furnished room, and there they spend between one and two hours with our research assistant. The session is tape-recorded, with the reader's permission. Each session begins with the experimenter asking the participant questions about reading likes and dislikes for instance: What was the most recent book you read? What did you like or dislike about it? Who was it by? (Danielle Steel and Stephen King are the runaway winners of this informal poll, incidentally.)
Then the participants are asked to read a short story, one page at a time. It is important to emphasize that we do not believe any one measure can adequately capture the complexity of reading, especially literary reading, and therefore we use a whole battery of measures, although not all of them in the same experiment. Some of the measures are in-process (occurring as close to the actual reading as possible), and some of them are after-the-fact, or retrospective.
Our primary in-process measure is a modified think-aloud procedure, related to the procedure used by Linda Flower and John R. Hayes in their studies of the composing process. In our case, however, the participants are prompted and the protocols fairly rigidly structured: at the end of every page the experimenter asks the reader to report anything that was going through her mind as she read that page. Among other in-process measures is the phrase selection task: at the end of every page the readers are shown a list of phrases taken from that page and asked to indicate which ones they noticed particularly.
Similarly, we use a number of different retrospective measures. After the story has been read, the reader is asked general, open-ended questions such as, "What do you make of this story?" We have found that a particularly useful measure is a "probe response" task which consists of the experimenter reading to the student a number of statements about the story. The readers are asked to say whether they agree or disagree with each of these "probes," and then to explain why. Other retrospective tasks include "highlighting" (the readers go back through the story and mark out anything that catches their eye), retelling, and recognition memory tests.
We turn now to three experiments that test some aspects of our model. Since the model specifies that reading is a transaction among readers, texts, and situations, it is convenient to categorize the experiments as explorations of these three different aspects (although in fact the experiments were not performed in this order). In other words, each experiment represents a "push" on one or another of the arrows shown in Figures 1 and 2.
Experiment 1: Readers
The first experiment described focused on readers. We predicted that different types of readers would demonstrate different reading processes, and therefore compared the reading of 12 undergraduates with 12 faculty members. The faculty members were drawn from 12 different disciplines: English and psychology, for obvious reasons, but also history, economics, mathematics, religious studies, and so on.
Each reader was individually asked to read "The Day We Got Our Own Back," a short story by the Irish writer, Maeve Brennan. During reading (i.e., after every page), they were given the think aloud task and the phrase selection task. After reading, they were asked open-ended questions, and given the probe task and the highlighting task.
The qualitative, protocol data from this experiment have not yet been fully analyzed. Nevertheless, at this point there appear to be substantial differences between undergraduates (here called "novice" readers) and faculty members ("expert" readers) on the retrospective measures, especially the probe statements. For instance, in response to the probe, "The story just leaves you hanging. You want to know whether her father ever did get caught, and how it all came out," the novice readers were much more likely to agree (on a 5-point scale, the mean scores were 3.75 and 2.00 for the novice and expert readers, respectively; higher numbers indicate stronger agreement). In all, nine probes were put to each reader. On the average, the novice readers' agreement scores (mean = 2.49) were significantly higher than the experts' (mean = 1.98). In short, the novices tended to answer the probes as if they were, in our terms, "story-driven," whereas the experts tended to respond as if they were "point-driven."
The main in-process measure we have looked at is the phrase selection task, in which the readers are asked to indicate which of a selection of phrases they noticed particularly. We expected that the experts would be more likely than the novices to choose statements that embodied evaluations in the sense discussed previously. However, we found no evidence for this. Thus, it appears at least from this preliminary glance that the main differences between experts and novices lies not in the first read-through of a text, but in later, post-reading processes: how they think about the story, what they make of it. Of course, it is also possible that the post-reading differences we observed could be due simply to the experts' greater "test wiseness"; perhaps they were just better at figuring out what the experimenter wanted to hear. In any case, we plan to pursue this matter further, most immediately, by coding the think-aloud protocols, looking for patterns of response that might distinguish the two groups.
Experiment 2: Texts
Turning now to the second element in our model, Experiment 2 was designed to give a push on the "text" arrow, and in particular to examine the influence of what we call "discourse evaluations."
Earlier we mentioned that evaluations are, roughly, those aspects of a text that embody the narrator's attitudes towards events, characters, or utterances. From the reader's point of view, a textual evaluation is an opportunity or invitation to share an attitude or a belief with the narrator. Generally speaking, evaluations are words, phrases, or events that are unpredictable against the norm of the text. When that unpredictability is due not to what is told but to the way in which it is told, we call it a "discourse evaluation" (cf. Riffaterre's "stylistic devices").
For example, in "The Day We Got Our Own Back," the narrator describes an incident in which some "unfriendly" men, carrying revolvers, make a raid on the family's house. This takes place in Dublin, in 1922; the narrator was five years old at the time. The narrator uses some unpredictable language to describe the men's actions. She says, for instance, "they crowded into our narrow little hall"; "they tramped around the house"; later "they camped around the room." The mother fears that the father, who is "on the run," will "be trapped, and that [the children will] see him trapped."
Now, as we read it, crowded, tramped, camped, and trapped are pretty clear cases of evaluative language. They represent invitations from the narrator to the reader to share a perception of the father as a noble underdog, someone hunted.
It may be objected that this is all quite obvious. And yet it is not obvious that our readers really notice or respond to this kind of material. In our previous work we found that what many of our students really seem to care about is the story what happens. From their point of view these discourse evaluations often have no importance. Perhaps for them it would make no difference whether the text were to read, "they camped around the room" or "they sat around;' because what happens the story event is, however it happens to be expressed, the same.
The question we were asking in this experiment is simple: Are discourse evaluations noticed by the reader? (In fact this was only one of the questions this experiment was designed to answer, but the others are not of immediate interest.) To find out, we had 96 undergraduates read "The Day We Got Our Own Back." Half of them read it in its original, evaluated version, whereas for the other half we prepared an alternate version, identical to the original, except that 21 discourse evaluations were replaced by paraphrases that conveyed the same story event but in a predictable, nonevaluative way. For example, the evaluation, they camped around the room was paraphrased as they sat around the room. Similarly, crowded was paraphrased as came, tramped as walked, and so on.
Students were asked to read this story, and after every page they were shown a list of phrases and asked to indicate which ones "caught their eye" or "struck them" particularly. Half the items on this phrase selection task were ones we considered critical, because in our terms they embodied discourse evaluations (in the case of the paraphrased text, of course, the corresponding paraphrases were "critical" in this sense). 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 were nonevaluated ones (e.g., he pointed out a blue-glass bead).
Our main concern was to find out whether the readers were "sensitive" to evaluative language, where sensitivity was defined in signal detection terms and computed as d'. Setting aside the procedural details of computing d', we found, as expected, that readers of the evaluated version showed greater sensitivity (d' = .030) to the critical items than did the readers of the paraphrased version (d' = - .019). This means, for example, that phrases such as "they camped'" were significantly more distinctive than phrases such as "they sat." In some sense, the two phrases represent the same story event, but there are important differences between them. And, although there are clearly many ways of describing such differences (word frequency, imagery, etc.), we would claim that one important difference is that camped is evaluative and sat is not: camped, that is, affords the reader the opportunity to share a perception of the men with the narrator.
In short, this experiment suggests that evaluations aren't just a figment of our theoretical imaginations. There is apparently something about evaluations a property of the text that is salient for our readers.
Experiment 3: Situations
In another experiment we tried to show that reading is affected by the context or situation in which it occurs. We predicted that different situations would produce different readings of literary texts. In this way we attempted to push the "situation" arrow. Specifically, we had undergraduates read three short stories. After every page of reading they were stopped and asked to report whatever they were thinking about. In addition to thinking aloud at the end of every page, they were given one of three tasks to perform. One group of readers was asked a short-answer question concerning some relatively minor detail that had been presented on that page; for example, "what was the color of the kitchen floor" or "how many boats did the men take?" The task for the second group was to give verbally a short summary of the main events that had taken place on that page. Recall that we distinguish, conceptually, between three modes of reading-information-driven, story-driven, and point-driven. It will be seen that by asking readers to answer detailed, factual questions, we were trying to encourage them to read in an information-driven way. Similarly, by asking them to summarize the plot we were trying to encourage them to read in a story-driven way.
The task used to encourage readers to read in a point-driven way needs a bit more explanation. People who read in a point-driven way impute intentionality to the text: they read in order to see what the narrator might be "getting at." Now, it goes without saying that we can't turn people into point-driven readers at one stroke. We can, however, introduce or "frame" the story in such a way that the readers are more likely to at least try to connect the story to human purposes and goals, rather than reading it as a series of events.
As an example, we will describe the frame used for the third story, which was "Indian Camp" by Ernest Hemingway. (The frames for the other two stories were less elaborate, but comparable). Interestingly, Hemingway himself framed this story. In his collected short stories, some of the Nick Adams stories, which take place in the Upper Peninsula of Michigan, are preceded by short vignettes concerning the battlefields of France during World War I. Our frame was similar to his, only more explicit. We told the readers that during the First World War, a soldier in France had written a letter home to a friend, who was a writer. We then showed the students this "letter," in which the soldier spoke essentially about a loss of idealism. The crucial passage was this:
How did I get to be the sort of person who expects he's going to be heroic and brave, but really turns out to be nothing more than a kitchen corporal who worries because he might have to start a new fire in the morning?By the sheerest coincidence, his friend, the writer, replied that he had "just finished" writing a short story that might have "something to do with the things you've been thinking about." The story was, of course, "Indian Camp," which the students then proceeded to read. And at the end of every page, they were asked if they saw any connection developing between the letter and the story, any reason the writer might have sent it to the soldier.
Thus 70 students read three stories, in one of three situations: the first group was asked detailed questions about it (this was intended to promote information-driven reading); the second group was asked to provide page-by-page plot summaries (this was intended to promote story-driven reading); and the third group was asked to think about the connection between the framing letter and the text (this was intended to promote point-driven reading). The students read three stories, one after the other, and performed the same task for all three stories. Following reading the students were asked openended questions and given the probe response task, and then several days later they completed recognition and free recall tests (all these tasks concerned "Indian Camp" only). Space does not permit us to report all these results here; instead we will focus on just one in process measure, reading time, and will present the results separately for each of the three stories.
The first story was "Old Man at the Bridge," also by Hemingway. As we formatted the text, it occupied three pages; average reading times for each group as a function of page are shown in Figure 3. The dependent measure here is words per minutes, so the faster the reading, the higher the point on the graph. There were two main findings. First, the detail group was slowest overall, and the plot group was the fastest. Second, the plot and detail groups showed little change in speed over time, whereas the frame group (the dotted line in the graph) slowed down noticeably over the last two pages.
Figure 3. Mean reading speed as a function of group and page
for Story 1, "Old Man at the Bridge"
The second story was "The Day We Got Our Own Back," by Maeve Brennan. Average reading times are shown in Figure 4. Here, there were three main findings. First, as before, the detail group was the slowest, and the plot group was the fastest overall. Second, all groups were faster on pages 2 and 3 than they were on page 1. Third, from page 3 to page 4, the frame group (and this time the plot group, too) slowed down, but the detail group did not.
Figure 4. Mean reading speed as a function of group and page
for Story 2, "The Day We Got Our Own Back"
The third story was Hemingway's "Indian Camp." Reading times are shown in Figure 5. Again, there were three main findings. First, the detail group was again the slowest, and the plot group the fastest overall. Second, there was again a speeding effect from page 1 to page 2 for all three groups. Third, over the last three pages, the plot and detail groups were fairly steady, whereas the frame group slowed down significantly.
In brief, the reading speed data from the three stories are remarkably consistent: For all three texts the detail group was significantly slower than the plot group, and the frame group slowed down over the last page or two. What do these findings mean?
We think the large difference in reading speeds between the detail group and the plot group suggests that these groups really were reading the stories differently. Presumably, the detail group was reading for facts, trying hard to acquire and store textual information, whereas the plot group was attending to the storyworld of events and characters a task that apparently is much less demanding. But it was the pattern of results exhibited by the frame group that was most interesting to us. The frame group was the only one to slow down over the last few pages in all three stories. We suppose that this slowing occurred because the frame readers were reading in order to make sense of the stories that is, to see if and how the text was connected to the letter-frame and this task became more urgent as the end of the story drew closer.
In summary, these experiments seem to indicate that when you push on the three different outside arrows-reader, text, situation different things happen inside different kinds of reading processes occur. Thus, in Experiment 1, we pushed on the "reader" arrow, and found some differences between expert and novice readers. In Experiment 2, we pushed on the "text" arrow, and found differences due to evaluative as compared to nonevaluative versions of the same text. In Experiment 3, pushing on the "situation" arrow, we found that different situations can produce different kinds of processing.
When an artillery piece is dropped from a tower, it is a fairly simple matter to determine the outcome of the test: either the gun fires or it does not. After performing our crash tests, however, it is difficult to say whether or not the model still "works." To us, it doesn't appear too horribly mangled, but that may reflect mainly our own prejudices. It's clear, in any case, that in at least one sense the model has survived: what our readers seem to do is, so far, consistent with what the model predicts.
Finally, perhaps the immediate point of doing this sort of work is not to prove a theory "right" or "wrong" but to see whether it helps us learn more about reading. In that sense our testing has definitely been a success. Allowing conceptual models to collide with hard evidence is forcing us to confront some difficult and uncomfortable questions about transactions among readers, texts, and situations.
Brennan, Maeve. In and Out of Never-Never Land: Twenty-two Stories. New York: Scribner's, 1969.
Dyson, Freeman. Weapons and Hope. New York: Harper and Row, 1984.
Flower, Linda S., and John R. Hayes. "The Dynamics of Composing: Making Plans and Juggling Constraints." Cognitive Processes in Writing. Ed. Lee W. Gregg and Erwin R. Steinberg. Hillsdale: Erlbaum, 1980. 31-50.
Gibson, James J. The Ecological Approach to Visual Perception. Boston: Houghton Mifflin, 1979.
Hemingway, Ernest. The Short Stories of Ernest Hemingway. New York: Scribner's, 1953.
Labov, William. Language in the Inner City. Philadelphia: U of Pennsylvania P, 1972.
McCormick-Leighty, Kathleen. "Psychological Realism: A New Epistemology for Reader-Response Criticism," Paper presented at the colloquium, "The New Rhetoric and the New Literary Theory," Carnegie-Mellon University, 18-20 April 1985.
Polanyi, Livia. "So What's the Point?" Semiotica 25 (3/4) (1979): 207-241.
Riffaterre, Michael. "Criteria for Style Analysis." Word 15 (1959): 154-174.
Rosenblatt, Louise M. The Reader, the Text, the Poem: The Transactional Theory of the Literary Work. Carbondale: Southern Illinois UP, 1978.
-----. "Towards a Transactional Theory of Reading." Journal of Reading Behavior 1 (1) (1969): 31-49.
-----. "Transaction Versus Interaction: A Terminological Rescue Operation." Research in the Teaching of English 19 (1985): 96-107.
Verbrugge, Robert R. "Resemblances in Language and Perception," Perceiving, Acting, and Knowing: Toward an Ecological Psychology. Ed. Robert Shaw and John Bransford. Hillsdale: Erlbaum, 1977. 365-389.
Vipond, Douglas, and Russell A. Hunt. "Point-driven Understanding: Pragmatic and Cognitive Dimensions of Literary Reading." Poetics 13 (1984): 261-277.