What makes a good story - Springer Link

45 downloads 63 Views 1MB Size Report
Feb 8, 1984 - protagonist), a good story is said to be composed of "problem-solving ... parse stories into informational nodes showing how a protagonist ...
Journal of Psycholinguistic Research, Vol. 13, No. 6, 1984

W h a t M a k e s a G o o d Story? 1 Allyssa McCabe 2 and Carole Peterson 3 Accepted February 8, 1984

Three different ways have been selected for analyzing a story's structure, each focusing on different types of information. (a) Episodic or story grammar approaches stories as problemsolving episodes, emphasizing goals and activities to achieve them. (b) Labov's high point structure emphasizes affective information and sees stories as organized around emotional high points or crisis events. (c) Deese's dependency analysis emphasizes linguistic complexity and, in particular, the way propositions are related to each other through a relationship of either coordination or subordination. Stories were scored according to how well they realized good structure in each system, and the three scoring systems were relatively independent of each other. Adults were asked to rate 3- to 9-year old children's personal narratives in terms of how good a story each was, and their ratings were compared to how the stories were scored in terms of all three systems. None of the systems completely explained the subjective quality ratings; rather, all three seemed to contribute in different ways to subjects' ratings, with the best narratives generally deemed sophisticated and complex in at least two of the three systems.

We all know a good story when we hear one. But what makes a story good? The answer to this depends upon whom you ask. We will consider three different hypotheses about what makes a story well formed, or good, and then see which of these hypotheses most closely matches the judgments of ordinary readers.

1The authors wish to acknowledge their great debt to James Deese and Christine Glenn. This research was supported by the National Sciences and Engineering Research Council of Canada Grant A-7911 to Carole Peterson. 2Address all correspondence to Allyssa McCabe, Department of Psychology, Southeastern Louisiana University, Hammond, Louisiana 70402. 3Memorial University of Newfoundland. 457 009076905/84/1100-0457503.50/0 9 1984 Plenum PublishingCorporation

458

McCabe and Peterson

Recently, a number of investigators describe stories as composed of episodes (Glenn, 1978; Mandler, 1978; Mandler & Johnson, 1977; Rumelhart, 1975, 1977; Schank & Abelson, 1977; Stein & Glenn, 1979; Thorndyke, 1977). After providing a setting (such as introducing the protagonist), a good story is said to be composed of "problem-solving episodes" (Rumelhart, 1977). Basically, something happens to protagonists, which causes them to respond to it or to set up a goal. Their actions or their attempts to accomplish the goal result in some resolution or state of affairs that terminates the episode. Episodes may follow one another sequentially or may be embedded within other episodes. Investigators differ in the details of their "story grammars," but all of them parse stories into informational nodes showing how a protagonist solves a problem; these nodes include such information as initiating events, goals, attempts to accomplish these goals, and consequences. Glenn and Stein (1980) have identified a number of story structures, some of which are diminutive and some more complex than the episode, although they are all related to the episode in some way. We have adopted all their structures in the following analysis. (For specific details of this and the other two analyses as we applied them, see Peterson and McCabe, 1983.) Labov and his colleagues (Labov, 1972; Labov, Cohen, Robins, & Lewis, 1968; Labov & Waletsky, 1967) have a rather different notion of what makes a story good. They see narratives as being structured around "high points" or "suspension points." They suggest that a good story begins by providing the listener with orientation about whom or what the experience to be described involved and when and where it took place. A good story then builds up to a high point through the recapitulation of events and then often suspends the action at this crisis point while its importance is highlighted. Then the resolution of this crisis is provided through more recapitulation of events. Good narratives are organized around one or more key points that are stressed by the narrator. Two functions in narratives are emphasized: reference--the listeners (or readers) are told what happened, and evaluation--the speakers (or writers) reveal their attitude toward the events of the narrative. The third analysis to be discussed in this work--dependency analysis--does not deal with stories per se but with discourse in general, and was developed by Deese (1983). In a number of ways, dependency analysis asks how coherent is any given discourse. Like high point and episodic analyses, dependency analysis views discourse as a tissue of propositions. But dependency analysis depicts the surface syntactic relations among the propositions rather than the nature of their content, which is the practice of the other two analyses. Whether a given proposition is

Good Stories

459

coordinate or subordinate to any other given proposition is the principal question asked by dependency analysis. Having undergone dependency analysis, propositions are displayed as a hierarchy, revealing the (hierarchical) syntactic complexity of a given discourse. Like any other discourse, then, a story is considered good to the extent that its propositions form a complex, ideal hierarchy. Each of these three analyses posits one structure as ideal the complete episode, the classic high point narrative, the ideal hierarchy. One might ask whether one of these analyses is superior to the rest in terms of describing what it is that makes a story seem good to most listeners or readers. So far, only one of the three approaches--the story grammar, or episodic, analysis--has been evaluated to any extent, and a number of different methods have been used to do so. Black and Wilensky (1979) take issue with formal considerations of the story grammar approach. They find only one version of story grammar to be formally adequate--namely, the forthcoming Johnson and Mandler revision. They also find that story grammars fail to generate many stories (including those that reverse the order of attempt and outcome) and do generate such nonstories as procedural expositions that instruct readers on how to do some task. In other words, the story grammar approach does not account for some well-formed stories, stories that would also be perceived as good by naive readers, while it does account for stories that would be perceived as bad by naive readers. A second approach has been to see whether the episodic structure is specific to stories from Western culture or whether it is a linguistic universal. In other words, do all cultures agree about what makes a story good? The evidence conflicts. Mandler, Scribner. Cole, and DeForest (1980) showed that when given episodic stories from both Western and native cultures, nonschooled Liberian children and adults (many illiterate) recalled those stories in a fashion very similar to schooled American children and adults (all literate). Those authors explicitly assume that the organization of simple stories is a cultural universal. In contrast, Kintsch and Greene (1978) make the opposite assumption--namely, that story schemata are culture-specific. They selected an American Indian story that was not episodic in structure and an episodic Grimm fairy tale. They found that American college students were capable of recalling significantly more propositions from and much more of the gist of the fairy tale compared to the Indian story. This supports the idea that we remember better the stories that fit our culture-specific schemata. Different cultures may very well have different ideas of what makes a story good, insofar as we tend to equate the memorability with the quality of a story.

460

McCabe and Peterson

Pollard-Gott, McCloskey, and Todres (1979) took a third approach. They asked people to sort propositions of various texts together. They submitted these sorts to hierarchical clustering analyses and compared the resulting structures to story schemata. They found that while some people grouped propositions in the episodic fashion proposed by Stein and Glenn (1979), Mandler and Johnson (1977), and others, there were other people who grouped analogous elements across the various episodes in a single story. In other words, even within our own culture, there appear to be individual differences in the perceived structure of stories, making it likely that there would be individual differences in judging what stories are good ones. But there are some who would argue that rather than prove any one of the approaches superior and others deficient, we should make use of all of them. For example, Jenkins (1974) argues that any notion of one wholly sufficient analysis of language or cognition is bound to be misguided since such "analyses mean something only in terms of their utilities for some purposes." He articulates this notion as a cornerstone of a more general approach to psychology called contextualism. In fact, the three approaches we have selected are only three among many others (e.g., see van Dijk & Petoefi, 1977). The question remains: What do ordinary readers think makes a story good? We propose to look at what stories various groups of people rate as good relative to others they rate as bad, and to see how these ratings reflect structural sophistication as conceptualized by each of the three analyses mentioned at the outset of this article. Although '~story" usually refers to a fictionalized account of something, we propose to use a set of children's narratives about their real experiences as materials for this study. Narratives can be described by the same analyses as stories, as we shall see, and therefore differ from stories primarily in the truth of their content. The narratives to be rated will be a set of spontaneously produced narratives of varying quality told by children aged 4 to 9 years. Children's narratives will be used because adults' narratives are too long to plausibly study in large numbers and too complex to be adequately categorized by the three systems. In the interests of generalizabi!ity, no attempt will be made to polish the narratives. We want to know what kind of narratives people consider to be the best among those they encounter in daily conversation.

Good Stories

461

METHOD Subjects Thirty-five people served in this experiment. Fourteen were undergraduate students of English, 14 were undergraduate students of psychology, and 7 were faculty members of the psychology department of Memorial University of Newfoundland. Approximately half of each group were women and half were men, for a total of 20 women and 15 men. All were Canadian.

Materials The stories used in this experiment were 288 transcriptions of oral narratives of personal experience produced by 96 children who were white, predominantly working-class, aged 3V2 to 91A years, chosen from a nursery school and an elementary school in a small town in Ohio. There were 16 children, half boys and half girls, in each of the six yearly age ranges, with mean ages of 4 years 1 month, 5 years 1 month, and 6, 7, 8, and 9 years 0 months. Children produced these stories in a conversation with an experimenter, where the experimenter produced a number of standard prompting narratives. These prompts did not contain evaluation (e.g., "That was scary"), as that is the only easily extractable, generalizable component of a narrative and we did not want the children to simply parrot us. The prompting narratives were about experiences that are unusual for any one child but that were likely to have happened to all children (e.g., trips, going to the doctor's, pets' visits to the veterinarian). Children were given an art project to complete during the course of the interview to minimize self-consciousness. For details of this procedure, see Peterson and McCabe (1983).

Scoring The three longest narratives produced by each child were submitted to each of the three analyses mentioned in the introduction. These arc

462

McCabe and Peterson

summarized below, but details of each scoring procedure can be found in Peterson and McCabe (1983). Episodic Analysis. Narratives were scored as containing one or more of the following structures: 1. Descriptive sequence: This is simply a description of the character and his or her surroundings and habitual actions; the statements in a descriptive sequence are not causally related. 2. Action sequence: The focus is on behavior, with a series of causally unrelated actions as well as external and internal states of the characters involved. 3. Reactive sequence: The focus is on changes in the narrative environment, where, typically, one change generally causes other changes. Something happens that causes something else to happen, although there is no evidence of goals. 4. Abbreviated episode: This describes aims of the protagonist, but planning generally must be inferred. Two components are required: (1) Some motive for action, either an event in the environment or an internal motivating state, must exist and lead to (2) a specified consequence that achieves or fails to achieve the protagonist's goal, which may or may not be specified. 5. Complete episode: This describes purposive behavior, but there is more evidence of planning. A complete episode must include at least three of the categories of events, motivating states, attempts, and consequence, with the consequence category obligatory. If consequences are omitted, the episode is incomplete. In addition, complete episodes may include the optional categories of settings, reactions, and judgments. 6. Complex episode: This involves complications of the basic complete episode. Some complex episodes involve reactive sequences or complete episodes functioning as single units in higher-order structures. Other complex episodes involve some kind of complication in the pursuit of the goal. 7. Interactive episode: There are two people who have goals and influence each other. Such a narrative may be analyzed from the perspective of either person, and one can specify a complete episode for each. Individual statements usually serve multiple functions, the functions differing depending upon the protagonist focused upon. 8. Multiple structure narratives: There is more than one structure in the story; the different structures may be of different types (e.g., a reactive sequence and a complex episode) or of the same type (e.g., two or more complete episodes).

Good Stories

463

The sophisticated structures in this analysis are complete, complex, and interactive episodes. Also, according to Glenn and Stein (1980), descriptive, action, and reactive sequences are considered diminutions of the complete episode. High Point Analysis. Each narrative was scored as one of the following structures: 1. Disoriented: In these narratives, the child is either confused or disoriented about the events being narrated or misuses language (e.g., misuses negatives) such that the narrative cannot be understood. 2. Impoverished: These narratives either consist of so few sentences that they form no recognizable or analyzable pattern, or they provide two successive events (the minimal definition of a narrative used both by us and by Labov) and then go over and over them, often providing extensive orientation and evaluation about these two events. 3. Chronological: These narratives contain only a-then-b event clauses and are merelY a description of successive events, temporally rather than structurally integrated. The narrative is not built around a high point. 4. Leapfrogging pattern: The child jumps from one event to another but clearly leaves out various major events, such that it is difficult to reconstruct the original events being described by the narrative, yet one can often infer that some one event would have served as a high point had the child been a skillful enough narrator to depict it as such. 5. Ending-at-the-high-point pattern: The narrator provides successive complicating actions until a high point is reached. The high point is dwelled on in a manner similar to that in the classic pattern (described below), but the narrative is promptly terminated at this point; there is no resolution of the high point event. 6. Classic pattern: The events leading up to a high point, or crisis, are recapitulated in a well-ordered series. At the high point the action is suspended and the high point evaluatively dwelled on, and then the events that resolved the high point are successively related. There are some variants of this pattern, which are described in Peterson and McCabe (1983). From the point of view of high point analysis, the classic narrative is the most sophisticated structure--in fact, the only sophisticated structure. Somewhat less sophisticated is the ending:at-the-high-point structure. Leapfrogging, chronological, impoverished, and disoriented narratives are considered to be structurally unsound. Dependency AnalYSiS. The dependency analysis devised by Deese

464

McCabe and Peterson

(1983) breaks any discourse into contrastive functional syntactic propositions and displays the dependencies of those propositions in a form much like an outline. Such an outline may reveal that a discourse expands in one of two ways: by proliferation (multiplying the number of propositions, but without much dependency of subsequent propositions on former ones) or by elaboration (multiplying the depth or complexity of the discourse by having propositions depending on other propositions, which in turn are dependent upon still other propositions). After such dependency analysis, each narrative was scored as one of the following structures (borrowing rhetorical labels from Christiensen & Christiensen, 1976): 1. Simple coordinate sequence: Propositions simply proliferate, with few dependencies of one proposition upon another, as, for example, in a narrative that consists simply of a sequence of events that are not elaborated upon. 2. Simple subordinate sequence: Expansion is only in the form of elaboration here; a series of (usually right-branching) contrastive propositions are successively dependent upon each other, as, for example, in an elaborate description of something. 3. Combination of a simple coordinate with a simple subordinate sequence: A series of unelaborated propositions is combined with a solitary series of propositions that successively elaborate some aspect of discussion. 4. Mixed coordinate sequence: A fair number of propositions consistently show moderate elaboration by means of dependent propositions. 5. Mixed subordinate sequence: A mixed coordinate sequence is combined with a spate of successively dependent propositions elaborating some aspect of discussion. 6. Ideal hierarchy: Propositions display elaborate proliferation of dependencies; there may also be more than one shoot of elaboration to unusual depths. Such a structure is quite a syntactically complex discourse. In this analysis, the structures are progressively more sophisticated, with the ideal hierarchy considered to be the most sophisticated structure. Reliability. Christine Glenn trained the second author in episodic analysis, and the former scored 25% of the narratives for reliability calculations. All narratives were scored episodically by the second author, and the agreement with Glenn was 84%. All narratives were classified in terms of high point structure by both authors, and agreement on that classification was 87%. Disagreements were resolved by consensus.

Good Stories

465

All narratives were scored in terms of dependency analysis by the first author, who was trained by James Deese. As dependency analysis is mechanical, the scoring was simply checked by Deese. He checked approximately 12% of the narratives completely and was consulted whenever any ambiguity in applying the rules of dependency analysis occurred.

Procedure The 28 undergraduates received half of the 288 narratives, while the 7 faculty members received all 288 narratives. Narratives were given in a random order, and subjects were not informed, nor did they see any indication, of the age or sex of any narrator. However, they did know that they were reading narratives produced by children. The subjects read through all the narratives they received to get some idea of the range of things they were to rate. After reading all of them, subjects rated each story on a 6-point scale, where 1 -- a very bad narrative and 6 = a very good narrative. Subjects were asked to use all points on the scale. Each narrative received 21 ratings, which were averaged.

RESULTS AND DISCUSSION

Table I shows the frequency of narratives classified according to their average rating of quality and according to whether or not they contained a particular type of episodic structure. Because of the problem of insufficient expected frequencies, rating categories had to be collapsed into bad (1-3), mediocre ( 3 4 ) , and good (4-6) in order to perform a chi-square analysis. Individual chi-square calculations were performed to test for the possible association between quality ratings and the presence or absence of each type of episodic structure because narratives could (and often did) contain more than one kind of structure. Ratings of quality are associated with the presence or absence of descriptive and action sequences (Chi square = 33.54, df = 2, p < .001; Phi coefficient = .34). More narratives containing such sequences are rated poor and fewer are rated mediocre or good than would be expected due to chance. Rated quality is also associated with the presence or absence of complete episodes (Chi square = 41.8, df = 2, p < .001; Phi coefficient = .38),

466

McCabe and Peterson

Table I. Relationship Between Mean Rating of Quality and Episodic Structure "'b i

Mean rating of quality

I

Descriptive and action sequences

Complex and interactive Multiple episodes structures

Reactive sequence

Abbreviated episode

Complete episode

Bad 1-3

31 (14.4)

33 (34.3)

5 (7.8)

13 (33.4)

7 (13.8)

13 (30.3)

Mediocre 34

15 (23.1)

54 (55.0)

14 (12.5)

55 (53.6)

23 (22.2)

40 (48.5)

Good 4-6

4 (12.5)

32 (29.8)

8 (6.8)

48 (29.0)

18 (12.0)

52 (26.3)

~Note that six separate chi-squares were calculated, one for presence or absence of each type of structure. bObserved frequencies are printed above (expected frequencies).

with the presence or absence of complex or interactive episodes (Chi square = 7.69, df = 2, p < .05; Phi coefficient = . 16), and with the presence or absence of multiple structures in a narrative (Chi square = 57.59, df = 2, p < .001; Phi coefficient = .45). More narratives containing such sequences are rated good, fewer bad, than would be expected due to chance. There is no association between rated quality and the presence or absence either of reactive sequences (Chi square = .40, df = 2, n.s.; Phi coefficient = .04) or of abbreviated episodes (Chi square = 1.56, df = 2, n.s., Phi coefficient -- .07). Apparently, people are sensitive to episodic structure when rating the quality of a narrative. Table II shows the frequency of narratives classified according to their average rating of quality and according to what type of high point structure they exhibited. Again, rating categories had to be collapsed into three for a chi-square analysis. Ratings of quality are associated with high point structure (Chi square = 63.01, df = 6, p