Journal of Pragmatics 6 (1982) 383-422. Smartly polished for upload, August 2005.

THE STORY OF GRAMMARS AND THE GRAMMAR OF STORIES

Robert de Beaugrande

[An earlier version of this paper was delivered at an invited symposium on discourse processing at the American Educational Research Association convention, Los Angeles, 1981. I am indebted to Dr. Jim Coots of the Southwest Regional Laboratory in Los Alamitos for organizing the symposium; to Robert Calfee, Charles Fillmore, Jean Mandler, and Robert Wilensky for appearing on the panel; and to Benjamin Colby, Jean Mandler, James Mehan, David Rumelhart, and Nancy Stein for further comments.]

Abstract

The theoretical and practical development of story grammars is analysed through a detailed comparison of some influential proposals. Progress in research suggests a gradual trend away from strict formal grammars toward psychological process models. However, several issues remain unresolved, such as the conceptions of ‘grammatical rule’ and ‘well-formed story’, and the requirements of interest and surprise in the interaction between storyteller and audience. Some prospects are aired for broader research.

1. The heritage of formal grammars

1.1. When a new notion is introduced into a field, a primary goal of research is to render it plausible and well-known. After a period of adjustment, we ought to take stock, and probe whether and how far investigators are proceeding in consistent and productive ways. If the notion has been pursued in different directions by its various adherents, the accruing body of research should present some signs of tension, diversity, and controversy.

1.2. Such is my assessment after a close reading of the literature on ‘story grammars’ over some five years, which promised significant contributions to the study of discourse processing, and a rich domain of exploration for psychologists, linguists, and educators alike. Yet in view of lively controversy, I think it may be useful to trace the historical and theoretical evolution of story grammars lest the main issues of contention be clouded over by disparities among claims and terms.

1.3. During the acceptance of formal grammars into linguistics and psycholinguistics, the general assumption prevailed that there was no significant level of formal organization higher than sentence structure. The sentence remained the chief object of inquiry and sustained widespread views about how grammars are or should be constructed (see now Beaugrande 1999). Thus, early attempts to design ‘text grammars’ favoured some type of analysis was essentially fashioned for sentences, with fairly minor modifications (cf. van Dijk 1972; van Dijk et al. 1972). Indeed, it was not clear at the time whether these modifications were sufficiently substantive to warrant setting up ‘text grammars’ as a domain independent of current or future sentence grammars (cf. Dascal and Margalit 1974; Ballmer 1975).

1.4. In this ambience, the proposals for a story grammar implied some discourse structure on a higher level than the sentence yet still analogous to it (cf. Rumelhart 1975: 211f). The question was how far the analogy would be either implicit or explicit, notably in regard to was the extensive borrowing of terminology (e.g. Mandler and Johnson 1977; Johnson and Mandler 1980). Recent controversies reflect the backlog of issues where stories do not seem so readily comparable to sentences. Moreover, the consensus has been gradually emerging that some seemingly secure categories of sentence grammars themselves are not so straightforward after all. We might therefore start off the present discussion with a brief look at the heritage of formal grammars in the study of language and communication.

1.5. The central enterprise of modern linguistics, as derived from the methods proposed by Saussure (1916) and his admittedly diverse followers, has been the description of structures. These methods were not offered up as an account of the structure-building operations of language in use. A clear demarcation was drawn between the abstract language system (‘langue’) and the uses of the system (‘parole’). The categories and procedures developed were accordingly simple analytical techniques employed by the linguist as investigator. The major effort was directed toward isolating units and toward classifying those units in a taxonomy, so as to describe structures as relationships obtaining between two or more units within a system. Consequently, the approach was often called ‘structuralism’ (surveys in Bierwisch 1966; Helbig 1974; Beaugrande 1984).

1.6. Structuralism was adopted by some early studies of the organization of prose (e.g., Propp 1928; Sklovskij 1929). This research undertook to isolate units that figured as structural components of plot lines in many samples. So Propp (1928) justly called his method not a ‘grammar’ but a ‘morphology’, i.e., a catalogue of meaningful forms. He only isolated the units which his set of folktales had in common, and did not offer much detail about the creation of the individual tale, nor about the role of the story-teller (cf. 3.17). (compare Lévi-Strauss 1960). He described ‘functions’ in terms of blocks of content, e.g. ‘villainy, departure, struggle, victory, and return’. Some of these units are by definition structural, i.e., determined by relations among units. For example, a departure would logically be stated before a return, and a struggle before a victory. Others are not so confined, e.g., ‘villainy’ could occur anywhere in a story, though as Benjamín Colby (personal communisation) points out, it most frequently appears near the beginning of a story to provide a clear motivation for the hero’s response (cf. Colby 1973; Sutton-Smith et al. 1976).

1.7. In retrospect, the limitations of structural description seem more evident than they did for quite some time. These limitations are part of the historical heritage of grammar-based approaches, and thus far bear on the current controversy over stories. Some researchers even proposed that structures could and should be isolated and classified without taking account of content (e.g., Harris 1951). Also, the analysis of structures was seen as an end in itself, divorced from the study of how structures get created or used. Taken together, these two tenets leave open the questions of what status the structures have, and how they may relate to the concerns of human communication.

1.8. These questions can be illustrated with the ‘discourse analysis’ proposed by Harris (1952) can illustrate these questions. Looking only at the distribution of formal units such as nouns and verbs, he suggested that discourse can be defined in terms of the ‘equivalences’ obtaining among its sentence patterns. But he noticed that his samples, e.g., an advert printed on a hair-tonic bottle, did not manifest very many such equivalences. He therefore introduced into linguistics the seminal idea of the transformation — not as a language processing mechanism, but merely as a technique for formal analysis. To get a match for (1), he ‘transformed’ (2) into (3) — a commonplace operation in subsequent ‘formal linguistics’):

(1) You will be satisfied.

(2) Satisfied customers

(3) Customers are satisfied.

This method is an unintentional but striking demonstration of a major problem permeating all varieties of structuralism: the investigator is creating the structures that are to be analysed and described. Martinet (1962: 58) expressed the matter pointedly, though few seem willing to grasp it then or now: ‘A structuralist is not one who discovers structures, but one who makes them’. When structuralists like Harris purported to ‘abstract away from’ the use, meaning, and purpose of language in communication, their own investigatory procedures were themselves an unacknowledged and rather untypical use of language. The formal analysis of sentences rest upon the linguists’ knowledge of how sentences can be used. I have concluded that the discourse of linguists merits a detailed analysis in its own right (Beaugrande 1991).

1.9. The prerogative to create structures entrained linguistics in a strenuous enrichment of the ‘well-structuredness’ of language by postulating theoretical substrata of structure, as when Chomsky (1957) took over Harris’s idea of the ‘transformation’ for a different purpose. Here, structural relationships among sentences were construed as defining not the discourse, but the entire language. One would set up a ‘grammar’ whose rules would both describe all the basic strings and transform those basic strings into any more complex structures (or back again). The grammar was thus a theory for establishing the well-structuredness of every ‘grammatical’ or ‘well-formed’ sentence. Predictably, considerations of language use were again excluded, but the exclusion of meaning was soon abandoned in favour of assigning it an ‘interpretive component’ (Beaugrande 1998).

1.10. Even in his early review of Skinner, Chomsky (1959) sounded ambivalent. On the one hand, he envisioned a ‘grammar’ as ‘a statement of the integrative processes and generalized patterns imposed upon the specific acts that constitute an utterance’; and ‘rules of the grammar’ as the ‘selective mechanisms involved in the production of a particular utterance’ On the other hand, he warned that ‘the construction of a grammar which enumerates sentences in such a way that a meaningful structural description can be determined for each sentence does not in itself provide an account of this behavior’ (1959: 56). This ambivalence has run throughout the later history of formal grammars: on one side, they are a formalism that directly represents human language; on the other side, they are a purely analytic technique for linguists who manipulate and classify data. Most researchers would favor the former enterprise; but the methods themselves obdurately tended toward the latter.

1.11. Consequently, each central notion of transformational grammar was routinely used in diverging senses. First, there was the formal definition stipulated during the construction of the grammar; second, there was a commonsense definition more attuned to human activities. For example, ‘generate’ means, in the formal grammar, ‘assign a structural description to’; in normal usage, it means ‘create’, ‘produce’. The ‘speaker-hearer’ was a formal device, a kind of abstract automaton with ‘perfect’ knowledge of ‘well-formed sentences’; in everyday parlance, it is a real human being using a language to communicate. Intriguingly, Chomsky (1965: 9) acknowledged that’ his ‘terminology’ had engendered ‘a continuing misunderstanding’, yet he could ‘see no cause for a revision’. Presumably, the clarification of the misunderstanding could have engendered pressure to reconstitute the whole enterprise.

1.12. I have briefly aired these difficulties to illustrate the principled divergence between formal grammars and human faculties. The grammar is set down via deduction (a priori stipulation) and centers upon formal definitions. The study of human faculties, in contrast, is set down by induction (conclusions drawn from experience) and centers upon empirical demonstrations. It seems plain that deduction and definition should be the second stage of research, not the first. Before we formalize a domain, we need empirical evidence about its nature. Reacting no doubt against the behaviorist emphasis on ‘observation’, Chomsky (1965: 19f., 15) dismissed ‘objective tests’ as a ‘matter of small importance’, unfit to bring any ‘gain in insight and understanding’; indeed, the discovery of ‘patterns’ in ‘observed speech’ was said to ‘preclude the development of a theory’.

1.13. It is, I think, among the most important tenets shared by researchers in cognitive science that there should be no gap between a formal representation and its empirical domain. For example, Rumelhart advocates ‘a grammar of the language’ that ‘fits together with the rules for processing the language’ (1977a: 12). If this kind of a grammar is to work, we will need closer integration between the formal-deductive approach and the empirical-inductive one. The two issues raised in 1.7 must be placed in the center of our deliberations. First, we must consider whether a grammar can and should imply or assert the independence of structure from content. To have a grammar at all, one must have both a set of categories for classifying elements, and a set of rules for arranging elements (Jean Mandler, personal communication). The question is then whether these categories and rules can be defined in terms of content, and if so, how far such a definition can be compatible with a definition in terms of structure. Second, we must consider whether formal rules are justifiable only if they claim to represent the structure-building operations carried out by human processors. If so, the design of grammars must be constantly updated as experimental evidence accrues. In the next section, I undertake to probe these two issues within the recent evolution of the story grammars approach.

2. The heritage of story grammars

2.1. Against the background sketched above, the current debate over story grammars might well have been foreseen to trouble the critics of story grammars (e.g., Black and Wilensky 1979; Black and Bower 1980): the separation of structure from content, and the empirical status of the grammars. Though I would concur that grammars with the limitations cited by the critics are problematic, I am not convinced that those deficiencies can properly be laid to the account of the story grammars under discussion.

2.2. Inspired in part by transformational sentence grammars, recent grammars went beyond the isolation of units typical of the early studies (cited in 1.6) and offered sequencing rules (e.g., Colby 1973: 645). Rumelhart’s (1975 [original 1973]) paper, following Colby’s approach, was among the most influential for the new trend in story grammars (e.g., acknowledged in Bower 1976: Mandler and Johnson 1977: 113; Thorndyke 1977: 19). He called his approach a ‘schema’ in the paper’s title, but used the term ‘grammar’ throughout the paper itself. Rumelhart (1975: 213) followed Chomskyan grammar in giving the generative (productive) role to syntax (structure) and the interpretive role to semantics (content). He didn’t seem to consider the trenchant critiques that had already been advanced in linguistics against this assignment (e.g. McCawley 1968; Lakoff 1971; vgl. Petöfi 1971). In psychology, Mandler and Johnson (1977: I 13f.) rejected it as ‘unwieldy to work and frequently redundant’. Thorndyke (1977: 78) introduced semantics into the syntactic domain by appealing to Fillmore’s (1968) ‘case grammar’, where structural categories are additionally defined by recourse to their content, e.g., the sentence slots reserved for expressions of location and time. And finally, Rumelhart himself later remodeled his own set-up by defining the schema in terms of both structure and content:

A schema is an abstract representation of a generic concept for an object, event, or situation. Internally, a schema consists of a network of interrelationships among the major constituents of the situation represented by the schema. (1977b: 267)

2.3. Black and Wilensky (1979: 214) single out Thorndyke’s ‘grammar’ as having only syntactic relations (cf. Black and Bower 1980: 228). It is true that Thorndyke (1977: 77) talks about ‘inherent plot structure’ that is ‘independent of passage content’. But his rules signal the contrary, e.g. (1977: 79):

STORY => SETTING + THEME + PLOT + RESOLUTION

SETTING => CHARACTERS + LOCATION + TIME

Categories like ‘setting’ ‘location’, and ‘time’ are semantically determined, not syntactically. Thus, these structural rules can be applied only with constant recourse to passage content; at most, they are independent of the specific details of passage content (e.g., where the location is, and what the time is.) Moreover, these rules are not fully syntactic because they do not all have to do with sequencing. Thorndyke (1977: 80) follows Rumelhart’s (1977: 213) use of the ‘+’ sign ‘to form two items in a sequence (cf. 2.7.5.1.). But it is obviously not the case that a story must give the location after the characters and before the time. The beginning of a standard folktale may have a quite different order, as in the opening of Snow White (Grimms’ version):

(4) Once it was the middle of winter [time], and the snowflakes fell from the sky like feathers. At a window with a frame of ebony [location] sat a queen [character] and sewed.

I shall return to this matter of order in the discussion of ‘rewrite rules’ (cf. 2.7.5.1, 2.14).

2.4. Conversely, Black and Wilensky (1979: 227) would each view their own story research, which coincides mainly just in their reservations about story grammars, as based entirely on content and ‘unguided by syntactic considerations’. However, their key notion (and that of many story grammars too, e.g. Rumelhart 1977b) is problem-solving, which is in fact a structural category just as much as it is a content-based one. A problem is at least a pair of states, an initial state and a goal state — possibly with intermediate states between them — where there is at least a reasonable probability that the attainment of the goal state will fail. The higher that probability, the more serious the problem, and hence the more interesting (Beaugrande 1980b: 27). Often, the problem-solver will have to set up subgoals because the main goal cannot be reached in a single push (cf. Meehan 1976). Depending on the method, processing may focus on a single complete path (depth-first search); or on the spread of alternatives for each phase (breadth-first search); or on the differences between the initial state and the goal (means-end analysis). From all this, it follows that we identify or understand a problem by relating content to a structural configuration: a network of states and their transitions.

2.5. It is therefore inappropriate to adopt an either/or stance on structure versus content, placing the story grammars on one side of the fence and the problem-solving story schemas on the other. I think we can safely concur with Mandler and Johnson (1980: 311):

What people within the story-grammar framework claim [...] is that in addition to content specifications, stories have specifiable structures and that people have knowledge about such structures which they use in the course of comprehension and retrieval.

This ‘claim’ may have been unclear in earlier work, where the analogy of story grammars to sentence grammars could still taken for granted without explicit critical evaluation. In formal linguistics, the sentence is a purely structural category, defined exclusively by reference to its format; but in actual communication, meaning and purpose are always implicated in the processing of utterances (which may or may not have precisely determinable sentence formats). The story, on the other hand, cannot be dually defined in this way. When Thorndyke (1977: 78) presents ‘the rules of the narrative syntax’ as ‘independent of the linguistic content of the story’, the distinction should be not between structure versus. content, but between higher-level versus lower-level categories in a hierarchy. The ‘linguistic content’ is then the detailed utterances of the story itself, as opposed to the organization shared by stories in general.

2.6. Johnson and Mandler (1980: 54ff.) soon qualified their analogy, expressly noting that ‘the semantic and syntactic aspects’ are more ‘intertwined’ in stories than in sentences in that ‘story categories are more dependent on context for their assignment than are lexical or sentential categories’; yet remarking that this same intertwining may be far more important in sentences than the ‘standard theory’ suggests (1980: 57). The claim that ‘the syntactic classes’ of sentence ‘units can be determined without analyzing the structure of the sentence in which they appear’ (1980: 56) is far too strong (especially for English); but, as I shall argue in 2.14, the dependency upon context is indeed different for story categories versus sentence categories.­

2.7. We might now inquire how the differences between sentences and stories might be reflected in all terms and notions of story grammars. For purposes of discussion, shall enumerate six perspectives regarding the status of grammatical rules..

2.7.1. The rules are stipulated solely by the formal construction of the grammar. This perspective pervades most work on the strictly ‘formal grammars’ upon which such linguistic conceptions as Chomsky’s are based. Such rules would have to specify the formal structure of all ‘grammatical’ stories and exclude all non-stories; and this delineation would have to be done by a purely mechanical application of the rules for well-formedness. A serious weakness of linguistic grammars is that human language users are demon­strably not able to determine in any mechanical way which sentences their language allows (cf. Greenbaum ed. 1977). The same empirical intracta­bility will almost certainly be found for stories. At most, people will be more or less confident about accepting a given story, or will rate samples as better or worse stories (Brewer and Lichtenstein 1981; Stein and Policastro 1982); on this basis, we might be able to assign a probability for storyhood to any sample. But in doing so, we exchange the purely deductive, formal outlook on rules in favor of an inductive, approximative outlook (cf. 1.12; 3.15).

2.7.2. The rules are analytic techniques to be applied by the investigator. In 1.8ff., I remarked that most of the early rules in grammar had precisely this status. For example, in Harris’s (1951) ‘discourse analysis’, transformations were carried out only by the linguist, and only to create equivalent syntactical patterns. An enduring perplexities with Chomsky’s followers is that they were not content with this version of transformations and tried to upgrade them into psychological processes carried out by language users (cf. Clark and Clark. 1977). Perhaps it is not impossible that real language processes do resemble the analysis carried out by professional linguists, and much psycholinguistic research has in fact assumed that resemblance (cf. Clark and Clark 1977). But for that very reason, the resemblance tended to be built into experimental designs rather than treated as a hypothesis to be tested.

2.7.2.1. In story grammars, the rules have been frequent tools for the investigator’s analysis. However, the detailed procedures for the analysis are much less specific than in sentence grammars, because we can agree much more easily on entities like ‘noun phrase’ than on entities like ‘theme’. Thus, although the value of rules as heuristics for story analysis seems secure, I doubt we can accept this as their only function. Surely, rules that reflect human processing should be considered more interesting than rules for analysing examples.

2.7.3. The rules represent human operations carried out when producing or comprehending a story. Here, the rationale for the rules is neither formal deduction nor heuristic analysis. Immediately, the constraints upon the creation of rules, as well as the constraints embodied in the rules, are of a wholly different nature. The rules must not only stipulate what elements are placed in what sequences, but also what limits there are. A purely formal rule-set might, as Chomsky (1965: 10) argued, be stipulated without regard for ‘rapidity, correctness, and uniformity of recall and recognition’; but a rule-set for human operations must be bounded by considerations of memory, attention, motor control, and so forth. It is no coincidence that story grammars have been most impressively supported by evidence of people’s story memory (e.g., Mandler and Johnson 1977; Thorndyke 1977; Mandler 1978; Stein and Nezworski 1978; Stein and Glenn 1979). Obviously, this shift brings important restrictions on such notions as recursion and embedding (to which I return in 2.17ff.). In effect, the ‘infinite’ set of strings which formal grammars could describe (e.g., Chomsky 1957) becomes the finite set of story structures which people will be likely to produce and understand, given the limitations upon processing capacities (cf. Beaugrande 1984a, 1984b, 1987).

2.7.3.1. The most alluring aspect of this perspective is that it is empirically testable in a manner that the perspectives depicted in 2.7.1 and 2.7.2 are not. As Johnson and Mandler (1980: 78f.) suggest,

Each model has been used to predict that departures from canonical form will make comprehension more difficult and that given a less than well-formed story to remember, subjects will tend to recall the story in a more canonical form than that which was presented. For the most part, tests of these predictions have involved generating a story which a given grammar predicts is well-formed, generating an alternative version of the ‘same’ story that violates the rules in some way (e.g., by deleting constituents or moving constituents out of order), and comparing performance for the two forms.

The loyal following that story grammars have attracted may be due to the success in such experiments (cf. Kintsch et al. 1977; Thorndyke 1977; Mandler 1978; Stein and Nezworski 1978; Stein and Glenn 1979). Johnson and Mandler (1980: 79) are not entirely happy with these confirmations, because the latter ‘involve assessments of relative, rather than absolute, grammaticality’ (emphasis added). This discontent hardly seems just. Even Chomsky (1965: 11) was ready to admit that ‘grammaticalness is, no doubt, a matter of degree’.

2.7.3.2. Consider in this connection the notion of the canonical form invoked in the quote from Johnson and Mandler. In formal grammars, the canonical form is plainly an idealization, independent of any realization. This should hold especially for stories (cf. Smith 1980: 216ff). We can refer to different versions of Snow White, but we can’t single out any one of them as the canonical version; at most, we could accept the oldest version, as was the tendency in nineteenth-century folklore research (but compare the protest in Housman 1911); or build a fictional version from the elements common to all known versions (Lévi-Strauss 1955). In any case, we are depending upon our powers of reconstruction, historical or ethnographic; and the ‘canonical’ version so obtained is always tentative, pending further evidence.

2.7.3.3. As representations of human processes, the rules proposed so far have some grave drawbacks. As Thorndyke and Yekovich (1980: 41) point out, as long as the detailed processing involved in using story grammars had not been specified in terms of rules, such grammars are ‘relatively impervious to disconfirming data while remaining capable of explaining post hoc most empirical results’. Another drawback is that so far, researchers appear to suppose that a theory of comprehension can be worked out and tested without a theory of production. Nearly all studies of story structure in period I surveyed work at the comprehension end, even though much of the data in recall is the result of story production carried out by the test subjects. The retelling of a story you have has received is by no means trivial, given the exigencies of text production (Smith 1980: 230; cf. Beaugrande 1984). Thus, many changes that the story undergoes when recalled by audiences may be due not just to ‘encoding’ and ‘retrieval’ of ‘story propositions’ (Thorndyke and Yekovich 1980: 35). Propositions must be reprocessed to meet exigencies of the linear media of a surface text (cf. Beaugrande 1984a, b, c). Moreover, as we saw in the quote from Johnson and Mandler (1980: 78f., cf. 2.7.3.1), many of the test stories have been written (‘generated’) by the investigators themselves — a practice with implications for the objectivity and generality being claimed for test results.

2.7.3.4. Finally, I see the drawback that story processing is probably not uniform in its details: different people (or the same person at different times) may do quite different things with one and the same story. There can be fluctuations of attention, interest, motivation, or background knowledge that materially alter the scheduling and thoroughness of processing — precisely in the domains of evidence we would want to seek out, e.g., reading times (cf. Haberlandt 1980; Uyl and Oostendorp 1980). In terms of design criteria (set down in Beaugrande 1981): how much freedom can a story grammar allow among different people and still count as a grammar?

2.7.4. The rules stipulate not all human processes, but only those dealing with the order of story constituents. We could evade some of the drawbacks enumerated above by limiting the rules only to sequencing procedures. We would for the time being disregard such matters as how words activate concepts, how story grammars are ‘instantiated’ during the act of comprehending, and so on. We do indeed find such a perspective, e.g., when Thorndyke and Yekovich (1980: 29) aver that ‘these schemata describe the syntax of narrative organization just as earlier phrase structure grammars describe the syntax of sentences’; and that ‘a schema provides expectations about the order of events in the story, and influences the encoding of events into their ideal order’ (1980: 37). Also, several of the recall tests involved disordered stories (e.g. Stein and Glenn 1979).

2.7.4.1. The drawback here is that numerous rules proposed so far would not work as sequencing rules. Thorndyke (1977: 80) expressly presents his rules with a ‘+’ symbol to ‘indicate the combination of elements in sequential order’ (cf. 2.3). But consider his very first rule:

STORY => SETTING + THEME + PLOT + RESOLUTION

Now ‘setting’ and ‘resolution’ are clearly in sequential order, appearing most often at the beginning and ending of a story, respectively. But ‘theme’ is defined as ‘the general focus to which the subsequent plot adheres’, and ‘plot’ as ‘an indefinite number of episodes’ (1977: 20). Obviously, it makes little sense to suppose that the theme and plot are ordered as sequential constituents; instead, both of them should pervade the entire story, including the setting and resolution (cf. 2.3). Similarly, I already remarked that the rule

SETTING => CHARACTERS + LOCATION + TIME

is not an ordering rule, and provided a counter-example. At most, we could have an unordered set, e.g.:

SETTING => {CHARACTERS, LOCATION, TIME}

On the other hand, Rumelhart’s (1975: 219) rules such as

EPISODE => EVENT + REACTION

ATTEMPT => PLAN + APPLICATION

APPLICATION => PREACTION + ACTION + CONSEQUENCE

would seem to be valid sequencing rules for a typical story: an event should precede the reaction, a plan should precede its application, and so on. The different story grammars, as we see, are making ‘rewrite rule’ do different things even though the rules look much the same (cf. 2. 14ff.). Note also that the time sequence of events in a story line need not be the sequence in which the events are narrated via the surface text (Conrad’s Lord Jim is an extreme example). And, time sequence often correlates with causality. (Stein and Glenn 1979). Hence, if we accept the restriction of the rules to sequencing, we need to reconsider the various dimensions implied in the ordering of constituents: temporal, narrative, and causal.

2.7.5. The rules do not account for the processing of stories, nor for the sequencing of stories, but only for abstract expectations about stories. In this perspective, the incompleteness and inconsistency of the rules as noted above seem less damaging. It is no longer claimed that story-tellers or their audiences must behave according to the rules, but only that they can use rules as a means of general orientation in dealing with stories. The grammar need not represent in detail what every audience does with the story under all circumstances; the grammar would only predict tendencies, such as the defaults and preferences, that people use for stories. For example, a story would normally be told with the events in their original time sequence, but the story-teller could do otherwise, provided appropriate signals were included in the text. The empirical studies on recall of disordered stories (cf. 2.7.4), where people put events back in order, could render such a perspective attractive.

2.7.5.1. However, this perspective weakens the predictive power and testabil­ity of the grammar. Since the rules need not be manifested in the surface text, a given test failing to reveal their effects need not constitute a refutation. We have no exact criteria for recognizing an application of the rules in actual recall protocols, aside from the highest-level categories such as the main constituents of episodes being present or being in a certain order (ef. 2.7.4). Thus, the relationship of the surface text to the underlying mental representation we are postulating for the whole story becomes flexible, if not vague.

2.7.6. The rules stipulate the ideal toward which stories evolve in repeated retellings. This perspective should follow logically from that depicted in 2.7.5 ­— the rules as a point of orientation. Mandler and Johnson (1977: 113) argue as follows:

An orally transmitted story will survive only if it conforms to an ideal schema in the first place or has gradually attained such a structure through repeated retellings.

The grammar would not stipulate which stories can or cannot be told, but only which stories are likely to be retold. The difficulty now is the historical dimension involved in ‘repeated retellings’. Empirical demonstrations could ultimately be done only over long periods of time, e.g., by comparing earlier and later versions of a folktale from different centuries. An experiment would be one tiny step in the long-range evolution of a story form: test subjects would presumably be ironing out structural difficulties (e.g., missing or disordered constituents) in much the same way as traditional story-tellers should do throughout the ages. But a particular subject (or group of subjects) might contribute very little to the overall evolution of a story without thereby proving that the rules don’t work in a gradual fashion. Or, we might want to do experiments in which a story is passed on from one test subject to the next in an extended series (compare Bartlett 1932), rather than having all the subjects read or hear the original story and produce a recall protocol. This procedure could rely entirely on oral transmission, provided reliable transcripts were made from recordings.

2.7.6.1. However, we would also need to find out why — if the rules have this evolutionary effect — all stories do not end up with the same structure. The steady migration toward a ‘canonical form’ (2.7.3.1) or an ‘ideal schema’ (2.7.6) should gradually level all stories into one basic pattern. What we actually find, however, is a startling diversity of story patterns, even within a single culture. One explanation would be that memory is too unreliable to keep the ideal (or canonical) format firmly in mind at all times; differences would then be due to memory lapses. This explanation seems unattractive, since studies such as Lord’s (1960) show that story-tellers, especially in pre-literate, oral cultures, command imposingly large and accurate memories. I would argue instead that story-telling depends vitally on variety as a means to maintain interest (cf. 3.5ff.; 3.24ff.). The ideal version — the single story that might be expected to arise from evolution — would not be interesting, or not for long. Story-telling is always situated in a dynamic context of social interaction (cf. Chafe 1980), and these dynamics mediate against the imposition of a static, ideal version (cf. Beaugrande and Colby 1979).

2.8. I have now enumerated six perspectives that might be adopted regarding the ‘rules’ of story grammars. I have tried to show how each one entails certain drawbacks, especially in regard to the rules proposed by the usual story grammars. In early research, it was unclear which perspective was being adopted. In consequence, there were no unifying principles for constructing a story grammar, nor for controlling the creation of grammatical rules. Yet if story grammars are to have psychological reality — and I believe few researchers would want to abandon the claim that they should (cf. 3.23) — there must be empirical arguments that determine how a grammar is set up. For instance, Johnson and Mandler (1980) had to decide between proliferating the rules for ‘base structures’ or else introducing transformational rules that can account for story variants upon the same limited range of base structures; they took the latter alternative in hopes of containing rule proliferation and thereby upholding what they called ‘descriptive adequacy’. But the ultimate arbiter should be the degree of cognitive, rather than formal, economy.

2.9. Anyone who works with formal grammars as linguistic and psychological models will soon incur the dangers inherent in the freedom to create rules. Formal grammars are decidedly too powerful wherever they take no account of limitations imposed by human processing capacities. For example, I can see nothing in the story grammars that would prevent the generating of stories whose length, recursion, and embedding are infinite; yet we would all agree that human restrictions would never tolerate an infinite story. Thus, we must build in limitations not only on the number of rules, but also on the extent to which the rules should be applicable.

2.10. Rewrite rules are, I fear, hard to control on both counts. The constructor of a grammar can equate any category on the left with any other category (or set of categories) on the right. Taken as a formalism, the grammar sets no limits upon what can be ‘rewritten’. Psychologically, however, a rewrite rule is a claim that two categories (or sets of categories) are, under given conditions, equivalent enough to be substituted for each other. This claim must seem very strong in view of the difficulties of defining conceptual categories (cf. Rosch and Mervis 1975), and in view of the intricate, changeable associative capabilities of memory (cf. Loftus 1980). Thus, the equivalence claim inherent in rewrite rules needs to be both supported and constrained by empirical evidence.

2.11. The same challenge can be raised about transformational rules (see now Beaugrande 1998). As a formalism, these rules (often notated as rewrite rules) allow the conversion of one structure into another. Psychologically, these rules imply that there is some sense in which processing actually takes the one structure and ‘transforms’ it into the other. This too must seem a strong claim, and one for which empirical support in studies of syntax has not been forthcoming, despite the best efforts of ‘psycholinguistics’. If Johnson and Mandler (1980) wish to create transformational rules for deleting or moving story constituents, they should make it clear who uses the rules — besides of course, the investigators making up the stories. A line of argument such as:

The conditions for deletion depend upon the principles of inferability which govern the ability of a listener to recover the complete underlying structure. (Johnson and Mandler 1980: 71)

clearly goes beyond the interpretation of the deletion transformation as a pure formalism by appealing to something that a listener does (on deletion rules, see 2.20).

2.12. The whole question of using transformations in a well-formedness grammar has, I think, been inadequately resolved, whether in linguistics and in psychology. On the one hand, demands are advanced that every sentence (or story) be well-formed. On the other hand, the transformational rules permit virtually endless tinkering with the forms of the data, so that the demand is implicitly vacuous, and the theory circular, creating the very well-formedness that it has postulated as its foundation. If the data do not fit the grammar, rules can always be added or manipulated to impose conformity. The transformations foreseen by the rules don’t actually explain the data so much as get rid of them. Only if there were a complete, definitive rule set that did not allow this freedom to add and manipulate could the basic well-formedness claim be strictly tested. Until then, it remains a circular assumption arising from the design of the approach and bears no principled responsibility toward any one corpus of obtainable data.

2.13. Another way to interpret a ‘transformational’ grammar is as a claim that any pattern for a sentence or story is reducible to one of a small set of basic structures. Thus claim is also vacuous and circular as long as there are no principled limits on the number and extent of rules that can be created. The grammar itself stipulates that all sentences must contain the same categories (e.g. ‘noun phrase’, ‘verb phrase’). Once that stipulation is accepted, it is immediately obvious that conversions from any sentence to any other can be carried out. It is not obvious, however, that a single, consistent rule set for doing so will also satisfy the well-formedness criterion. On the contrary, the ad-hoc rules that manage a desired conversion might also allow many undesirable ones. It is also questionable whether the relationship between sentences claimed to share the same base structure is syntactic. In my view, the relation between, say, an active sentence and its corresponding passive or interrogative is syntactic only in the trivial sense that they all contain the same structural categories; but in communication, the decision to use a passive or interrogative is always semantically or pragmatically motivated. Thus, if the transformations are to be psychologically relevant, we must build these motivations into the apparatus of rules.

2.14. I would argue that these principled objections to formal grammars are more pressing than the details of the rules themselves. The current debate has, for instance, revolved on the various kinds of rewrite rules, which critics of story grammars use to classify current them as ‘finite state’, ‘context free’ and so on (cf. Black and Wilensky 1979: 215ff.; Black and Bower 1980: 229f.). Their classifications would be entirely correct if the rewrite rules of story grammars contained only formal symbols. But instead we encounter fuzzy, content-based categories on both sides of the rules. Thus, categories like ‘setting’, ‘theme’, and ‘plot’ (cited in 2.7.4.1) are in no way formal symbols, but at most entities we can define or recognize only because they occur in something we already know to be a story. No story grammar proposed today can be ‘context-free’, even though the rewrite rules have only one category on the left, as long as that left-hand category is determinable only in its context. Consider here Thorndyke’s (1977: 79) rule:

where ‘attempt’ is indeterminate except in the context of a plan and goal for some agent. Thus, context is omnipresent, but hidden away in the interpretation of the symbols (cf. Johnson and Mandler 1980: 56) — a recourse disallowed for a formal context-free grammar (cf. Ginsburg 1966).

2.15. Not only do the symbols designate fuzzy content-based categories, but so do their junctions. As I remarked in 2.7.4.1, the ‘+’ symbol is being made to do too much and too varied work. In a formal grammar, it could only signify that two symbols appear adjacently. In a story grammar, this junction depends upon at least three kinds of linkage: (a) the sequence of mention in the surface text; (b) the temporal progression of story events (whether or not they are mentioned in temporal order); and (e) the causality whereby events and actions depend upon each other (again, whether or not this is reflected in the order of mention) (cf. 2.7.4.1). Several story grammars have clarified this matter by labeling the links between constituents, e.g. Johnson and Mandler’s (1980: 60) ‘and’, ‘then’, and ‘cause’; or Stein and Glenn’s (1979: 59ff.) ‘allow’, ‘initiate’, ‘motivate’, ‘result’, ‘and’, ‘then’, and ‘cause’ (cf. 2.24). Beaugrande and Colby (1979: 48) further proposed to distinguish causality according to the type of implication: cause (first event creates necessary conditions for a second); enablement (first event creates sufficient, but not necessary conditions for a second); reason (an event motivates a rational human reaction); and purpose (some event is the motivation for someone’s prior actions).

2.16. As the nature of the categories and junctions in the rules becomes better defined, a purely formal representation seems more and more impoverished. Neither finite state grammars nor context-free grammars would allow us to define junctions in terms of temporality and causality. For this reason, the criticisms of Black et al., though just in themselves, do not apply to the story grammars I am reviewing. Perhaps the whole controversy would be assuaged if we could phase out rewrite rules altogether, and state the rules in plain English, as in Beaugrande and Colby (1979: 54f.):

Notice the main CHARACTERS, and their PROBLEMS and GOALS.

Relate ACTIONS and STATES to PROBLEM-SOLVING and to GOAL-directed PLANS

Such a method would at least make it clearer what the rules require.

2.17. The difficulties I have raised can be further illustrated with the notion of recursion, a common feature of formal grammars. The ‘infinite’ set of strings a formal grammar can supposedly enumerate (cf. Chomsky 1957) hinges on infinite recursion of a finite set of categories. There could be no ‘longest sentence of English’ in any such grammar because some prankster could always come along and dob in another modifier, embed another clause, or whatever. But for a realistic grammar based on human capacities, we don’t want this much ‘power’. I have found that a recursion containing more than three members is comparatively rare in naturally occurring clauses or sentences (if you aren’t Spenser, Faulkner, Tom Wolfe, or the like). Each recursion makes the next one steadily less probable — a kind of ‘self-destruct’ mechanism that normally terminates recursion fairly soon.

2.18. A recursive story plot is also limited. A case in point is the quaint tale of the Old Farmer and His Stubborn Animals (used in Mandler and Johnson 1977: 127; Rumelhart 1977b: 273; Thorndyke 1977: 105f.; etc., etc.). (The original bard of this momentous tale was apparently Rumelhart, who trans-gendered and trans-animaled Ye Olde Englisshe folktale, The Old Woman and Her Pig; see Rumelhart 1977: 273.). A superannuated agrarian deploys one tactic after another to impel his contumacious donkey into a ‘shed’. Some of his sallies are pure recursions, where two actions are unrelated and could readily be moved without trouble, e.g., pushing the unmannerly quadruped from behind (a parlous act, I grant you) versus beseeching the dog to bark discordantly and affright it into locomotion. Other attempts are embedded recursions, where one action is initiated in order to enable another, e.g., bribing the cat with milk to scratch the dog so that the latter will bark with alacrity; or bribing the cow with hay to give the milk; luckily, no need to bribe the haystack. All these recursions are unlike those in a formal grammar for at least two reasons. First, we have not mere repetition, but linkages of temporality and causality among the individual recursions. Second, there is always a goal waiting to terminate the recursions as soon as it is attained — such this is the recursion of general problem solving (GPS), to wit:

Difficulties lead to goals to overcome them. This device accounts for much of the recursive flavor of GPS, since its reaction to being unable to apply a goal is to set up the goal of applying it. (Newell and Simon 1972: 447).

A third reason is not found in the Old Farmer Story, but in many stories studied by Schank’s merry consortium (cf. Schank and Abelson 1977; Wilensky 1980a). Here, each action is a little more drastic than the one before it, e.g., when Bill first ‘asks’ John for the latter’s bicycle, then ‘bargains’ with cash, and finally ‘threatens’ to ‘break his arm’, which last eventuates in the donation (Wilensky 1980a: 28). This mode of progression was dubbed planbox escalation by Beaugrande and Dressler (1981: 170ff.). For a clearer example, our old farmer might first conjure the obdurate donkey in language courtly and melodious, then in language menacing and obscene; then proffer a bouquet of the most delectable carrots that ever mammal munched; next administer an unceremonious poke in the wotsit with a pike-pronged pitchfork; all these no whit availing, he could secure an iron ring around the donkey’s neck and hook up with a thick chain to a powerful windlass inside the shed and, hey presto! Success at last! In this progression, each method is more forceful and likely to work until the last one is guaranteed to convey the stubbornest donkey into its foredestined shed (and probably for good).

2.19. A fourth reason why story recursion is not comparable to recursion in a formal grammar is that audiences’ attention span is limited. If a spotted and inconstant farmer launches too many sallies and comes a cropper alway, audiences get jaded and yearn to the see old codger give up or get kicked into oblivion, and their minds (or their whole selves) will wander off. So there must be a limit to the number of times either the dogged farmer or the surfeited audience can endure sundry dickerings with unfeeling dogs and cats; frankly, I find the story already a wee bit tiresome. And my head positively swims over a Russian version where a harried peasant, in quest of a humble drink of water to keep her husband alive, is required to bribe a river with a lime-tree leaf, and the lime tree with some thread (Johnson and Mandler 1980: 64ff.). Whether the senselessness of these bribes would astound those hapless peasants doomed to make requests of the Russian bureaucracy, I cannot say; I myself was courteously showered with libations in Russia, and the Moscow River is not drinkable for grisly reasons, God knows. But clearly, the actions in the story become correspondingly interchangeable, and their order immaterial, which is not true for the story of the old farmer and his ingratitudinous menagerie, nor in my escalating version thereof. One should always essay honeyed words before the steely pitchfork.

2.20. Just as recursion rules may allow too many additions, deletion rules may allow too many removals (cf. 2.11). In formal grammars, deletion rules are occasionally needed to amend ungrammatical sequences created as a side-effect of other rules. Also, deletion can show the grammatical equivalences between outwardly diverse (though equally inane) sequences like Clark’s and Clark’s (1977: 61):

(5a) The pen which the author whom the editor liked used was new.

(5b) The pen the author the editor liked used was new.

However, deletion is too powerful to the extent that it encourages not the proliferation of invisible data: an absent category would be present, had deletion not excised it. I recall a bold proposal to incorporate all sentences into speech act theory by arguing that there is a deleted ‘performative’ (e.g., ‘I assert’) in front of every sentence (assessment in Lyons 1976: 778ff.). The founding of a general theory almost exclusively on invisible data proved unconvincing, and the project was not pursued beyond the proposal stage. Similarly, if any category in a story grammar were deletable, it would be hard to justify calling it a ‘grammar’ at all. This issue has colored the recent controversy, insofar as story lines are closely knit in temporal and causal connections that can be left implicit, but left up to the audiences’ powers of inferencing (cf. Black and Wilensky 1979: 221; Johnson and Mandler 1980: 70ff.). For instance, many folktale protagonists have a homey ‘stay-alive’ goal which few story-tellers would bother to mention; (6) is certainly odd:

(6) Jack heard the army of giants clumping back and decided he would like to stay alive, so he hid inside a colossal and odoriferous shoe.

Similarly, if we hear that there is a beauteous princess to be rescued from a maleficent sorcerer, or a cankered hoard of strange-achievèd gold to be freed from a fire-tonselled dragon, we can safely assume by default that the protagonist hero will have a go. The goal need not be expressed in the story, whereas the attempt for the goal may not be deleted (Mandler and Johnson 1980: 308). I shall venture an account for this constraint later on (3.8).

2.21. The story tree, a popular schematic diagram for story structure, is also derived from the ambience of formal grammar. This formalism, like the rewrite rules (2.10), is being made to stand for several different things at once, especially when the tree is merely a graphic equivalent of the rewrite rules, which is the case for example in Thorndyke (1977) and Stein and Glenn (1979). Thorndyke’s (1977: 81) own rendition of the Old Farmer story, blown in every eye, shows a high level whose branchings are neither distinct from each other, nor ordered within the story sequence (cf. 2.7.5.1): the ‘theme’ probably includes the ‘setting’, and the ‘plot’ certainly contains the ‘resolution’. Further down, the branchings for ‘episodes’ are in a logical sequence (‘subgoal’, ‘attempt’, ‘outcome’), whilst those for ‘characters’ are not. On the lowest level are propositions, or, more accurately, clauses or sentences mechanically numbered just by their position in the story text. What all this succeeds in explaining is out of my welkin.

2.22. Rumelhart (1977b: 272) treed his more recent favorite, Mary and the Ice-Cream Man. In this suspenseful epos, the pristine heroine Mary, hearing the jingle-jangle approach of that many-flavored huckster, is stricken with desire for his confections; she dredges up from the floors of memory the funds wherewith she had been endowed on a recent birthday and hies into her abode with all the single-minded impetuosity of the sugar-addicted poppet. Unhappily, the story breaks off, leaving the denouement enshrouded in mystery (aptly symbolized by question marks in the tree), because Rumelhart wished to test audiences’ expectations about the (seemingly obvious) future actions of the cone-crossing pair. (In another, more terrible, version, she finds no funds and fetches a ‘revolver’ to secure the delectable ices [Rumelhart and Ortony 1977: 115]). The tree is decked out with general nodes like ‘cause’, ‘select’ and ‘try’ but also with specific ones like ‘rush’, ‘spend’, and ‘buy’, all strewn across vertical branches like so much tinsel at Yuletide. In contrast to Thorndyke’s tree, Rumelhart’s represents the protagonist’s planning, such as mediates between ‘hearing the ice-cream man’ and ‘rushing into the house’. No account is tendered of whether this mediation devolves upon a story-teller whose rendition were less infused with the poignant simplicity of Rumelhart (1977b: 265): ‘Mary heard the ice-cream man coming down the street. She remembered her birthday money and rushed into the house...’

2.23. Mandler and Johnson (1977: 120) propose to unveil the ‘underlying structure of the Dog story’. This canine object lesson (from Aesop) recounts the woe-begotten wages of greed and could thus edify any experimental subjects left unmoved by the Old Farmer and the youthful Mary. As you doubtless recall, a ravening dog carries a piece of (stolen?) meat across a bridge; seeing its reflection in the stream, opes its jaws to seize the meat of that same glassy essence and loses all — the importunate victim of optics and gravity in malignant collusion. Yet none of this gripping, insensate drama shines through in a tree made of arid nodes like ‘beginning’, ‘development’, and ‘ending’; not Aesop himself would recognise his fable so filtered.

2.24. Stein and Glenn (1979: 61) in turn would enchant their test subjects with Melvin the Skinny Mouse. Here was in sooth a sobering chronicle, recounting the misdeeds of a mouse who falls from righteousness upon encountering ‘a box of Rice Crispies’ unaccountably ensconced ‘underneath a stack of hay’. In a trice, the box is despoiled, and the erstwhile ‘skinny’ mammal, now ‘very fat’, lies engulfed in darksome contemplation of his pitfall: ‘Melvin knew he had eaten too much and felt very sad’. (Why this mouse should fall prey to pangs of guilt is also out of my welkin; any other would be chuffed to be stuffed.) Yet any embarrassment to us ourselves is spared by a tree with nothing more sorrow-laden than ‘initiate’, ‘motivate’, and ‘result’ — as compared, say, to ‘junk food’ or ‘pig out’.

2.25. Merely for the sake of discussion, I make bold to display a transition network drawn up by Beaugrande and Colby (1979: 60) for the Tom Tit Tot story, an eponymous ‘Suffolk tale’ dating from 1878 and recounting the Rumpelstiltskin name-guessing bargain. We wanted a formalism that stresses the continuity of event and state sequences in a story as viewed not just from temporal and causal sequences, but also from the perspectives of the characters.

In addition, we included events that were anticipated, but not realized, on the grounds that what happens at one point in a story is most richly comprehended in terms of what might have happened, but didn’t. We need to grasp the importance of interest and surprise value in the telling and hearing of stories, which slip through the rough fronds of story grammars (cf. 3.24). Since every ‘attempt’ has at least the two outcomes of failure or success, both might reasonably appear in a model dealing with comprehension and recall of plot lines. To appreciate this Rumpelstiltskin legend, an audience must know what would (but doesn’t) befall the ‘gatless’ (shiftless) and gutless ‘gal’ if an intolerable deal of ‘flax’ is not ‘spun’ every day for a month (the flax-mad king will ‘behead’ her); or if the name of the ‘black impet’ who does the job is not guessed (he will carry her off for a fate much worse than death). More worrisome yet, if she guesses the name before the month is out, she will lose both his aid and her noddle regardless. So, the most contrived coincidences must be fabricated. Whilst on the final eve she sits upon her stool, sullen and discombobulated, His Highness just happens upon Tom who is spinning lustily and lustfully in ‘an old chalk-pit’ in the forest and just happens to be singing a ditty with his own name, which the royal flax-collector then innocently repeats to the girl...and the Happy End crashes down like a whole ton of flax. Only a sour-pussed spoilsport would point out that the same mad marathon is due again in eleven months, during which she would be wise to take some serious spinning lessons.

2.26. The issues reviewed in this section should illustrate the enduringly uncertain sense in which ‘story grammar’ relates to formal grammar. I suggested that the separation of structure from content is not feasible (2.1-6). I noted that the notion of ‘grammatical rule’ can be seen in a different perspectives, each involving its own difficulties (2.7.1-2.7.6.1). The unduly great ‘power’ built into formal grammars was compared to the important limitations upon human processing capacities, as illustrated by the concerns of rewrite rules (2.10), transformations (2.11-13), recursion (2.17-19), and deletion (2.20). The categories of a story grammar will necessarily remain fuzzy and content-based, quite unlike arbitrary formal symbols. Finally, the construction of graphic representations as trees or networks should follow from clear decisions about how the structure of the story is built up in real time, however many ways it might be analysed after the fact (2.21-25).

3. Future scope and ecological value of story grammars

3.1. Having reviewed a range of issues raised by past story grammars, we might now weigh the outlook for the future. The ultimate merits of a story grammar might be sought in two criteria. The scope would be the number and range of stories the grammar covers. The ecological value would be the grammar’s contribution to our knowledge about how stories are told and understood by human beings. We saw in section 2 that, from a purely formal standpoint, there can be a substantial variety of possible grammars or grammatical constructs. Thus, we need to step outside the grammar itself to judge its merits.

3.2. The question of scope defines what counts as an example or counter-example for any theoretical construct. A grammar that could ‘generate’ only the sagas of the foiled farmer, the sweet-toothed Mary, the rapacious dog, and the gourmandising mouse, for all their stark grandeur of ‘events’, ‘causes’, ‘motivations’, and so on, would not be satisfying. Mandler and Johnson (1977: 113; 1980: 306; Johnson and Mandler 1980: 84) aim to encompass all folktales in the oral tradition — ‘oral’ because, as the quote in 2.7.6 reveals, they relate the need for a stable structural pattern to the limitations of memory. I would note here that, although some story grammarians addressed oral presentation and retelling (Mandler and Johnson 1977: 142; also, Stein and Glenn 1979: 75), others addressed either oral or written (Thorndyke 1977: 86), oral accompanied by written (Stein and Policastro 1982: msp. 46) (6), or exclusively written (Rumelhart 1977b: 283). This matter should be controlled, since the processing of oral and written discourse is obviously subject to differing conditions and limitations (Beaugrande 1984). Besides, the texts may have been ‘traditional’ but they were at best condensations and manipulations. Whether such alterations might affect the processing of the stories is another matter for closer scrutiny.

3.3. Rumelhart (1975: 213) first offered a grammar for ‘the structure of a wide range of simple stories’, and deployed both a made-up story of the ‘Mary’ type and an Aesop’s fable. Later, he added the basic notion of ‘problem-solving behavior’ as ‘a surprisingly simple motif underlying a remarkable number of stories’ (1977b: 269). However, his categories ‘plan’, ‘preaction’, ‘attempt’, and ‘application’ in the early paper (1975: 222) provide the essentials of problem-solving before the term itself appeared. Then, Rumelhart (1980a: 315) expressly limited his grammar to problem-solving stories as opposed to both (a) problem-solving texts that aren’t stories, and (b) stories that have some other basic organization.

3.4. An optimal story grammar should have in its scope all those texts which a given culture regards as stories. Cultural consensus rather than the story grammar would properly be the final arbiter (cf. 3.8). Unlike Mandler and Johnson (1980: 307), I would prefer to include novels and literary short stories, because these forms evolved ultimately from the oral tradition and thus should bear structural similarities to folktales, despite. Distinctions in the detailed execution of the text in tribute to literary conventions.

3.5. What then might all stories have in common? First, they must have at least two states, plus an event that leads from the first to the second (cf. Labov and Waletzky 1967; Prince 1973). A single state cannot be a story (Carroll 1960: 126:

(7) ‘Once’, said the Mock Turtle at last, with a deep sigh, ‘I was a real Turtle’. These words were followed by a very long silence [...] Alice was very nearly getting up and saying. ‘Thank you, Sir, for your interesting story’, but she could not help thinking that there must be more to come. (Alice in Wonderland)

Moreover, the transition event from the first to the last state should not be a matter of course. These samples would hardly be acceptable:

(8) It was autumn. Then the season changed, and it was winter. And spring was far behind.

(9) Socrates was born. He lived and died.

There must be at least two alternatives (and often, there are a great many) for the story plot, defined as a succession of states and events or actions (cf. 2.25). A dynamic way for story audiences to interact with a story-teller is by identifying with a story character and wondering how they would react to his or her choices and circumstances. They would favour stories with the formal character of problem-solving, a problem being defined as two states whose intermediate transition uncertain (2.4). It might not be a problem the protagonist explicitly sets out to solve, however; it might arise and be solved by the intervention of natural forces or fate (as frequently in the Arabian Nights). However, a protagonist who must actively solve the problem should elicit more active audience participation, as envisioned by some story grammarians (cf. Rumelhart 1977b: 269; Rumelhart and Ortony 1977: 113).

3.6. The requirement that stories have either implicit or explicit problem structure seems rooted in the very nature of information, which requires at least two possible transitions from one state to the next (cf. Shannon and Weaver 1949). Samples (8) and (9) violate this requirement and so are utterly devoid of interest. I have elsewhere suggested that informativity is a necessary condition of textuality which all actually occurring texts meet in some degree (Beaugrande 1980a). It could be argued that the temporal and causal connections in story lines are derivative from the dual need for coherence and informativity. The two states of the minimal story must be related in time, usually as earlier versus later; and a causal relationship strengthens coherence, even when the effect is not anticipated or desired. An unexpected causality brings a rise in information, as when magical events create novel causalities. In one Native American tale, a bridegroom drank rather more river water than was entirely prudent and got transformed into an enormous fish — the all-time best alibi for missing your own wedding. It blocked the passage, and his grumbling confreres had to portage their canoes around, until it was moved to retire by the bride’s well-meant if hardly fish-fitting gifts of moccasins and tobacco.

3.7. If all texts possess informativity, a problem-solving structure will not uniquely identify stories, and further criteria must be sought. One such criterion would be the presence of at least one animate agent — not necessarily a person, but some force-possessing entity that can act and react with the world (Prince 1973; Stein and Policastro 1982). Since the audience must understand the problem in terms of a recoverable goal, the agent is usually human or quasi-human enough to entertain human-like goals, as in the tales used by the story grammarians (cf. 2.21ff.). Animals may have character traits that fit their popular reputation (smart fox, greedy wolf, stupid cow), but the representation of the character will still be in human terms (e., thinking and speaking. The ability to recover causalities and to identify with an agent and with his or her goals presupposes a cultural consensus whereby the story audience is knowledgeable and can participates (cf. Johnson and Mandler 1980: 80). In contrast, the Native American tale War of the Ghosts used in Bartlett’s research was poorly comprehended by Anglo-American readers, especially the death of the protagonist who has been shot by ghosts on the warpath but drops dead only back home at sunrise (cf. Bartlett 1932; Kintsch and Greene 1978). I also noted in 2.19 the special inferencing needed for the recursive story where animals demand senseless bribes. Conversely, an appealing story has goals that people easily recognize as highly desirable. Indeed, the mentioning of the goal might be omitted precisely because of this easy recognition; such is true of all three characters in Tom Tit Tot, e.g., why in earth the king wanted all that flax. The attempt cannot be deleted, though, because it sets in motion the goal-path and injects the uncertainty and interest. This account agrees with the constraints on deletion cited in 2.20

3.9. Further constraints on the event configurations that qualify as stories are needed in regard to the identity of agents and the mood of verbs. To devise a text type that fits the grammar and yet remains a non-story, Black and Bower (1980: 231) made up a set of instructions on ‘how to catch a fish’. Before launching into the steps to follow, they provide a setting and initiating event:

(10) It is fishing season in Illinois and a friend asks you to go fishing, but you do not know how. Well, I am going to tell you how to catch a fish. First you need to get some fishing equipment. [etc.]

‘This text’, they remark, ‘contains a setting, theme, plot, and resolution as demanded by Thorndyke’s or Rumelhart’s story grammars’ (1980: 23 1). Stein and Policastro (1982: msp. 19) object that this ‘text’ cannot be a story, lacking a ‘specific protagonist’, as well as ‘an overt attempt, consequence, and reaction’. More precisely, to preclude sets of instructions as stories, the grammar would have to be tightened up to disallow both possibilities: (a) protagonist agents whose identity is neither determinate nor relevant, such that processing cannot recover the plans, goals, and attempts needed for audience participation (cf. 3.5); and (b) the conditional and imperative mood in verbs expressing the main events, such that temporal and causal relationships remain hypothetical (cf. Freedle and Hale 1979: 124ff.). Here, we may see some grammatical constraints of linguistic nature, though their underlying motivation stems from world-knowledge about events and actions.

3.10. As the grammars evolve, the extent to which they are or should be formal and specific may prove a decisive issue. This point is illustrated by the relationship between research on story grammars as compared with that on story schemas. In previous research, the two areas were not sharply differentiated. Rumelhart’s (1975) paper has ‘schema’ in the title, but ‘grammar’ from then on (2.2). Notice also the similarity of these definitions:

We will use the term ‘story schema’ to refer to a set of expectations about the internal structure of stories which serves to facilitate both encoding and retrieval. (Mandler and Johnson 1977: 112)

‘Story grammars’ [...] describe the types of information that listeners expect to encounter in a story and the organization they tend to impose on that information. (Johnson and Mandler 1980: 51)

Perhaps we might clarify the respective domains by carefully assessing the differences between the grammar approach and the schema approach. A comparison of the literature indicates that a story grammar can be viewed as a rule-set for relating the ordering of surface-text categories to the underlying schema (cf. 1.13). Thus, the grammar is a theoretical formalization that operates upon the knowledge organized within the schema, with major focus on the arrangement of categories in sequences. Prince (1973) and Thorndyke (1977), among others, assume that a story text must be composed of sentences, though this requirement is not met in many oral samples I have collected. No doubt influenced by sentence grammars, this assumption adds an well-formedness constraint which cannot be essential to the notion of a story grammar as such — an ordering of categories, not of sentences. This matter is not just a ‘subtle distinction in usage’ between ‘proposition’ and ‘syntactic clause’ (Thorndyke and Yekovich 1980: 29). Even my fairly authentic ‘Suffolk tale’ is not just sentences, viz.

(11) Well, once upon a time there were a woman and she baked five pies. And when they come out of the oven, they was that overbaked, the crust were too hard to eat. So she says to her darter: ‘put you them there pies on the shelf an’ leave ‘em there a little, an’ they’ll come agin’ — she meant, you know, the crust’d get soft. But the gal, she says to herself, ‘Well, if they’ll come agin, I’ll ate ‘em now’. And she set to work and ate ‘em all, first and last. Well, come supper time the woman she said: ‘Go you and git one o’ them there pies. I dare say they’ve came agin now.’ The gal she went an’ she looked, and there weren’t nothin’ but the dishes. So back she come and says she, ‘Noo, they ain’t come agin.’ ‘Not none on ‘em?’ says the mother. ‘Not none on ‘em’, says she. [etc.]

Grammar or no, the ill-fated bakeress and her strong-jawed offspring could hardly day: ‘not none on ‘em ain’t came agin’.

3.11. The literature also reveals what we saw in section 2: that numerous grammars can be proposed for the essentially same schema. The outcome is not only a variation in rule sets, but also a variation in the text sets that are to count as stories. Unlike studies of the schema, the literature on grammar is highly preoccupied with the distinction between stories and non-stories. It is questionable whether the schema itself is (or can be) defined precisely enough to carry this distinction. We would need to decide what the status of a non-story might be:

3.12.1. In purely formal terms, a non-story would violate the rules of the grammar. However, no grammar proposed so far claims to be either explicit or complete enough to provide such a test. Nevertheless, some artificial counterexamples have been constructed whose status as non-stories seems intuitively clear, even without consulting a grammar, e.g.:

(12) The ball was red and white. The ball was stored in the gym. The ball had a small hole in it. The ball was used for volleyball. The ball could fit into a person’s hand. The ball was dirty on one side. The ball was three inches around. (Stein and Policastro 1982: msp. 37)

Still, the Black and Bower example in 3.9 showed that the decisive criteria for excluding non-stories (e.g., sets of instructions) are not well defined in current grammars. Interminable revisions might be needed before the grammar is robust enough to exclude all non-stories.

3.12.2. A non-story might be incomprehensible, because its constituents were out of reasonable order or were lacking in ¡my temporal and causal relatedness. This measure is also not likely to be exact, because different audiences possess their own capacities and dispositions for making sense out of a presented text, for example, performing inferences to offset discrepancies or discontinuities. One variant of Wilensky’s (1980a: 29) gripping narrative about John lusting for Bill’s bicycle (cf. 2.18) takes a strangely fruity turn:

(13) John told Bill he would break his arm if he didn’t let him have it. Bill ate a banana.

In Wilensky’s vision, an audience might infer that eating bananas makes Bill preternaturally strong (like Popeye with his spinach); or that Bill plans to run and have John skid on the banana peel and land on his asphalt. When confronted with apparent incoherence, people are more likely to seek a solution than to output a message ‘ungrammatical’ and abandon all hope of understanding. Hence, whether a sample is a non-story could depend on the audience’s disposition, not just on the surface text. After all, I suppose somebody must read Finnegan’s Wake.

3.12.3. A non-story might be comprehensible, but only with substantial cognitive strain. This measure escapes the quandary raised in 3.12.2. We can test reading time and recall accuracy to estimate difficulty of comprehension. The lower performance on ‘less than well-formed stories’ (Johnson and Mandler 1980: 79) would be one effect of this strain. Nothing prevents anyone from telling or understanding such stories, though there is more effort demanded. But we should bear in mind that many people enjoy a story precisely because it challenges their understanding, witness the popularity of intricate detective stories told with omissions and displacements (e.g., Conan Doyle’s Valley of Fear). Thus, we might obtain the paradox that many celebrated, enduring stories get a marginal rating from the grammar.

3.12.4. A non-story might be comprehensible, but only if the audience performed substantial rearranging and inferencing. This definition supplies the reason for the ‘strain’ suggested in 3.12.3. The experiments showing that disordered stories had been reordered in recall to fit the grammar more closely (Stein and Nezworski 1978: 187) provide good evidence to support this view. A study of mine where tenth-graders heard an O. Henry story with its typical flashbacks, omissions, and surprise ending, and then had to retell it yielded similar results. However, the tenth-graders normalised their versions and thus unravelled the intricate texture of the story that made the surprise ending possible and effective. Again, there is a contradiction: tension between the patterning suggested by the grammar and the entertainment value of the story (cf. 3.12.3).

3.12.5. A non-story might be comprehensible, but devoid of all interest, so that nobody could endure it and would forcibly eject the story-teller. The audience might have no trouble understand­ing such a story, but would forcibly eject the story-teller or run madly to the back of beyond to avoid suffering through it. This possibility has not been addressed openly in any story grammar — a good thing, considering the lethally unprepossessing pallor of their ‘stories’ about characters who, not coincidentally, resemble the ‘John’ and ‘Mary’ of untold linguist’s examples, whiling away their ‘grammatical’. lives A grammar might specify that a ‘well-formed’ story should contain challenging problems, violent conflicts, grave dangers, high suspense, and so on. Propp’s (1928) approach stressed elements such as ‘struggle’ and ‘victory’; Colby’s (1973: 646) included elements like ‘betrayal, challenge, con­frontation, provocation, attack, and escape’; the users of Meehan’s (1977: 96) story program enjoyed ‘making the problem very hard’; Bruce (1978) pointed up the importance of ‘conflict’; and Brewer and Lichtenstein (1981) found that the degree of suspense influenced audiences’ ratings of sample stories. These criteria entail values judgment about story quality, not just about story form; and well-formedness correlates with, rather than collides with, interest and entertainment (cf. 3.12.3-4).

3.13. A suitably adjusted outlook on non-stories would affect not only the scope of story grammars, but also their ecological value (cf. 3.1). Several investigators outside the story-grammar approach have recently objected to the latter on the grounds that a story must have some element of surprise, a building and release of expectations (Beaugrande and Colby 1979; Kintsch 1980; Morgan and Sellner 1980; Brewer and Lichtenstein 1981). Stein and Policastro (1982: msp. 15) retort that ‘the notion of an unexpected event, even though not explicitly stated in the definition of the story, is implied by the inclusion of an emotional response on the part of the protagonist’. This argument will not hold up, however, even if we accept its doubtful premise that ‘emotional reactions to events occur primarily’ when ‘the event has violated the pro­tagonist’s expectations’ (1982: msp. 15) — certainly, emotions can accompany expected events as well, such as the sugar-caked weddings which wind up uncountable stories as if the plot had succumbed to an overdose of sanctified domestic joy. Besides, the story grammarians themselves assume that this response of the protagonist can be (and often is) deleted from the story (Mandler and Johnson 1977: 121; Johnson and Mandler 1980: 72); and Stein and Policastro’s (1982: msp. 68) own findings with adult judges confirm the assumption. Of course, those who stress the role of surprise would insist that the unexpected event is precisely what must not be deleted.

3.14. The very inception of story grammars makes it unsuited for dealing with surprise. First, a theory which posits ‘well-formedness’ as a ‘precondition for all members of a category would naturally incline to view a violation of expectations as an ungrammatical occurrence, and hence outside the domain of concern. Second, the grammars attempted to formalize only the given events of the story and their impact on a character; except for Rumelhart’s fragments (cf. 2.22), they were not concerned with hypothetical events in whose context the given events seem more or less surprising. Third, a categorical grammar is unlikely to exploit the concept of probability. On the contrary, by emphasizing direct causality, the story grammars make the event chain seem logically inevitable. But an audience will participate in a story most intensely if the goal is very difficult and the desire to attain it is very strong, e.g., winning the lily-like hand of the matchless daughter of a legendary sultan only after performing spectacular feats of strength and endurance that kick the stuffing out of classroom physics.

3.15. The que