In this essay, we argue that stories about one's experiences, and the experiences of others, are the fundamental constituents of human memory, knowledge, and social communication. This argument includes three propositions: 1) Virtually all human knowledge is based on stories constructed around past experiences; 2) New experiences are interpreted in terms of old stories; 3) The content of story memories depends on whether and how they are told to others, and these reconstituted memories form the basis of the individual's "remembered" self". Further, shared story memories within social groups define particular social selves, which may bolster or compete with individual remembered selves.

Our style of presentation is discursive and probably prone to overstatement, as we seek to emphasize the differences between our position and competing views in cognitive psychology and cognitive science. Where suggestive empirical research is available, we adduce it. However, we do not believe that a definitive body of empirical evidence is presently available on one side or another, and we have not attempted to fashion a fully data-based theory here.

In the first major section, we lay out the shape of our overall argument. In following sections, we discuss each of our premises in more detail.

I. STORY TELLING AND UNDERSTANDING: THE BASIS FOR HUMAN MEMORY

For thousands, maybe millions of years, people have been telling stories to each other. They have told stories around the campfire, they have traveled from town to town telling stories to relate the news of the day, they have told stories transmitted by electronic means to passive audiences incapable of doing anything but listening (and watching). Whatever the means, and whatever the venue, story telling seems to play a major role in human interaction.

However, the role of story telling and story understanding is far more significant in human memory than simply being an example of one kind of human interaction. The reason that humans constantly relate stories to each other is that stories is all they have to relate. Or, to put this another way, when it comes to interaction in language, all of our knowledge is contained in stories and the mechanisms to construct them and retrieve them.

Philosophers, psychologists, artificial intelligence types, and occasionally even linguists concern themselves with discussions of "knowledge." We talk about what people know, attempt to formalize what they know, make rules about what can follow from what they know and so on. But, with the exception of one part of the AI community, the subject of how people use what they know rarely comes up. It is in this discussion of the use of knowledge that the idea of knowledge as stories becomes significant.

Simply put, humans engage in two broad classes of actions involving language that depend upon knowledge. They try to comprehend what is going on around them and refer to what they already know in order to make sense of new input. And, they attempt to tell things to others, again referring to what they already know in order to do so.

Knowledge is Functional

It is important to recognize that knowledge is functional; it is structured not to satisfy an elegant logic, but to facilitate daily use. However, when we say that all knowledge is encoded as stories (plus mechanisms to process them), we must deal in our analysis with bits of apparent knowledge that don't seem to be stories, such "Whales are mammals," "I was born in New York," or "Stanford is in California." This we do in Section II. In Sections V through VIII, we cover the facilitory types of knowledge necessary to process stories.

The idea that knowledge is inherently functional, that it exists to be used for some purpose, imposes a constraint on how we talk about knowledge in this essay. We do not talk about what people know, but about the processes they engage in that utilize what they know. To this end, we can ask what people do that utilizes knowledge. Here are some of these things:

people answer questions
people make plans and inform others of them
people comprehend what others are saying
people inform other people of events that have taken place
people give advice to other people

This is not intended to be a complete list of what people do in their mental lives, but it is intended to characterize a great deal of what people do mentally that involves the use of language. In each and every one of the situations listed above, the knowledge that people use to help them is encoded in the form of stories.

The stories behind "facts"

When we find ourselves saying that "I was born in New York" we could be doing so for any of the reasons for talking stated above. We could be answering a question. We could be prefacing some advice that we are about to give (perhaps about what to see in New York). Or this could be part of an explanation of some events that have occurred. Whatever and whenever such a phrase is used, it is, fairly obviously, an abbreviation of a much longer story.

Human memory is a collection of thousands of stories we remember through experience, stories we remember by having heard them, and stories we remember by having composed them. Any story in memory could have gotten there in one of these three ways. The key point is that, once these stories are there, they are relied upon for all that we can say and can understand. Obviously, one can't recall where one was born. One is told the story of one's birth, and when required, one can tell others. There are many ways to tell this story, including the very short version -- "I was born in New York" -- that sounds like a mere fact instead of a story..

Such very short versions are not the results of "table-look up in memory" -- that is, finding the birthplace slot and reading its value. This may work for computers but it makes no sense psychologically. Search in human memory is a search for stories. The linguistic expression of those stories can range widely. We can say as much or as little as we like of the story. When we know that a story is funny or weird, we might digress and tell the best part. In fact, when one of us (RCS) is asked about his birth by someone who seems willing to listen he can tell two stories, one about why he was born in Manhattan rather than Brooklyn where his parents lived and the other about his short lived middle name (Wilco) that came from having overzealous Air Force officers for parents.

There are several stories of one's birth, and many more stories that comprise one's life. These stories get strengthened in the telling. The memories become more real because we tell them. When we tell them in the abbreviated way, they are simply that, abbreviated stories. Such really small stories should not be confused with factual knowledge. We propose that there is no factual knowledge as such in memory.

What about "Stanford is in California?" Surely this a fact in memory? Actually, it may be more of a derived fact than a fact that exists as such in memory. One of us was a professor at Stanford for five years. Ask him about Stanford and he can think of hundreds of stories, any one of which one could use to derive the needed information. Just as there is no slot for birthplace in memory, neither is there a slot for Stanford filled in with various characteristics of the place. It is possible to have memorized stuff about the acreage of the campus for example, but such memorized knowledge is hardly the usual sort of knowledge that we rely upon daily. The issue is not whether we can memorize facts, but whether that process has much to do with normal memory functioning.

Deriving static factual knowledge from stories we have in memory is of course quite possible to do. But, the fact that we can do this should not confuse us. This is yet again a very abbreviated story. In fact, even the phrase "Stanford is in California" is a story for one of us. When he was at Stanford he took a course in Yiddish. The first sentence the students learned to say was "Stanford is in California" since, as it happens, that sentence sounds exactly the same in Yiddish as it does in English.

Even such banal facts as "whales are mammals" can be stories. We need to guard against thinking that what we are talking about here is the attempt to understand such a sentence. We will deal with understanding later. Here we are discussing where and in what form that knowledge might reside in memory. In any discussion of static knowledge, particularly amongst those who believe in semantic memory, the whale issue is significant. For instance, knowing that whales are mammals we can predict things about how they suckle their young. But, as it turns out, we have never needed to use such knowledge. Whale suckling has simply never come up in any conversation we can remember. What has come up are discussions of how to represent knowledge. In these discussions we have always maintained that, despite what biologists have to say to the contrary, whales are better treated as fish since an intelligent system would learn much more about them than this one non-fact. The problem here again is the formalist's view of knowledge that stands in contradistinction to the functionalist's view. Formalists make up names like "mammal" in an attempt to make rules about knowledge that will be predictive. This is nice for them but has little to do with what human memory is actually like. In human memory whales are fish if they have to be anything at all, which they probably don't. The real role for whales in memory is as a part of Jonah stories, or Sea World visit stories, from which we really derive what we know about whales. Everything else is just rote memorization that we did in school.

Applying Old Stories to New Situations

What we know that seems factual is actually derived from personal stories. Similarly, what we can say about things that we believe is usually adapted from personal stories as well. When we ask someone for their opinion on a subject, and they produce what seems to be a truly creative response, that is, a response that surely they have never uttered before, careful examination tells us that old stories in memory are the ingredients of the seeming novelty.

To explore this idea, we asked the students in a graduate seminar why Swale had died. Swale was a racehorse that had won many important races and one day was found dead (at the age of three). This was told to the students and then they were asked to explain what had happened. They came up with some very creative hypotheses, including murder plots, drug overdoses, stress problems, and so on. The more we pushed, the more creative they became; for example, the "Janis Joplin Memorial Reminding" which basically said that Swale was too rich too young and couldn't stand the success so he overdid it by living "life in the fast lane."

The point is that even such new responses, explanations of totally new events, are really just rewrites of existing stories in memory, adapted to fit new circumstances. All of there new responses were already available in memory, or to put this another way, the responder already knew an answer. His problem was selecting from the stories he knew.

People have opinions about a large range of topics which are derived from the stories that exist in memory. They know what they think, and more importantly, they have already thought up what they are likely to say long before they say it. The respondent's actual task here was to determine which of the many already known answers was relevant to the question at hand.

Understanding the world means explaining what has happened in it in a way that seems consonant with what you already believe. Thus, the task of an understander who has a memory that is filled with stories is to determine which of those stories is most relevant to the situation at hand. He then uses the old story as a means for interpreting the new story. Doing things in this way makes a seemingly unmanageable task much simpler. Understanding the world is phenomenally complex. Finding a story that is like the one you are now seeing is much easier. The fewer stories you have in memory, the easier it is. To a man with a hammer, everything looks like a nail. But if you have many tools, you had better have a good system for knowing when to use them.

Scripts

In the mid-seventies at Yale, in our work on designing programs that understood English, or natural language processing (Schank & Abelson, 1977), we applied the concept of a script. A script is a set of expectations about what will happen next in a well-understood situation. In a sense, many situations in life have the people who participate in them seemingly reading their roles in a kind of play. The waitress reads from the waitress parts in the restaurant script, and the customer reads the lines of the customer.

Scripts are useful for a variety of reasons. They make clear what is supposed to happen and what various acts on the part of others are supposed to indicate. They make mental processing easier, by allowing us to think less, in essence. You don't have to figure out every time you enter a restaurant how to convince someone to feed you. All you really have to know is the restaurant script and your part in that script.

Scripts are helpful in understanding the actions of others as long as we know the script they are following. Scripts also enable computers to understand stories about stereotypical situations. When a paragraph is about a restaurant, we can realize with very little effort that we need not wonder why the waitress agreed to bring what was asked for, and we can assume that what was ordered was what was eaten. To put this another way, not everything in the world is worthy of equal amounts of thought, and restaurant stories are readily understandable by a computer armed with a good enough restaurant script. In fact, not too much thinking has to be done by a computer or a person if the right script is available. One just has to play one's part, and events usually transpire the way they are supposed to. You don't have to infer the intentions of a waitress if her intentions are already well known. Why concentrate one's mental time on the obvious?

Taken as a strong hypothesis about the nature of human thought, scripts obviate the need to think; no matter what the situation, people may do no more in thinking than to apply a script. This hypothesis holds that everything is a script, and very little thought is spontaneous. Given a situation, there are rules to follow, the way things are supposed to be. We can follow those rules and not think at all. This works for all of us, some of the time. People have thousands of highly personal scripts used on a daily basis that others do not share. Every mundane aspect of life that requires little or no thought, such as sitting in your chair or pouring your daily orange juice, can be assumed to be a script. In fact, much of our early education revolves around learning the scripts that others expect us to follow. But, this can all be carried a bit too far. Situations that one person sees as following a script may seem quite open-ended to another person. The more scripts you know, the more situations in which you feel comfortable and capable of playing your role effectively. But, the more scripts you know, the more situations you will fail to wonder about, be confused by, and have to figure out on your own. Script-based understanding is a double-edged sword.

Scripts are also a kind of memory structure. They serve to tell us how to act without our being aware that we are using them. They serve to store knowledge that we have about certain situations. They serve as a kind of storehouse of old experiences of a certain type in terms of which new experiences of the same type are encoded. When something new happens to us in a restaurant that tells us more about restaurants, we must have someplace to put that new information, so that we will be wiser next time. Scripts therefore change over time and embody what we have learned. For this reason, my restaurant script won't be exactly like yours, but they will both include information such as one can expect forks to be available without asking unless the restaurant is Japanese. Thinking in most contexts means finding the right script to use, rather than generating new ideas and questions; so, essentially, we find it easier to apply scripts than to reason out every new situation from scratch. But scripts aren't the complete answer. Obviously, we can understand some novel experiences even if no script seems to apply. We do this by seeing new experiences in terms of old experiences.

Beyond scripts: Indexed stories

When a prior experience is indexed cleverly, we can call it to mind to help us understand a current situation. This process can lead to brand-new insights. All people reason from experience. The differences among reasoners depend upon how they have coded their prior experiences in the first place. We are not all reminded of the same things at the same time.

If a prior experience is understood only in terms of the generalization or principle behind the case, we don't have as many places to put the new case in memory. We can tell people abstract rules of thumb which we have derived from prior experiences, but it is very difficult for them to learn from them. It is hard to remember abstractions unanchored in specific experiences, but it is relatively easy to remember a good story. Stories give life to past experience. Stories make the events in memory memorable, to others and to ourselves. This is one of the reasons why people like to tell them.

We are more persuasive when we tell stories. For example, we can simply state our beliefs, or we can tell stories that illustrate them. If John explains to Bill that he is in a quandary about whether to marry Mary or Jane, and if after listening to John's description, Bill responds Mary, his reply would usually be seen as useless advice. We need justifications for the beliefs of others in order to begin to believe them ourselves. If Bill responds, Mary because Mary is Irish, and Irish girls make good wives, he is being more helpful but not necessarily more believable. But, if Bill responds with a story about a similar situation that he was in, or that he heard about, and how the choice was made in that case, and how it worked out, John is likely to be quite interested and to take the advice offered by the story more to heart. Why?

Thinking involves indexing. In order to assimilate a case, we must attach it someplace in memory. Inaccessible information is not information at all. Memory, in order to be effective, must contain both specific experiences (memories) and labels (indexes) used to trace memories of experiences. The more information we are provided with about a situation, the more places we can attach it to in memory and the more ways it can be compared to other cases in memory. Thus, a story is useful because it comes with many indexes. These indexes may be locations, attitudes, beliefs, quandaries, decisions, conclusions, or whatever. The more indexes we have for a story that is being told, the more places it can reside in memory. Consequently, we are more likely to remember a story and to relate it to experiences already in memory. In other words, the more indexes, the greater the number of comparisons to prior experiences and hence the greater the learning.

Telling Old Stories

The "grandfather model"

With this in mind, let's consider what we might call the grandfather model of memory. The "grandfather" is one who seems to tell the same story over and over again. Everyone has heard his favorite stories many times, but they indulge him in hearing them once again. Every now and then, a new story appears, or at least one that no one is sure he remembers hearing before. When this happens, even the grandfather is surprised, and he enjoys telling the story more than usual. In some sense, the grandfather model is really just a crystallization of something we all do.

The logician model

The grandfather model of memory is in opposition to the more common logician model. This model is all too common in scholarly attempts to understand the nature of knowledge. For centuries people have been composing lists of what we know and rules for deriving logical inferences as if such models had something to do with normal memory processes. People are not really logical at all. The idea of logical people is parodied in the popular mind by the Mr. Spock character in Star Trek for example, and, in general, is an image associated with scientists. The idea behind the logician model of intelligence is that every problem can be reduced to first principles and decided on the basis of some logic. Knowledge, in this view, is about rules of inference, principles, not stories. In popular fiction, such logical characters often fail to have some emotional quality that makes them human in the audience's eyes. In fact, any live logician model would have two more serious problems in real life. The first is an understanding problem, and the second a discovery problem.

The understanding problem is simply that humans are not really set up to hear logic. People tell stories because they know that others like to hear stories. The reason that people like to hear stories, however, is not transparent to them. People need a context to help them relate what they have heard to what they already know. We understand events in terms of other events we have already understood. When a decision-making heuristic, or rule of thumb, is presented to us without a context, we cannot decide the validity of the rule we have heard, nor do we know where to store this rule in our memories. Thus, what we are presented is both difficult to evaluate and difficult to remember, making it virtually useless. People who fail to couch what they have to say in memorable stories will have their rules fall on deaf ears despite their best intentions and despite the best intentions of their listeners. A good teacher is not one who explains things correctly but one who couches his explanations in a memorable (i.e., an interesting) format.

Story telling and understanding

In the end all we have are stories and methods of finding and using those stories. Knowledge, then, is experiences and stories, and intelligence is the apt use of experience, and the creation and telling of stories. Memory is memory for stories, and the major processes of memory are the creation, storage, and retrieval of stories. To build models of intelligence, or simply to understand the nature of intelligence, we must understand the role that stories play in memory. We must know how events become stories and how these stories are stored and later retrieved. We must know the indexes we construct that label stories, and we must determine how and why such indexes are created. A good theory of mind must include theories about how the stories of others are decoded to find indexes to enable retrieval and storage as well as to determine how and why our own stories appear in our minds in response. It must also contain a model of how and why we create new stories and of what happens to experiences that do not get encoded as stories. What we know is embodied in what we tell, and as we shall see, what we tell strongly determines what we know. We know what we tell and we tell what we know.

The process by which this all works is the reminding process. Each story we hear reminds us of one that we know and, in a conversational situation we tell that story if it seems appropriate to the audience, adapting it as we go to make it more relevant.

In this way, it is clear that storytelling and understanding are functionally the same thing. Conversation is no more than responsive storytelling. The process of reminding is what controls understanding and, therefore, conversation. Thus, seen this way, conversations are really a series of remindings of already-processed stories. The mind can be seen as a collection of stories, collections of experiences one has already had. A conversationalist is looking to tell one of his stories. He is looking to tell a good one, a right one, but to do this he must be reminded of one of the ones that he knows.

Generating relevant stories

It is almost as if we never say anything new; we just find what we have already said and say it again. But, we don't do this freely or randomly. There is a method to this system. We are always looking for closest possible matches. We are looking to say, in effect, well, something like that happened to me, too or, I had an idea about something like that myself. In order to do this, we must adopt a point of view that allows for us to see a situation or experience as an instance of "something like that." In other words, we must evaluate experiences with an intention of matching them to what we already have experienced.

The story-based conception of generation presupposes that everything you might ever want to say has already been thought up. This is not as strange an idea as it seems. We are not suggesting that every sentence one will ever say is sitting in memory word for word. Rather, an adult has views of the world that are expressed by ideas he already has thought up and has probably expressed many times. When asked for your view of Reagan, for example, you don't usually consider the problem for the first time. What you say, however, may not be something you have ever said before. Nevertheless, what you say will have a certain familiarity to it and will be something that you have thought before and possibly expressed in other words. Your views evolve, so what you say one time will not be identical to what you say the next time. But, the relation between the two will be strong and will occur to you as you begin to construct your new thoughts. New ideas depend upon old ones.

So the main issue in generation is really the accessing of whatever you already think about something and the expression of those thoughts. When we tell a story, we are doing the same thing. We are accessing the gist of that story and then re-expressing that gist in English. The words we use are not identical each time, but the ideas behind them are more or less the same. We take the gist of a story as it exists in memory and then transform that gist into an English expression, one that perhaps leaves out one point or embellishes another. The words we choose may depend upon the audience. The ideas expressed may depend upon our re-interpretation of past events in light of events that have occurred since the story that is being told took place.

We get reminded of what has happened to us previously for a very good reason. Reminding is the mind's method of coordinating past events with current events to enable generalization and prediction. Intelligence depends upon the ability to translate descriptions of new events into indexes that help in the retrieval of prior events. One can't be said to know something if one can't find it in memory when it is needed. Finding a relevant past experience that will help make sense of a new experience is at the core of intelligent behavior.

II. TYPES OF KNOWLEDGE OTHER THAN STORY-RELATED KNOWLEDGE

In our introduction, we stated a very strong position on stories as knowledge, claiming that all of what people know is in the form of stories, and structures that expedite the storage and telling of stories. There are, of course, other positions. For Tulving (1993), there are five kinds of stored memories, of which the two most prominent are episodic memory and semantic memory. In these terms, we are saying that there is no distinct entity called semantic memory. This stance grows out of a point of view in artificial intelligence, where it has proven more fruitful to simulate episodic memory processes (Schank, 1982). Old episodes are applied to such tasks as explanation, planning, learning, and summary description. If there is any semantic content in the propositional knowledge base for these tasks, it comes in the form of generic episodes growing out of clusters of similar events.

A Variety of Theoretical Positions

The "kinds of memory" issue is provocative, and both cognitive psychologists and computer scientists debate it hotly. It would be nice if the empirical evidence were clear on this issue, but it is not. Even striking case studies of brain injured amnesiacs can be differently interpreted for their implications about episodic and semantic memory. On the one hand, amnesia can occur in chronological chunks, e.g., good memory for everything before a given date, but no memory of any experiences after that date. (See e.g., Sacks (1985), ch. 2.) This phenomenon is consistent with an episodic memory position, although not definitive. Tulving (1993), for example, asserts that amnesiacs suffer from loss of semantic memory.

Our very strong position on the episodic side of this debate will raise some hackles. In this section we discuss a somewhat milder position that will still maintain the central role of stories. (The one thing we won't do is to opt for the theory that all knowledge is propositional, leaving episodes out entirely.)

Facts

It is common in considering two theories, one stronger than the other, to refer to them as "the weak theory" and "the strong theory" (e.g., Searle, 1980). We prefer to follow the scheme used in grading the sizes of olives, thus contrasting "strong theory" and "very strong theory". In this section, we will attempt to outline a position that is (merely) strong We will do this by considering proposed types of knowledge that on the face of it seem unrelated to stories.

One category of knowledge often offered in contrast to the episodic type is knowledge of facts. We discussed this briefly in Section I, where we took the stance that facts either: 1) are learned by rote, and never used again (except to recite by rote, perhaps to impress the listener with one's mastery of the subject); 2) or function as indexes to stories, for example, about one's birth in New York or life in California.

This position may seem somewhat cavalier, but we think it holds true for isolated facts in social situations where people are genuinely communicating with one another. If we consider bundles of facts, such as knowledge of the street map of a particular town, and we ask what such bundles are used for, we run into episodic considerations again. If you stop on the intersection of Main and Lakeside Street, and ask a passerby how to get to the Fourteen Carrots health food store, she will probably recite for you a little script for getting there -- continue west on Main Street, bear left at the Y-junction, and in three-tenths of a mile you'll see a shopping center on your left, and then...

Such a response can -- arguably -- be interpreted as a narration of experience rather than as a manipulation of propositions about the local geography. The argument here essentially reflects a clash of views about expertise and "expert systems (Feigenbaum, et al. 1971). Should experts be modeled as users of "case-based" reasoning on stores of relevant examples (Kolodner, 1991), or as symbol processors employing "rule-based" reasoning? We are squarely in the first camp.

A more indirect challenge to a strong or very strong position on knowledge as stories arises in the case of beliefs. What, if any, relation could there be between beliefs and stories?

Beliefs

We will not attempt a systematic categorization of types of belief, but we will consider two major varieties, belief in the existence of some entity or phenomenon, and ideological beliefs. Together, these should give us a good feeling for the issues involved. (There are some distinctions between belief and knowledge (Abelson, 1979), but these are not especially relevant in the present context.)

Existence beliefs

Popular existence beliefs include belief in God, in the reality of UFOs, in the Big Bang theory of the origin of the universe, and so on. In each case, the core belief is that something that seems fantastic is really there, is really true. Let us take a fourth example, belief in the existence of extra-sensory perception (ESP).

Some years ago, one of us (RPA) understood an exploratory interview study of the nature of belief in ESP, soliciting volunteers with strong belief in the existence of psychic phenomena. A point of major interest was whether the respondents, when asked, "Why do you believe in ESP?", would respond by reference to some source of data, or the feats of a purported psychic, or, rather, would they report a convincing personal experience.

External "facts" never turned up. One after another, subjects told stories. One woman reported that at noon one day, after a fight in the morning with her husband, a construction worker, she suddenly had a feeling of foreboding that her husband had suffered a terrible accident. Sure enough, at 11:59, a heavy crane had toppled on him and killed him.

An aspiring artist gave an account of an experience that led him to realize that he was clairvoyant, he said. He was on the fifth floor of a well-insulated building when he felt himself suffused in the sound and mood of a particular Mozart quartet. On going downstairs to leave the building, he discovered that someone on the ground floor was playing a record of this very quartet. He attributed this experience to his artistic sensitivity, one aspect of which was the ability to perceive the world in ways going beyond the ordinary five senses.

A third respondent, a man with a six-month old son, claimed that a few weeks before, he noticed when he came into the living room that his son was sitting facing three-quarters the other way. He tried sending a telepathic message to the baby to turn his head and look at him, and it worked! He said that since this incident, he had tried repeating his telepathic commands, and though he wasn't always successful, he felt that his technique was improving.

And so it went, one personal experience after another. these people insisted that ESP was possible, because they had witnessed it first-hand. Now, a skeptic could fairly readily debunk these reports; for instance, the construction worker was at a site two short blocks from his wife's office, and it is entirely possible she heard a crash. Furthermore, her ready admission that they had been fighting suggested that she might often have wished him harm--perhaps repeatedly having (gleeful?) forebodings. We might also regard these respondents as naive or neurotic, or whatever. But such skeptical comments are not to the point. The important observation is that we found ordinary people readily giving autobiographical experiences, not abstractions, as support for their belief in ESP. In fact, we can find very well-educated individuals giving comparable testimonials. Irvin Child, a respected scholar who supports empirical study of ESP (Child, 1973), writes that his view of the possibility of psychic influences was radically altered by an experience reported by his uncle, who accurately predicted a tragedy 30 miles away.

We suspect that several other existence beliefs--those whose truth is known to be subject to intense debate--are also apt to form indexes to stories. A popular sort of competition among UFO buffs consists in who can tell the most exciting story without violating present lore on the proclivities of space aliens. (You can't say that you were taken to Alpha Centauri, for example.)

A belief that commands high cultural consensus, on the other hand, such as belief in God, seems less likely to index striking personal experiences. In our acquaintance, when believers are asked by they believe in God, they tend to say that there must be a God, or that everyone in their family believes in one and they never questioned it, or give other propositional responses. The conclusion for our strong story-based position, therefore, is that in general, existence beliefs are closely bound to stories, but that there may be exceptions.

Ideological beliefs

Ideologies, whether political or religious, have morality plays at their core (Roseman, 1993). There are good guys and bad guys, and the bad guys, using illegitimate methods, are trying to bring about an evil state of affairs. This can only be averted if the good guys mobilize their forces, recruit people from the sidelines (who are in danger of being seduced by the bad guys), and press forward to glorious victory. Such a scheme was found in interviews by Roseman to characterize both anti-nuclear activists and anti-Communist hawks--with different content, or course. Abelson (1973), in a conceptual and computer simulation of the Cold War ideology of Senator Barry Goldwater, outlined a master story (or "Master Script"), which made possible a large number of coherent (if unsophisticated) hawkish responses to foreign policy questions.

The perspective on ideological systems is consistent with a storytelling model of belief. In effect, when an ideologist is challenged, he will tell you a story--the plot of the morality play or master story. The supportive "facts" given will tend to be stories, too, episodes illustrating the venality of the opponent or the effectiveness of mobilization of the virtuous.

Which comes first: the belief or the story?

It might be argued that the beliefs could stem from other sources, or hidden motivations, and that the personal or political stories are merely justifications for beliefs otherwise formed. A strong case can be made that people have no access to the real reasons for their behavior (Nisbett & Wilson, 1977), and research on cognitive dissonance theory indicates the facility with which subjects rationalize foolish or insensitive words or acts (see, e.g., Aronson, 1969). Politicians are even more facile at justifying almost anything (as we will see in Section VII). Which, then, comes first: the belief or the story?

With respect to the point of view we urge in this essay, it makes no difference whether the story is cause or consequence of the belief. They end of packaged together, so that the belief indexes the story, and the story supports the belief. Mention ESP to the wife of the construction worker, and she will tell the story of her husband's death; ask her instead what the foreboding about someone death might signify, and she will report her belief that it very well could be the omen of an actual tragedy. Ask the model, or the real Barry Goldwater, to defend his foreign policy beliefs, and he will tell you a story of the struggle between good and evil. Tell him a story of communist malfeasance, and he will immediately point out the confirmation of his beliefs. (Tell him a story of benign Communist behavior, and he won't believe it.)

We pursue the role of beliefs in indexing stories further in Section III.

Lexical Items

Words

How does knowledge of the meanings of words figure in our position? The lexicon seems to be preeminently of a semantic rather than episodic nature, and this would compromise the very strong position that "all knowledge is stories."

There are a fair number of cases in which individual words, while not themselves stories, do serve to index stories. In Section VII, we discuss "betrayal" as an index favor a generic story skeleton, to which a number of particular stories attach. And many names of individuals and places evoke stories: the Alamo, Paul Revere, and Henry VIII, to name three examples. Yet such example-giving merely forestalls the inevitable conclusion that the lexicon is really essentially semantic. To claim that every word is a story would lead to the unsupportable position that interpretation of the meaning of phrases and sentences is achieved through a process of combining stories. (A whole sentence might index a story itself, but it wouldn't be made up of mini-stories.)

Posing such challenges to the very strong version of our position, however, is like the political game of "Gotcha!" It misses what we consider to be the important issues by switching the level of analysis. Our concern is with meaningful social communication, and such communication does not ordinarily consist of single words. To see the significance of this observation, suppose that conversational turns were restricted to a single word at a time. Then these utterances, we claim, would become a vocabulary of story indexes. Picture a group of rumpled academics, sipping port in the common room. The collage has just passed a regulation that no member may say more than one word at a time. What sort of conversation can they have?

One scenario might go like this: "Revolution!" says the first speaker, trying to stir up something. A second says, "French", suggesting a stock example. "Cahiers", chimes in a third, who likes to put on airs. (He is referring to the bills of complaint sent in by groups of ordinary citizens during the French Revolution.) "Ineffective, " opines a fourth. And so on. Each word indexes a story, or extends a previous story.

Or imagine a word-at-a-time exchange touching on daily concerns. "Grading," says the first professor. "Staircase?!", suggests the second. The third shakes his head. "Dean," he warns. Again, each word is a story. Our basic point is that whatever the size of basic conversational chunks, each chunk will usually be story-relevant. Linguistic units smaller than a chunk shift the focus to a lower level of analysis.

Numbers

Numerical elements, like words, don't seem to be story-related. For convenience, let us focus on whole numbers. The situation here is similar to that with words. We certainly would not deny that people have arithmetic knowledge that has little to do with stories.

We can't resist, however, pointing out that particular numbers can index stories. As Bill James, the baseball statistician says, "Children...love those baseball numbers, even though they might hate math: They love the stories the numbers tell...the great rookie year followed by years of unfulfilled potential...a trade...the slow decline of the aging player..." (James, Albert, & Stern, 1993).

Later in this essay, we refer to the classic story of the prisoners who refer to jokes by number, thus allowing jokes to be efficiently retold simply by calling out numbers. One of us haphazardly chose to illustrate the prisoners' joke-telling with the number 42, which reminded the other of a story given in Adam's (19 80) zany book, The Hitchhiker's Guide to the Galaxy. It seems that a committee of scientists had programmed the most powerful computer in the universe to solve the riddle of the meaning of life, the universe, and everything. the computer whirred and crunched, and worked on this conundrum for several million years. Finally, with a breathless crowd waiting for the answer, the computer printed out, "Forty-two."

The astonishing number theorist, Srinivasa Ramanujan, was the protagonist of several number stories. One of these tells of the occasion when a colleague visited him and remarked that he had come in a taxi with the dull identification number 1729. Ramanujan instantly replied that on the contrary, 1729 was a very interesting number, being the smallest integer that equals the sum of the cubes of two integers in two different ways. (See Humez, Humez, and Maguire (1993) for this anecdote and many other tales revolving around numbers.)

As was the case with words, we concede the reality of a type of knowledge that--except for a few entertaining examples--is basically different from story knowledge. But again, numbers and numerical operations, if engaged in by people at all, do not ordinarily enter into social interaction.

Rule Systems

Another type of knowledge category that does not seem to involve stories might be called "rule systems". by this we mean a symbolic domain with a set of constraints on permissible transformations of state. The influential work of Newell and Simon (1972) promoted the concept of mind as an information processor analogous to a computer. This analogy focused attention on the particular class of mental activities involving rule systems, such as theorem proving, puzzle solving, and chess playing.

The limitation of all this is that very few people spend time trying to prove theorems (Fermat's Last notwithstanding), and when they do, they don't ordinarily talk about it. The same holds essentially true (with less force) for problem solving and chess playing. And when people do talk about these things, they tend to become more story-like.

Grammar

A better candidate for a symbolic domain, universally practiced, in which remembered experiences play little or no role is grammar. Most people learn grammatical rules without awareness of that learning, and are typically unable to articulate what those rules are. Why does one say, "I can hardly wait", but not "I can wait hardly"? Is it because adverbs go before verbs? No, "I can walk slowly" is all right. There must be a perfectly simple answer, but we confess we don't know what it is. Nor, like most people, do we particularly care (except out of idle curiosity). For purposes of dealing with spoken or written language, we can place reliance on subconsciously developed intuitions about what is grammatically correct, and what is not.

Perhaps the best way to characterize our position on memory is that virtually all its interesting features arise within the context of stories. Stories based on personal episodes have both cognitive and social advantages. Variously experienced by different individuals or groups, they are eagerly shared in conversations and mass communications with other individuals and groups. They span historical time and social space, spreading object lessons and encouraging social solidarity.

Stories are very flexible; they can be copiously indexed to help interpret a variety of new circumstances. They can be jointly constructed by two or more individuals. They are the stuff of conversation. By contrast, we are hard pressed to imagine useful conversations about theorems or grammar or whether a flounder has gills.

Linguists who admire the elegant formal properties of language will disagree (perhaps violently) with our judgment. They might argue, for one thing, that without knowledge of grammar, communication could not occur, and that therefore, grammar is the most important knowledge we have. Our counter argument would be that first of all, communication is certainly possible with a minimum of grammar. Make ideas nevertheless scramble him words. Second, grammar is only a vessel for conveying messages, not to be confused with the messages themselves. When someone gives a great speech, we do not praise the microphone and the amplifier (nor the speaker's grammar). Third, grammar has very little social function, other than the unfortunate use of the quality of grammatical construction to guess the social class background of the speaker.

If the reader feels that our view of the very dominant role of autobiographical experience in the totality of people's knowledge is overstated, please wait and see where this position leads us.

III. UNDERSTANDING MEANS MAPPING YOUR STORIES ONTO MY STORIES

People are only able to comprehend a small part of what is being said to them. Most of what we attempt to understand contains many aspects, including ideas, people, events, opinions, and so on. It all comes by at a very rapid rate, and our minds can only do so much with this barrage of information. We get reminded by everything, names, places, words, beliefs, situations. All these remindings proceed in parallel until we can stand no more, until we can listen no more. Once we find an interpretation, we have made our choice. We cannot think about all the possible ramifications of something we are being told. We pay attention to what interests us. We settle on a story we have been reminded of, and in effect, we hear no more.

We select the mental paths to take on the basis of what interests us. We express our interests by focusing on certain indexes, those that we can say we have been looking for, ignoring the potential indexes we are not prepared to deal with. Since we can only understand things that relate to our own experiences, it is actually very difficult to hear things that people say to us that are not interpretable through those experiences. In other words, we attend to what we are capable of understanding. When what we are attempting to comprehend relates to what we know, what we care about, or what we were prepared to hear, we can understand quite easily what someone is saying to us. If we have heard the same story or a similar story before, we can also understand more easily what we are being told.

Old Stories

Understanding, for a listener, means mapping the speaker's stories onto the listener's stories. One of the most interesting aspects of the way stories are used in memory is the effect they have on understanding. Different people understand the same story differently precisely because the stories they already know are different. Understanders attempt to construe new stories they hear as old stories they have heard before. They do this because it is actually quite difficult to absorb new information. New ideas ramify through our memories, causing us to have to revise beliefs, make new generalizations, and perform other effortful cognitive operations. We prefer to avoid all this work. One way to do this is to simply assume that what we are seeing or hearing about is just the same old stuff. The real problem in understanding then, is identifying which of all the stories you already know is the one that is being told to you yet again.

The number of old stories

In the shallowest form of understanding, a hearer has only one story he wants to tell. No matter what you say to him, he tells you this story --like the grandfather in an earlier section. He understands what you say as something that reminds him of the story he wanted to tell in the first place. Thus, his understanding algorithm need have no more in it than a detector for when you have stopped talking, and perhaps he doesn't even need that. One typical case of this kind of understanding involves people who we might label as crazy or senile, people who just rattle on without regard to the world around them.

A less shallow form of understanding takes place when a listener with many stories to tell pays enough attention to what you have said so he can relate the story in his repertoire that is most closely connected to what he has heard. In a sense, this still seemingly shallow understanding may be all we can really expect most of the time. This view may seem rather radical. After all, we do see and hear new things every day. To say that we never have to understand any brand new story may be overstating the case. But often, we understand new stories mistakenly.

Specifically, we don't understand them as new stories. They may be new enough, but we nevertheless persist in seeing them as old stories. To understand what we mean here, consider the possibility of this hypothesis in its strongest form. Let us assume an understander who knows three stories. No matter what story you tell him, he will tell one of his three stories back. If understanding means matching the story we are hearing to the stories we have already heard, the strong form of our hypothesis states that an understander must decide which of the three stories he knows is most applicable. When he has found some way to relate the new story to an old one he knows, we can claim he has understood the new story as well as could be hoped for.

Looked at this way, the strong hypothesis appears somewhat silly. Why should we label as understanding a process that merely differentiates among three stories? In some sense, we shouldn't. But, let's consider the same situation if the understander knows 10,000 stories. When he selects one as a response to the new one he has heard, he will most likely seem more profound than the understander who has only three stories. If he has used sound principles for selecting a story to tell from his data base of 10,000, we are unlikely to dispute his having understood the original story. But, the process of understanding in both cases is identical; only our subjective judgment allows us to decide that one understander seems to have "really" understood. We cannot look inside people's heads to see what the difference in their understanding of a new story is; therefore, from an objective evaluation of the output alone, we still can only measure understanding by how effectively and reasonably we think the responsive story relates to the input story.

The selection of old stories

Our argument here is that what someone is doing when he understands is to figure out what story to tell. Thus, the understanding process involves extracting elements from the input story that are precisely those elements used to label old stories in memory. In other words, understanding is really the process of index-extraction. This is an idiosyncratic process that depends upon what stories you have stored away and what indexes you have used to label those stories. In some sense, then, no two people can really understand a story in the same way. You can't understand a story that you haven't previously understood because understanding means finding (and telling) a story you have previously understood. Finding some familiar element causes us to activate the story that is labeled by that familiar element. In this way, we find things to say to those who talk to us. These things differ considerably from person to person, thus accounting for the very different ways in which two people can understand the same story.

People have powerful models of the world. Through these models, which are based on the accumulated set of experiences a person has had, new experiences are interpreted. When new perceptions fit nicely into the framework of expectations that have been derived from experience, an understander believes himself to have understood the experience. Understanding something is, after all, a relative issue. We only understand part of what could have gone on in a situation that is being described to us, for example. But, we can act as if we have understood when we can derive information from the description that we know how to place properly in our memory store. In other words, understanding, rightly or wrongly, usually means being able to add information to memory.

Anomalies in Stories

But we dislike failing to understand. When what we have been asked to understand is anomalous in some way, failing to correspond to what we expect, we must reevaluate what is going on. We must attempt to explain why we were wrong in our expectations. A failure to have things turn out as expected indicates a failure in understanding. People desire very much to remedy such failures. We ask ourselves questions about what was going on. The answer to these questions often results in a story.

People are constantly questioning themselves and each other in a quest to find out why someone has done what he has done and what the consequences of that action are likely to be. Thus, in order to find out how we learn, we must find out how we know that we need to learn. In other words, we need to know how we discover anomalies. How do we know that something doe not fit?

The premise here is that whenever an action takes place, we need to discover what might be anomalous about it. Anomalies occur when the answers to one or more of those questions is unknown. Then we seek to explain what was going on, and then we learn.

To get a handle on this process, we must attempt to sort out the kinds of anomalies there are. Knowing the kinds of anomalies gives us two advantages. In order for us to see something as anomalous, we must have been unable to answer a question about some circumstance. So, first we must discover the questions that are routinely asked as a part of the understanding process.

Every time someone tells us something or does something, an observer, in his attempt to interpret what he is observing, checks to see whether the actions involved make sense. But, actions do not make sense absolutely. That is, we cannot determine whether actions make sense except by comparing them to other actions. In a world where everyone walks around with his thumb in his mouth, we don't need to explain why a given individual has his thumb in his mouth. In a world where no one does this, we must explain why a given individual does have his thumb in his mouth. Clearly, making sense, and thus the idea of an anomaly in general, is a relative thing.

Relative to what? Naturally the answer is, relative to the stories one already knows. We are satisfied, as observers of actions, when the stories we hear match our own stories. When the match is very similar, we tell our version of the story. When the match is hardly a match at all, when we have a contradictory story, we tell it. Actually, the middle cases are the most interesting--when we have no story to tell. What do we do then? We look for one. We do this by asking ourselves questions.

People are not processing information with the intent of finding out whether something is anomalous and needs explaining, at least not consciously. In fact, quite the opposite is the case. An understander is trying to determine the place for an action that he observes. To do this, he must find a place in memory that was expecting this new action. Of course he may not find one since not everything in life can be anticipated. A tension thus exists between the attempt to find a story that allows us to think no more, and the desire to see something as new, and worth thinking about. So, an understander asks himself, unconsciously of course: Do I know a story that relates to the incoming story, and is it one that will allow me to rest from mental processing or one that will cause me to have to think?

Examples

Consider the following two stories:

Once while watching the demolition of a building in Chicago, I was struck by how ineffectively the work was being done. The wrecking ball hit one of the concrete supports near its center again and again with little result. It was frustrating to watch the lack of progress.

This poorly executed demolition reminded me of the time I saw a bull-fight in Spain. The matador kept dealing out blows to the bull with his sword with seemingly little effect. The failure to execute a "clean kill" made the whole affair grotesque.

The index that links these two stories has something to do with the way in which the goals of the two observed agents, wrecking-ball operator and matador, were not being achieved. It also has to do with the failure of a prediction task, namely, the prediction that the column (and the bull) will fall. The index is formed from the anomaly of a prediction failure. Anomaly is at the heart of index formation for reminding. Any anomaly is set against the backdrop of expected relationships among the salient details of a story.

Anomalies serve as the trigger for memory access. Because we have found an anomaly, we are forced to think about it. We want stories to be non-anomalous, to match directly to stories we have already understood. In a sense, finding a story anomalous forces us to find stories in memory also, but now we are looking for a different type of remembered story, this time one that we have not understood so well. In a sense, there exists a memory base of previous stories that fall into two classes; the first includes stories we feel we have understood. An understood story is one for which we have many examples. One could claim that these stories represent our beliefs. A belief under this view is a point of view we can illustrate with a number of good stories. (They tend to be of the form: "You know X never works out. Remember that old example of X we both knew? And there was another type of X I saw once, too.")

The second type of previous story we use to help us understand consists of stories that we found interesting but somewhat incomprehensible. The "steak and the haircut" story (given below) fits into this class. Most obvious remindings, the ones that make us feel that we have been reminded, as discussed in Dynamic Memory (Schank, 1982) are of this type. They rely on expectation failure. Expectation failure derives from anomaly. However the stories of which we are reminded during expectation failure have a different role in understanding. Their role is to begin the process of belief formation.

Thus, in the first case, non-anomalous story reminding, we are reminded of a story by a new story and feel compelled to tell that story as a response. In a sense, our old story is the means for understanding the new story, so overpowering the new story that we remember little of it. In this sense, we cannot understand anything new.

In the second case, anomalous story reminding, we are reminded of an old story, and it feels somewhat peculiar to have that reminding. Rather than feeling compelled to tell our reminded story, we feel curious about the reminding, wondering how we happened to think about it. In the "steak and the haircut" example, one of us was lamenting that his wife never cooked steak as rare as he liked, and the other was reminded of an experience in England years before, when a barber seemed unable to cut his hair as short as he wanted it. This type of story reminding is not reflective of existing beliefs. In fact, we don't yet believe anything because we are at the stage of pre-beliefs. We will soon believe something because we are comparing similar anomalous stories in the attempt to form a new generalization, which will likely yield a belief. In the steak/haircut example, the emerging belief has to do with the behavior of people who have adapted to a habitual range of variation in activities they perform for others. Requests made outside that range cause disbelief that the request could possibly be so extreme.

The demolition/bull-fight reminding is somewhat more elaborate. It may be seen in a variety of ways:

1. The frustrated observer (laborious destruction, botched kill)
2. The inept agent (crane operator, matador)
3. The blocked goal (destroy building, kill bull)
4. The noble object of destruction that holds up against all odds (pillar, bull)

While all these interpretations are plausible, only one consistent and coherent rendering is responsible for the formation of an initial indexed probe of memory. For the demolition/bull-fight example, aspects of the building demolition situation may give rise to many kinds of causal theories about the origin of the arrangement of collapsed building components, theories of decay to explain the pealing paint of the crane, explanations for the noise of the diesel engine, etc. However for the specific reminding in this example, the anomaly is in the theory of planning which generates the expectations that a worker with a task to do will select an appropriate plan with the right resources (methods and tools) to execute the plan expeditiously. It is also the viewer's inference that the crane operator has selected the same causal model of planning, but has picked an inappropriate method of the plan. Seeing the futility of the crane operator's and matador's actions, in the presence of an aesthetic goal to see the job done cleanly, produces frustration or disappointment, a type of behavior expectation failure.

In general, an anomaly-based index is a system of interrelated goals, expectations, events and explanations. The index needs to include causal information about events and expected outcomes as contrasted to real outcomes. It is the contrast between causal theories and causal realities that serves as the means for finding prior stories with the same anomalous character.

Commonalities between old and new stories

We might be tempted to imagine that we only create questions for ourselves when our curiosity is aroused by confusion about something in the world that we have observed, but we are also often forced by social circumstances to create questions for ourselves to answer. When somebody says something to us, we are supposed to say something back. But what? Is there always something worth saying? Whether or not we have something important to say, given that we have to have something to say, and given that this happens to us all the time, we have developed various methods of coping with this situation. We ask ourselves questions.

Clearly the questions we ask ourselves about what others have told us cannot be solely dependent upon what we have heard. In a sense, since we are both asking and answering these questions, we need to know that we can answer a question before we ask it. That is, the questions we ask serve as memory calls, requests to get information from memory that will be of use in the formulation of a response to what we have heard.

We are concerned here with input stories that are responded to by stories. In other words, we want to know what questions one might ask oneself in response to a story that would allow a story of one's own to come to mind. We can start simply by considering the paradigmatic case where the response to a story is a story where one says, "The same thing happened to me." However, the call to memory that might retrieve a story where the same thing happened is not likely to be: look for the same thing.

Obviously, the exact same thing never actually happens to anybody. What actually occurs are episodes in memory that bear some superficial similarity to the input story. Probably more differences exist in an "identical story" than similarities. Certainly places and people, time and context, are usually quite different. What is the same then?

Sameness, at the level we are discussing, exists with respect to plans, the goals that drive those plans, and the themes that drive those goals. Thus, when someone tells you a story, you ask yourself, Are there any events in my memory where I had a similar goal for a similar reason? In other words, when we hear a story, we ask whether at the broadest possible level of interpretation we already know a story like the one we are hearing. But, possibly the similarity would not exist at the level of plans. That is, one could easily recognize a similar situation and suggest an alternative plan that might have been followed, based upon one's own experience. So we can match new stories to old ones on the basis of identical goals. Therefore one question we can expect people to ask themselves is: Do I have a story in memory where the main goal is the same as that being pursued in the story I am hearing?

One thing we do when we understand a story is to relate that story to something in our own lives. But to what? One thing seems clear. Potentially, we can see any story in many ways. What does it mean to see a story in a particular way? It means that one has constructed an index that characterizes a "point" that one derives from the story. This point is usually highly idiosyncratic. Usually the points one derives from stories relate to goals, plans, beliefs, and lessons learned from a story. Here we are back to interestingness again. We don't consider points that fail to relate to beliefs we already have. We look instead for stories that verify beliefs we already have. When a new story can be absorbed into our memories as a "natural fit" with stories we already know, then we feel we have understood the story.

Consider for a moment what it might mean for an understander to believe something and not be able to give evidence justifying why he believes what he believes. Certainly inarticulate people have difficulty with that sort of thing. And, in fact, we do treat the ability to justify one's beliefs as a measure of intelligence and reasonableness. In other words, we expect intelligent people to have a story to tell that explains why they believe what they believe. But, how can they do this? The mental mechanisms that are available must be ones that connect beliefs to stories. The fact that we can do this is obvious. It follows, therefore, that beliefs are one possible index in memory. Construct a belief, and you should be able to find a story that exemplifies that belief.

We find what we want to say effortlessly and unconsciously. But, to do so, we must construct complex labels of events that describe their content, their import, their relation to what we know and what we believe and much more. It is effective indexing that allows us to have stories to tell, and enables us to learn from the juxtaposition of others' stories and the stories we are reminded of.

How Do We Understand?

A key point is that there is no one way to understand a story. When someone hears a story, he looks for beliefs that are being commented upon. Any story has many possible beliefs inherent in it. But how does someone listening to a story find those beliefs? He finds them by looking through the beliefs he already has. He is not as concerned with what he is hearing as he is with finding what he already knows that is relevant. Picture it this way. An understander has a list of beliefs, indexed by subject area. When a new story appears, he attempts to find a belief of his that relates to it. When he does, he finds a story attached to that belief and compares the story in memory to the one he is processing. His understanding of the new story becomes, at that point, a function of the old story. Once we find a belief and connected story, we need no further processing; that is, the search for other beliefs is co-opted. (This is the essence of Kelley's (1971) "discounting principle.") We rarely look to understand a story in more than one way. The mind cannot easily pursue multiple paths.

Telling stories back implies understanding

We tell stories for many reasons, one of which is to indicate to our listener that we have understood what he has said to us. Assessing understanding by assessing the relevance of a story that we hear in response may be the only avenue open to us. If we choose to measure understanding by the ultimate impact of a story on our permanent memory structures, we may be very disappointed. To the extent that people do understand anything at all, we can identify three different features of understanding:

1. Matching indices for story retrieval
2. Adding aspects of a new story to empty slots in an old one
3. Seeking further evidence for stories that were only tentatively regarded as correctly understood

Thus, the very strong hypothesis is that as understanders we are always looking for stories to tell back. We do this by extracting indices from what we hear and by using these indices to find stories we already know. When we find them, processing stops, and we wait to tell our story. We only incorporate what we have heard into memory when we feel that our own stories are inadequate in some way, for example, if a story is missing a piece. Such pieces can be supplied by other people's stories. We may find a story inadequate when we use it to exemplify a belief that we are not quite sure we hold. We are willing to consider new stories as evidence for or against those beliefs and, therefore, to record and to remember better the stories of others.

Scripts

We all understand differently--this much is obvious. The reason we understand differently is that our memories are different. Our experiences simply aren't yours. In order to understand anything, we must find the closest item in memory to which it relates. In Schank and Abelson (1977), we claimed that understanding required one to find the correct knowledge structure and to use that structure to create expectations for what events were likely to take place so that new events could be understood in terms of what was normal. Thus, when a story about a cocktail party was being told, an understander brought out his cocktail party script which told him about what ordinarily happens at cocktail parties, and he used that script to guide his understanding of the story he was about to hear.

Idiosyncratic scripts

In Schank (1982), this concept was extended to allow a more idiosyncratic view of understanding. The concept of a dynamic memory was proposed, one that changed in response to what it had understood as time went on. The conception of understanding developed there was that one's knowledge structures were more idiosyncratic than just standard scripts. Each of us has his own conception of a restaurant, formed after numerous restaurant visits. Although we all know what kind of standard expectations there are, we know what information is shared across a culture about restaurants; we also know that sometimes what we expect to happen next comes entirely from our own personal experiences. We get reminded of past experiences by current ones, and we use those past experiences as a kind of guide to help us process new experiences.

The reason we get reminded while processing something new is to help us by providing the most relevant knowledge that we have in memory. If knowledge of restaurants in general is useful when you enter a restaurant, then it follows that knowledge of Taillevant is useful when you enter Taillevant and that knowledge about prior circumstances where you have taken a date to a fancy French restaurant in order to impress her is especially useful when you are about to try the same thing again.

Beliefs and ideas

Reminding is very useful for planning and therefore for understanding the plans of others. When someone tells you a story, however, he is talking not only about plans but often, as we have noticed, about beliefs. When what is to be understood in a story is about beliefs, the kind of guidance we need changes. We don't need to know what will happen next. When we hear such stories all we are trying to do is understand them. If we are passively viewing a movie for example, understanding the movie means being able to follow what is going on by relating what we are seeing to what we know, learning something from the movie, in a very weak sense of learning. In a conversation, understanding means being able to respond to a story. In both of these cases, then, understanding means attempting to extract indexes such that old stories can be related to new ones. For movies the intent is recognition, for conversation, the intent is to be able to respond.

When stories are about ideas rather than plans, the problem for the hearer is to respond to the ideas. But ideas are much harder to get a grip on than plans. We may all agree on the plans a murderer was following when we finish a mystery novel, but such agreement is more difficult to come by when we attempt to discuss the key ideas in a novel about people and their relationships. A murder mystery has a plot, which means an involved set of plans, so when the understanding process involves plan extraction from a text, the process is fairly straightforward and not especially idiosyncratic. But when a novel has no plot, when no clear plans are being stated and followed, finding the ideas that are being expressed becomes a problem of belief extraction. This extraction of beliefs can be especially difficult because often actors and even the writers who create fictional actors don't know what their beliefs are. Actions can express beliefs, and so as understanders, our job is to find the beliefs that are inherent and implicit in a given action.

When the understanding process gets complicated, the primary mechanism we have available to us to guide understanding, namely reminding, must work especially hard on rather scanty evidence to find something to get reminded of. The main fodder for reminding in such circumstances comes from beliefs that have been extracted from a text. Such beliefs cause our own personal stories to come to mind when those beliefs happen also to be indexes in our own memories. But then a funny thing happens -- we feel compelled to tell those stories. Why we desire so strongly to tell our own stories is something we have already discussed in part and to which we shall return later. The point here is that once we have found our own story, we basically stop processing.

The reason for stopping is partially based upon our intentions in the first place. Since most of the time we were really just looking for something to say back in response, having found something, we have little reason to process further. But more important, what we have found usually relates to an arguable point, an idea subject to challenge, a belief about which we are uncertain. As understanders, one of our goals is to gather evidence about the world so that we can formulate better beliefs, ones that will equip us better to deal with the real world. Once we have found a match between someone else's experience and our own, we are excited to begin thinking about the connections so that we can add or subtract beliefs from our own personal data base.

The paradox of understanding

There is an odd side effect to all this. We are not likely to directly learn from other people's stories. In getting reminded of our own stories, ones which of course have more poignancy and more rich detail than the ones we are hearing, we tend to get distracted into thinking more about what happened to us. The incoming story can get recalled in terms of the story of which we were reminded, but in the end, we rarely recall the stories of others easily. More often than not, other people's stories don't have the richness of detail and emotional impact that allows them to be stored in multiple ways in our memories. They do, on the other hand, provide enough details and emotions to allow them to be more easily stored than if the teller had simply told us his belief.

So we are left with a paradoxical picture of understanding. We do not easily remember what other people have said if they do not tell us in the form of a story. We can learn from the stories of others, but only if what we hear relates strongly to something we already knew, and causes us to rethink our own stories. We hear, in the stories of others, what we personally can relate to, by virtue of having heard or experienced, in some way, that story before. Understanding is an idiosyncratic affair. Our idiosyncrasies come from our stories.

IV. BEHAVIOR IN RESPONSE TO NEW EXPERIENCES

The idea that new experiences get interpreted by adapting old stories is quite closely related to the psychological concept of stimulus generalization. The original form of the idea, in Pavlovian conditioning, had to do with the decline in strength of a conditioned response as the conditioned stimulus was shifted along a continuum of stimulus similarity away from its original value.

In this and other simple, well-defined situations, the decline is said by Roger Shepard (1987) to follow a strict mathematical form, which he believes warrants being called the First Law of Psychology. As a psychophysical statement, based on quantifiable responses to quantifiable stimuli, the First Law is very powerful. Unfortunately, it doesn't help us much when we ask how interpretations of stories change when they are altered away from prototypical stories. In this context, the idea of similarity is vague, but nevertheless compelling. In addressing the nature of similarity of stories, we will cite a couple of lines of research, and also tell a story.

Interpreting What You Don't Understand

Chance situations with "skill" indexes

In her seminal work on the "illusion of control", Ellen Langer (1975) proposed that people have great difficulty interpreting chance processes such as lottery drawings and roulette wheel spins. The lumpy disorder of chance processes is hard for the ordinary individual to appreciate, much less apply to a single chance event. Accordingly, success and failure at gambling on strictly chance events is typically stored in memory as a set of experiences (such as, "the time I hit eight in a row at the roulette table at the Taj Mahal"), rather than as summary statistics (such as, "in roulette, betting on red or black, I won 203 out of 400 bets, but by a chi-square test, this is not significantly different from the expected number, 191 of 400").

Langer goes on to hypothesize that unintelligible abstract chance situations are mentally transformed into skill situations with which the individual is familiar. There are many features of chance situations that correspond, or can be made to correspond, with features of skill situations. The list of such features would include facing a competitor, having choices, feeling involved, and being familiar with the paraphernalia of the game. These may be thought of as low-level indexes to clusters of past experiences. The more skill-related features a chance situation has, the more it will be understood in terms of skilled activity. A consequence of this misinterpretation is that people come to believe they can to some extent be skilled, or at least systematically lucky, in lottery drawings and other totally random events.

In testing this hypothesis, Langer ran a series of experiments following a standard design. A lottery was announced to a pool of subjects. Then, in selling them the tickets, only a random half of the subjects were exposed to one of the skill-related factors. For example, for the "choice" factor, the lottery tickets consisted of bubble gum cards of athletes. The experimental subjects were allowed to choose which athlete they wanted to represent them in the lottery drawing, whereas control subjects were assigned an athlete. Later, under a pretext, all subjects were invited to sell their tickets to one of a group of newcomers who belatedly wanted to enter. For a $1 ticket, the average asking price was above $8 for the choice group, but under $2 for the control group. In other words, chosen tickets were much more highly valued than assigned tickets, despite their objectively equal chances to win the lottery.

In another paradigm, subjects were pitted in a simple card selection game against either a nattily-dressed, confident opponent, or a sloppy, indecisive, nerdy competitor. On each trial, the players selected a card blindly from a deck of playing cards, and the player with the higher card won or lost from the experimenter the amount of money he had been willing to bet on that trial (amounts ranging from 5 to 25 cents). Subjects pitted against the confident player risked far less money than those facing the nerd, despite the objective fact that confidence has nothing to do with the proclivity for drawing high cards.

Langer's experiments well illustrate the phenomenon of interpreting poorly understood situations in terms of similar, familiar ones. This research further demonstrates the usefulness of conceptualizing the similarity of new and old experiences by means of "common elements" (Tversky, 1977 )-- in our terms, shared indexes.

A later experiment by Ayeroff and Abelson (1976) applied the same reasoning to an even more unfamiliar experience: a mental telepathy experiment. Subjects were brought into separated rooms and told that they would take turns "sending" a series of impressions of simple objects depicted on a deck of cards, as a test of a controversial aspect of extra-sensory perception.

These subjects faced an extraordinarily cryptic experience. Not only were they confronted with the trial-by-trial vicissitudes of chance, but almost none of them had ever before participated in a systematic, "scientific" mental telepathy test, and none of them knew exactly how to send or receive telepathic symbols.

Now, what type of familiar experience is similar to this unfamiliar experience? Ayeroff and Abelson asserted that the task is reminiscent of the interpersonal experience of having "good vibes" with someone -- feeling that communication is effortless, that two spirits are one. Accordingly, for half the subject pairs, an experimental manipulation was introduced to produce good vibes between the partners while preparing to try telepathy. These pairs were given five warm-up trials in which they could practice their telepathizing while conversing into open microphones. A typical conversation would go something like this:

Sender: I've got it focused now. What are you getting?
Receiver: A top hat?
Sender: That was my second choice. This one is bumblebee.
Receiver: Oh, yeah. Focus it a bit more.
Sender: Okay...(Pause)...Now I'm beaming it like a laser.
Receiver:...(Pause)...Oh, wow! I've got a buzzing bee! Great!

Pairs in the control condition also had five warm-up trials, but instead of talking, they practiced with the microphones off, and wrote down what they were thinking. This condition was designed to restrain vibes.

On the 100 real trials following the warm-up, the subjects' belief in their success was measured by having each member of the pair indicate to the experimenter whether he thought they had scored a hit on that trial, or not. Since there were five equally numerous symbols in the deck, the chance level of success was 20%. (The actual success rate was 19.6%.) On the average, subjects in the good vibes condition claimed hits on 54% of the trials, compared to the no-vibes subjects' mean estimate of 37%. The authors' interpretation of this result was that the unfamiliar experience of attempted telepathy was interpreted in terms of the very familiar experience of two people attempting to understand one other in a social interaction. When the similarity to successful social episodes was increased by providing warm-up trials with good vibes, the participants' subjective estimates of telepathic success increased significantly.

Thus far, we have reported research on the (mis)interpretation of unfamiliar situations by analogy with familiar situations. We advanced the proposition that the more aspects the new situation has in common with the prototypic familiar one, the greater the reliance on the latter.

Let us now consider what specific common aspects (indexes) might be rather compelling, that is, likely to dominate the choice of the particular familiar experience used in interpreting the cryptic one. (Here is where we tell a story.)

The misunderstood Peace Corps volunteers

In the early 1960s, an erstwhile Harvard cognitive anthropologist named Volney Stefflre spent some time at Yale working on various eccentric research projects.

His (unpublished) magnum opus was a study of the reactions of Indian natives in remote villages in Chile to the arrival of handfuls of Peace Corps volunteers. These volunteers, fresh from a month of insufficient training, were thrown into strange territory with no clear conception of how to use their limited engineering, public health, or organizational skills to help somebody down there. (Anybody!) And their Regional Coordinator had long since gone native, and could not be located.

Nevertheless, they were earnest and undaunted, and kept recommending projects to the Indians. These natives, for their part, were puzzled. They had never met any visiting people who behaved quite like this. Why had they come, offering help and advice in such a polite manner? What was the story?

As befits this essay, Stefflre's hypothesis was that people in unfamiliar situations behave as they would in the familiar situation most similar to the new one. Prior to the arrival of the Peace Corps, the natives had been visited by soldiers, teachers, and government officials, among others. But none of these groups behaved in the eager, earnest, talkative manner of the Peace Corps volunteers. The natives observed for a while, and finally decided that these new people were ministers. Thereafter they acted with appropriate reverence but paid no attention whatsoever to advice on medical care, birth control, sewage treatment, etc. After all, what would ministers know about medicine, sex, and digging ditches?

We see in this story a tension between different bases for experiential similarity. On the one hand, the personal manner of the unfamiliar volunteers was a very good match to the style of known ministers. On the other hand, the content of the volunteers' apparent interests did not fit with that of the ministers. It is to be expected that in some ways, the new experience would match the old, and in other ways, not. If an old experience matched a new one perfectly, then the new experience would not be unfamiliar.

A mix of similar and dissimilar features is found also in metaphor (e.g., Gentner 1983; Ortony 1979). Thus one might be tempted to say that remindings are metaphorical. Whether it helps in constructing a theory of reminding to say this, we are not sure. The key question is what aspects of experience (which indexes) are dominant in matching a reminding to a new situation.

The characteristics of the story actor

We have a small hint in the Peace Corps example that matching the personal characteristics of the actors in the new story to actors in the old one may be an especially strong principle of reminding, and thus understanding. A study by Lamb, Lalljee, and Abelson (1992) makes the same point. Subjects in their study were given thumbnail sketches of four men, each a different prototypical criminal: a purse-snatcher, an embezzler, a terrorist, and a bank robber. Then stories were presented in which each of these men was involved in an ambiguous action sequence which might or might not be interpreted as a particular crime. For example, one character (the embezzler) was previously described as a smooth-talking executive who had been involved in a celebrated stock swindle, and was now in desperate need of money to cover gambling debts. This character then appears in a story as a man sitting on an airplane aisle, with an odd-looking package under his seat, and an elderly woman with heart problems sitting next to him. He calls the stewardess to give her a note he has just written. She looks troubled, and hurriedly walks front to the pilots' cabin. The task for the subjects is to interpret what is going on. Is it a hijacking, or a medical emergency concerning the elderly woman?

In other combinations presented to different groups of subjects, a prototypical terrorist (interested in political causes, loyal to a group in the Middle East, etc.) appears in the airplane story, and the prototypical embezzler appears in a story about a company Vice President who often works late at night at a computer terminal.

There was an almost unanimous tendency for the subjects to infer the presence of a crime when the actor was prototypic for it, but much less of a tendency when the actor was nonprototypic (e.g., the terrorist was judged to be hijacking the plane, but the embezzler usually wasn't; he was getting aid for his sick mother).

To carry further this notion of story interpretations driven by the perceived personality characteristics of the actor, such an effect seems of a piece with what Ross (1977) calls the fundamental attribution error. When we perceive an actor behaving a certain way in a given situation, we tend to perceive the cause of the behavior as something about the individual, rather than something about the situation. An alternative label for this effect might be personality-driven reminding. In such a reminding, the personality characteristics of the main actor in the new experience serve as useful indexes for previous experiences with that type of person. This possibility is not always relevant (the lottery example, for instance, has no main actor), but we suggest that when available, personality-driven remindings tend to outcompete remindings driven by other features of the situation.

The Story Model of juror decision making

A vein of relevant research in this latter category arises from Pennington and Hastie's (1992) Story Model of juror decision making. They maintain that jurors' natural inclination in making decisions in criminal cases is to organize the evidence into a story structure with initiating events, goals, actions, and consequences. In several mock jury studies, factors facilitating story structures were found to lead to more confident verdicts in the direction of the bulk of the evidence. Interestingly, although subjects in story-inducing conditions organized their stories well, their recall for facts in the case was no better than that of other subjects. This result is consistent with much of the research on "transmission set". However, the orientation of jurors toward the information about a crime is a mixture of a transmission set--they know they will have to share their judgments with other jurors--and a receiving set--they know they should wait for all of the evidence.

Salience

Social psychologists have often relied on the concept of the salience of different features of a situation. In unfamiliar or unstructured settings, there are different interpretations the individual can come up with. What the person is most compellingly reminded of will, we have proposed, guide his subsequent behavior. But the reminding process is delicately triggered -- an individual placed a second time in the same new situation (with amnesia for the first) might not be reminded of the same prior experience. In part, the salience of one reminding rather than another is due to situational influences: the individual's interpretation of what is going on can be manipulated by hints and symbols. A blatant literary example of this is the influence of Iago over Othello. Iago drops innuendoes of Desdemona's unfaithfulness each time an ambiguous behavior of hers is seen or heard of by Othello, until at last the tragic Moor accepts this totally false interpretation.

Experimental manipulation of salience

Salience has been systematically manipulated in psychological research on persuasion, impression formation, and other topics. Price (1989), for example, showed that members of a group can be led to the interpretation that their interactions with another group are conflictful, by making salient some issue between them. "Humanities and science majors clash over curriculum change," says the bogus headline, and presto -- students who happen to be humanities majors start resisting persuasive communications from science majors on other issues. Such demonstrations of the effects of reminding [sic] people of their group identities go way back to the 1950s, when Kelley (1953) showed that Boy Scouts sitting at a Boy Scout meeting in full uniform were much more resistant to a speaker who made fun of hiking and camping activities than were Boy Scouts at an undesignated meeting who were exposed to the same speaker.

Availability

Modern research workers in the field of social cognition usually use the terms availability and activation to distinguish two sources of salience. Availability (Tversky & Kahneman, 1973) refers to the ease with which a given idea, judgment, or evaluation comes to mind. For the present paper, we would translate this into "the ease with which a given experience comes to mind". Availability arises from repetition of an experience. Thus scripts in particular have high availability -- they tend to be triggered as first-choice interpretations of any situation in which script-related cues are present. Also, strong attitudes have high availability (Powell & Fazio, 1984), which increases with repetitive exercise, especially via the expression of those attitudes (Downing, Judd and Brauer, 1992). This implies, for example, that the bigotry of the bigot increases as his expressions of intolerance increase. In the context of story interpretation, telling someone a story about your experiences might be expected to increase the availability of that story on future occasions.

Activation

In contrast to availability, which signifies the internal priorities of different experiences, activation refers to external factors affecting the choice of interpretation. Iago might be said to have activated Othello's jealousy story. The Boy Scout and humanities majors studies also merit the label of activation. In the former case, the presence of peers in uniform presumably activated Boy Scout camping experiences. If these had been fun, this would undercut the speaker's trash talk about camping.

A currently very popular class of experimental methods for producing activation is "priming". This refers to the incidental or sometimes subliminal introduction of an idea in order to activate extensions or related ideas. As cognitive psychologists and social cognitionists will know, the priming concept was introduced by Meyer and Schvanevelt (1971) in a study of "semantic priming". These investigators presented subjects with short strings of letters on a succession of trials, and the subjects' task was to decide as rapidly as possible whether each string was a word or a non-word. The experimenters had the hypothesis that words would prime (activate) the perception of similar words that immediately followed, reducing reaction times to the second word. Sure enough, as illustrated by their now standard example, reaction time to NURSE was faster when preceded by DOCTOR than when preceded by an unrelated word such as LAWYER. Other examples with the same structure also produced the same effect. Activation effects have since been obtained in many areas, the most popular of which appears to be the priming of expectations about the traits of an unfamiliar person (Higgins, Rhodes, & Jones, 1977).

Priming also occurs with sentences, and even paragraphs as stimuli. Sharkey and Mitchell (1985) showed that scripts could be primed with introductory sentences such as, "The children all brought their presents to Mary's house." Such introductions speeded the lexical decision response to script-related words presented immediately afterwards (e.g., subjects were faster to recognize CANDLES as a word when it followed the introduction of the party script, as opposed to a non-party introduction). Seifert et al. (1980) demonstrated a priming effect of proverb-related stories on other stories illustrating the same proverb. Indeed, priming can be conceptualized as the (usually implicit) activation of particular story indexes.

Psychological theorizing and research is amply supportive of the idea that new stories are understood in terms of similar old stories. Search for old experiences from memory requires some kind of indexing scheme for picking relevant stories, and we have suggested three conditions affecting the implicit choice of indexes: 1) When applicable, indexing in terms of the personality characteristics of the main actor in the new story tends to be preferred; 2) When aspects intrinsic to a particular type of old experience are added to the scenario of a new experience, search tends to be influenced by these common aspects; and 3) In any experience, situational factors or the actions of other people can affect the interpretation of that experience by selective activation or priming.

V. STORYTELLING AND MEMORY

Stories, as we have noted, are the basis of our understanding. Understanding means retrieving stories, and applying them to new experiences. The consequence of this is profound for models of memory. If active memory is really a beehive of activity involving story retrieval and story application, then what it means to remember needs to be reinterpreted.

Researchers have often viewed memory as a kind of warehouse, a place where events are placed and then retrieved when needed. The problem with this point of view is that it assumes that events are discrete items, made up of discrete parts. Thus we assume that when something happens to you, it is labeled as event497, with parts 1-27. A view of memory of this kind essentially buries two key aspects of memory. First, memories for events are indexed by our understanding of the events themselves. Our trip to Maine is not labeled event456 or even "trip to Maine, 1975." Rather, the event would be indexed by a characterization of the events that it comprises. This means that "the time I almost drowned" or "the best lobster I ever tasted" might be two of the many indexes such an event might have in memory. Second, and most important, we must consider how such indexes might have been constructed in the first place, since in their construction in memory and by memory, they have a serious effect on the stories we subsequently tell.

Constructing the Story of an Experience

When we construct an index we do so not by saying "Well, what would be a good name for this event that I wish to store away in memory?" People are not so conscious of their internal processes as to be able to ask such a question. Rather, we proceed through life having experiences and in no way attempting to store them away. We are not trying to remember, at least not consciously. But, of course, we are remembering. How?

Telling is remembering

The answer to this is: by telling stories. Storytelling is not something we just happen to do, it is something we virtually have to do, if we want to remember anything at all. No one who has recalled a lobster dinner as "the best one I ever had" could have possibly have done so without consciously thinking that thought. And, most likely, that person would have expressed that thought as well, to a dinner companion perhaps, or to someone who asked about his trip to Maine. While it seems less likely that it would be absolutely necessary to tell someone about a near drowning in order to remember it, people nevertheless do seem to feel compelled to tell about such salient experiences. It is in the story telling process that the memory gets formed.

This happens in the following way. We form an opinion, a viewpoint, or indeed, a good story about what happened to us. We retrieve other details of the story, such as where we stayed, and who we were with in that context. But, over time, we have difficulty remembering the surrounding details. We tend to just remember the story we have constructed. Everything else has to be re-constructed. To put this another way, the stories we create are the memories we have. Telling is remembering. Everything else, what we fail to tell, gets forgotten, although it can often be reconstructed. The effect of all this is very interesting. Not only do our memories become a function of what we talk about, but if these stories aren't sufficiently rehearsed, that is, told often enough, we begin to forget them as well. And because of this a curious thing happens.

The dangers of misremembering

In our desire to tell a story in the first place, we resort to certain standard story telling devices. Those devices are part of our cultural norms for story telling and they reflect what is considered to be a coherent story in a culture. Since, in telling one's story to others, one wants to be coherent, one has to structure one's story according to these norms. This means, in effect, that one has to lie. Nothing in life naturally occurs as a culturally coherent story. In order to construct such a story we must leave out the details that don't fit, and invent some that make things work better. This process was seen in Bartlett's (1932) work on Eskimo folk tales which were remembered by British subjects many years later as coherent stories while the original was certainly not coherent in a British context. This same process is at work when we tell our own stories. We tell what fits and leave out what does not. So, while our lives may not be coherent, our stories are. The danger here is that we may come to believe our own stories. When our stories become memories, and substitute for the actual events, this danger is quite real. We remember our stories and begin to believe them. In this way, stories shape memory profoundly.

When something important happens to you, you feel compelled to tell someone else about it. Even people who are reticent to talk about themselves can't help telling others about significant events that have just happened to them. Let's consider how this process might work.

Imagine, for example, that you have just returned from a vacation or that you meet someone who knows that you have recently been on a date that you were especially looking forward to. In either of these situations, when you are asked how it went, you can respond with a short pithy sentence or two, or you can begin to tell a story that captures the essential parts of the experience.

Changes in Stories When Retold or Not Retold

Now imagine that another person asks you substantially the same question. How different is your second story likely to be from the initial story? Of course, the time you have in which to tell the story, or differences in intimacy with the person you tell it to may affect the telling, but the likelihood is that on a gross level, the subsequent stories you tell will leave out or emphasize the same things. The stories will be, from a index point of view, substantially the same. In other words, while telling about a trip to a great restaurant, if you don't tell about the lovely park you visited beforehand, the park episode will eventually cease to be part of the story.

The process of story creation, of condensing an experience into a story-size chunk that can be told in a reasonable amount of time, is a process that makes the chunks smaller and smaller. Subsequent iterations of the same story tend to get smaller in the re-telling as more details are forgotten. Of course, they occasionally get larger when fictional details are added. (The old psychological terms for these two alternatives are "leveling" and "sharpening" (Allport and Postman, 1945).) Normally, after much re-telling, we are left with exactly the details of the story that we have chosen to remember. In short, story creation is a memory process. As we tell a story, we are formulating the index to the experience which we can use to create a story describing that experience.

Losing access to details

If we don't tell the story soon enough or often enough after the experience, or if we don't tell the story at all, the experience cannot be coalesced or indexed since its component pieces begin to mix with new information that continues to come in. We cannot remember a great restaurant if we keep eating in ones quite like it day after day.

Losing generalizations

In other words, while parts of the experience may be remembered in terms of the low-level memory structures that were activated--a restaurant may be recalled through cues having to do with food, or with a place, or with the particular company--the story itself does not exist as an entity in memory. Thus, without telling a story, any generalization that might pertain to the whole of the experience would get lost. We could remember the restaurant, but we might forget that the entire trip had been a bad idea. We might be able to reconstruct generalizations about the trip as a whole, but this process would require doing exactly what one would have had to do in the first place. That is, reconstruction with an eye towards generalizations creates indexes as stories. In other words, we tell stories in order to remember them.

Motivated forgetting

The opposite side of the coin is also true. We fail to create stories in order to forget them. When something unpleasant happens to us, we often say "I'd rather not talk about it" because not talking makes it easier to forget. Once you tell what happened to you, you will be less able to forget the parts of the story that you told. In some sense, telling a story makes it happen again. If the story is not created in the first place, however, it will only exist in its original form, i.e., in a form distributed among the mental structures used in the initial processing. Thus, in the sense that it can be reconstructed, the experience remains. When the experience was a bad one, that sense of being in memory can have annoying psychological consequences. If we encounter a particular setting or prop, unhappy remindings may well occur when not expected.

Adding embellishments

When you begin to tell a story again that you have retold many times, what you retrieve from memory is the index to the story itself. That index can be embellished in a variety of ways. Over time, even the embellishments become standardized. An old man's story that he has told hundreds of times shows little variation, and what variation exists becomes part of the story itself regardless of its origin. People add details to their stories that may or may not have occurred. Why should they be able to remember? They are recalling indexes and reconstructing details. If, at some point they add a nice detail, not really certain of its validity, telling that story with that same detail a few more times will ensure its permanent place in the story index. In other words, the stories we tell time and again are identical to the memory we have of the events that the story relates. Stories change over time because of the process of telling, because of the embellishments added by the teller. The actual events that gave rise to the story in the first place have long since been forgotten.

Memory for Daily Events

The man with the special day

Let's imagine a day in the life of a man living alone in a city. He works by himself and for himself. He sees and talks to no one about his particular experiences during the course of one day. He gets a haircut. He buys some groceries. He shops for new shoes. He fills out tax forms at home and watches some television. The next day he resumes a more normal life, interacting with people and talking about his experiences, but for some reason he never speaks to anyone about the day we have just described. Now, the question is, what can he remember about that day?

The answer to this is complex because we don't know two things. First, how unusual is this day for him? And, second, how much rehearsal has occurred? Let us explain why each of these questions matters.

What makes an event memorable is both its uniqueness and its significance to you personally. For example, we easily remember the first time we do anything of significance. So, if this man has never spent a day alone, or if he was deliberately trying out such a life style, or he had been designated "King for a Day? he would probably remember the day. Or would he?

At first glance, it seems probable that he would remember such a unique or significant day; therefore, how easily can we imagine the man never telling anyone about it? If people are incapable of not telling others about significant events, then this man, too, feeling that the day was important, would be likely to mention it to someone.

This brings up the question of rehearsal. One phenomenon of memory is that people talk to themselves, not necessarily aloud of course, but they do tell themselves stories, collecting disparate events into coherent wholes. So, let's imagine that while this man talked to no one about his day, he did talk to himself. What might he have said? If rehearsal entails storytelling, he would have had to compose a story with some pertinent generalizations or observations drawn from the day. Moreover, he would have had to keep re-telling himself that story in order to remember it.

What happens if he fails to tell anyone, including himself, about his day? Does he fail to recognize the grocery store where he shopped when he sees it again next week? Does he fill out his tax forms again? Of course not.

Obviously, we can remember events that we have not discussed with anyone. But how? How are events like going to the grocery store remembered? Certainly such events never become stories, so they are not maintained in memory by repeated telling. How, then, are they maintained?

Many psychologists have claimed that memory for facts must be organized hierarchically in a semantic memory. Others have argued for a memory which is more episodically based, and still others have suggested combining the two. A neatly organized hierarchy of semantic concepts is easy to imagine, but the world is full of oddities and idiosyncratic events that fail to fit neatly into a preestablished hierarchy. For example, we may "know" from semantic memory that female horses have teats, but we may more readily access this fact from an episodic store if we witnessed our pet horse giving birth and then suckling its young. Our first memories of playing ball may very well come to mind when the word "ball" comes up, and the properties we ascribe to "ball" may well be ones that a particular ball we remember actually had. In short, the semantic-episodic distinction does not seem to be sharp and clear.

Story-based vs. event-based memory

A more useful, process-based distinction can be drawn between story-based memory on the one hand and a generalized event-based memory on the other. To understand this distinction, let's go back to the question of where our hero's grocery store, tax form filling, and reading experiences are stored in his memory.

We know that he can recall what he did in each instance, so how is this ability to recall different than his ability to tell a story? Probably, he cannot tell the story of his day, while he can recall certain aspects of his day. This difference reflects itself in a kind of abstract idea of "place" in memory. To "recognize" the grocery store visit means he knows he has been there. Had something interesting happened there, especially an event that taught him something new about the operation of grocery stores, for example, we can feel sure that he would remember it. But, how can we make this assertion when he probably won't be able to recall this day if he never talks about it to anyone? We seem to have paradox here, but, in fact, we do not.

When someone has an experience in a grocery store, they update their knowledge of grocery stores. This is how grocery store scripts grow and change over time, in response to actual changes in grocery stores that occur over time and as a way of organizing pointers to personal odd experiences that have taken place in grocery stores. When shelf arrangements change in a particular grocery store, the patron's memory must change as well. Sometimes it takes a few trials--a patron might keep looking for milk in its old location for some time after a change in the placement of the dairy section, but eventually changes in memory follow changes in reality. So, people learn from their experiences, but where does this learning take place in the mind?

People need a file of information about grocery stores that includes specific information about where their favorite grocery store keeps the milk and what it wants from you in order to cash a check. This file must also include general information about grocery stores apart from your favorite, however. When we enter a new grocery store, we want to be able to utilize expectations about our favorite store that will help us in the new one. For example, we want to predict that the milk will be near the cottage cheese in any grocery store and that a new one might not take our check. In other words, we are constantly drawing upon our file of knowledge about grocery stores and adding to that knowledge when new experiences teach us something worth retaining.

What we are decidedly not doing, however, is updating our memories on what we might call a daily unit basis. That is, we are not making note that on October 16, we bought a quart of milk and six oranges. We could try to do this, of course, but we would have to try very hard. We might make up a poem about what we bought on that day and then memorize the poem, for example. But if we do not take some extreme measure like that, we will simply fail to remember the experience unless something rather strange or important occurred at the same time. Why can't we remember what we bought in the grocery store on October 16, 1982?

One obvious answer is that it would be absurd to remember such a thing. Human memory must be selective to function well. One aspect of this selectivity in memory is the recognition of the distinction between events that are to be added to one's internal storehouse of generalized events and those which need to be summarized and indexed and told as new stories to be added to story-based memory. Memory is looking for knowledge that tells it something about the nature of the world in general. This storehouse of knowledge is on its face analogous to the notions of semantic memory. But, of course since it is based in actual experience, it is really quite episodic. Although the notion that semantic memory should be devoted to such general knowledge seems inherently correct, the notion seems equally wrong that such knowledge would not have at its core a seriously idiosyncratic component. We may all know that a flounder is a fish and that a fish has gills, but we do not all know that our father used to eat flounder every Tuesday, and therefore, so did we, and we refuse to eat it ever again. Yet, this latter fact is just as much a part of the definition of flounder for us as is the fact that a flounder has gills--maybe more, since one fact is far more real to us than the other.

Building up event memory

Any general storehouse of knowledge, then, is likely to depend very strongly upon the expectations about various objects and situations that have been gathered over a lifetime of experience. Thus, when a new experience occurs that speaks to what we already know about something, perhaps updating it, perhaps overriding it, we add that experience to our memories. This is why we remember filling out the tax form. We add the experience of filling out the tax form to our general storehouse of similar experiences. That experience then becomes part of our general knowledge of tax-form filling and updates what we already know.

Similarly, when we read something, the facts we garner from our reading go to particular places in memory, to the structures that we have that are repositories of information about those subjects. Information about restaurants updates what we know about restaurants, stories about travel to exotic places causes our memory to add new information about those places to existing knowledge about those particular places and to general information we may have about exotic places. Of course, the actual updating of knowledge structures is much more complex than this. In Dynamic Memory (Schank, 1982) it was proposed that pieces of memory structures, once altered, update those same pieces as instantiated in other structures. Thus, for example, if you learn something about paying with credit cards in a restaurant, that new information needs to update how to pay in a department store and at an airport as well. The way this happens is through sharing of standardized smaller knowledge structures of which "paying" would be one, and "credit card paying" would another even smaller structure.

Through such structures, and through the sharing of smaller structures by larger structures, we build up event memory. Every time we use a particular body of knowledge in our interactions with the world, that knowledge gets altered by the experience. We cannot fill out a tax form without using the prior experiences we had in filling out tax forms as a guide to help us through the experience. But because that knowledge is being used as a guide, it changes. We add new information about tax forms, about the experience of filling them out, that overrides what we previously knew. When we are finished doing anything, therefore, our memories are altered by the experience. We don't know what we knew before.

The process of updating our event-based memory every time we have a relevant new experience has an odd side effect, however. The construction of a memory that organizes information around repetitions of events events destroys the coherence of any particular sequence of events. The dynamic nature of event memory causes the experiences of walking to be placed with prior ones of walking, those of shopping to be placed together with others of shopping, and so on. Constant updating of a memory for events causes a general storehouse about typical events to be built up by destroying the connectivity of one particular event to another particular event. A particular event of walking, therefore, becomes disconnected from its intended purpose of enabling one to go shopping at one particular time, for example, thus rendering our hero useless when asked how he got to the grocery store. His only recourse is to make an educated guess: I must have walked; it's not far, and I usually walk if it's a nice day, and it was June after all.

The need for story-based memory

Because of this need of memory to effect a constant disconnection of events from those that follow, we feel a need to undo this process when something of significance occurs. We can stop the dynamic disconnection from taking place and remember events in sequence by consciously giving our memories an event to remember that is a unit, specifically, a unit that we have rehearsed, sometimes frequently. In this process, the role of stories in memory comes into play, and motivates the concept of story-based memory that is the core of this essay. Stories are a way of preserving the connectivity of events that would otherwise be disassociated over time.

For stories to be told without a great deal of effort, they must be stored away in a fashion that enables them to be accessed as a unit. If this were not the case, stories would have to be reconstructed each time they were told, a process that would become more and more difficult with time as the connections between events fade from memory. Telling a story would require a great deal of work to collect all the events from memory and to reconstruct their interrelation. Further, stories would be quite different each time they were told. Reconstruction would not be the same each time, and instead, different stories would result depending upon what parts of memory were looked at during the time of telling. This kind of story-telling does occur, of course, especially when stories are being told for the first time, but most storytelling requires so little work and is so repetitive, each version so much like the other, that many stories must be stored and retrieved as chunks.

A different type of memory process, then, must be active here. Our hero who fails to tell one or more stories from his isolated day will understand and remember what has happened to him in the sense that the facts will be available to him. But they will be available to him only when the various segments of his day are accessed for some reason, when someone asks him about his favorite grocery store for example. What he will lose is the ability to tell a story about that particular day. The day will disappear as a unit from memory, as will various aspects of the day. In other words, events of the day will no longer be accessible by asking oneself about what happened today; after a while they can only be found in the other parts of memory which will have subsumed them. What is remembered, then, will be in terms of what one knows about grocery stores, not a story in and of itself arising from the events of that day. To find that kind of story in memory, one must have put it there consciously in the first place, either by telling it to somebody or to oneself.

Story-based memory, then, is a different kind of memory from the memory that contains general event knowledge. Story-based knowledge expresses our points of view and philosophy of life. It depends upon telling and gets built up by telling. The consequences of this process are interesting when one considers what we tell and why since we are, quite unconsciously, making decisions about what to remember.

VI. THE SOCIAL CONTEXT OF STORY TELLING

Human beings have large collections of stories. They accumulate stories over a lifetime, and when they are given the opportunity, they select an appropriate story and tell it. They determine appropriateness by a variety of factors, such as familiarity, emotion, the potential for shared viewpoint, and the need for approval. As we have seen, the story indexes can be selected for a variety of different purposes. The stories we tell are strongly affected by whom we are telling them to. In any situation where we finds ourselves telling an old story, we might reasonably wonder why we have chosen to tell a particular story.

The Teller and the Listener

The influence of the listener

One must decide on the appropriateness of a given story, and one usually seeks the approval of the listener either to elaborate or to tell another story. The listener, then, performs a very important role for a storyteller. He reveals, usually implicitly, which stories he wants to hear. He may like ones that show how important or powerful you are, or he may think such stories are exaggerations or simply ought not be told. He may want to learn some specific thing from your stories, or he may simply want you to finish up so he can tell his story in response.

Consequences for the teller

The trick for any listener is to send out the right signals, those that encourage the telling of the stories that the listener wants to hear. In the selection and evaluation process, eliciting the listener's approval is very important. We want to please our listener, but pleasing is a fairly complicated idea. If we know that we are our listener's hero, we might tell stories about our successful exploits in the world. If we know