It happens very often that we see, hear, and otherwise experience things happening around us, that we later remember these things, and that for one reason or another we are motivated to use language to communicate what we remember to other people. In other words, we often talk about recalled experience. In so doing we convert into verbal form something that may have been partially or even wholly nonverbal to begin with. How do we do this? How are we able to take nonverbal experience, store it away for a while, and subsequently recall it and turn it into words that convey to other people something of what the experience was like? If I pretended to have an answer to this question, I would be pretending to know the solution to problems that have plagued investigators of the human mind for thousands of years. My intention is much more modest: to try to make it clear that the question is an important one for linguists to consider, and that we really cannot expect to understand much more about language without committing ourselves wholeheartedly to an attack on the vexing problems that the question raises.
It is necessary to have some kind of model, however imperfect, on which our discussion can be based. Fig. 1, which ignores a number of factors I will not bring up here, is intended as such a simplified model. On the left in this diagram is a box labeled ‘stimulus’, which is meant to represent the physical input to our sense organs when we first experienced the incident we later tell about. There is an arrow leading from this box to the one labeled ‘consciousness’. This second box is meant to represent our phenomenological awareness of the experience in question, or, if you prefer, the presence of some information related to the experience in our short-term memory. The arrow is labeled perception’, and my intention with this label is to emphasize that what enters consciousness is not a faithful replica of the stimulus (whatever that might be), but rather an interpretation of it. To take a simple and familiar example, when we look at something like Fig. 2 we are likely to see, not a collection of horizontal, vertical, and diagonal lines—which in a sense is closer to what the physical stimulus contains—but a transparent cube. With some effort, too, we can interpret the cube in two different perspectives, with either the face in the lower left or that in the upper right being closest to us. These are different interpretations our conscious minds are able to give to one and the same physical stimulus. What is in consciousness constitutes a kind of belief about what is in the outside world. This belief may be derived from what is presented to the senses, but it is heavily influenced by contextual expectations as well as cultural and individual predispositions. All these things combine to produce in our minds some kind of interpretation of what is going on outside. The key notion for us here will be that perception is interpretive. (See Hochberg, 1964, for a useful introductory survey.)
Consciousness evidently has a very limited capacity (cf. Miller, 1956, for a suggestion of how much we can hold in consciousness at one time). At any given moment we can be aware of only a narrow range of phenomena. But at least some of what has been in consciousness can pass by way of the arrow labeled ‘storage’ in Fig. 1 into the box labeled ‘memory’. Using different boxes to represent consciousness and memory may be misleading, since it is likely to suggest that two different areas of the brain are involved. But undoubtedly it is preferable to regard these as two different states in which information or knowledge can be at different times. I like to think of what is in consciousness as being knowledge that is activated or ‘lit up’ at any particular time, while what is in memory as inactive or not lit up at the moment, but still ‘present’ in some fashion in the mind.
Another arrow leads back from memory to consciousness to show that information can be ‘recalled’, or reactivated. There are thus two inputs into consciousness: one through perception and one through recall. When an incident is first experienced, we are of course conscious of it through perception. But when we talk about the experience later it enters our consciousness through recall. Although occasionally we may talk about something at the same time we are perceiving it, such discourse is relatively unusual; most of the time we talk about things retrieved from memory. Most experience is recounted in the past tense, which is used for information that has traversed the ‘recall’ arrow in Fig. 1.
As linguists we are presumably most interested in the arrow in Fig. 1 labeled ‘verbalization’, which is meant to include all those processes by which nonverbal information is turned into words. Some readers may question the fact that this verbalization arrow is pictured as leading out of consciousness, suggesting that while we speak we are conscious of nonverbal material. There are evidently individual differences here. Many people, as they talk about recalled experience, are conscious of mental imagery related to the experience. For them the placement of the verbalization arrow in Fig. 1 should seem correct. Others claim to be conscious only of language itself. In that case it might seem that memory contains only material that is already verbalized, and that recall consists of bringing this already formed verbalization into consciousness. Later in this chapter I will return to some considerations that may be relevant to this issue. In the first half of the chapter, however, I want to discuss what I think are some of the processes involved in verbalization.
In a gross way, verbalization processes may be assignable to three major types. First, there are processes having to do with the organization of content. Verbalization requires a speaker to structure the recalled experience itself in certain ways. Second, there are processes having to do with the speaker’s assessment of the addressee’s current state of mind and processing capacities within the particular context of the discourse. Elsewhere (Chafe, 1975) I have lumped together these processes under the term packaging’, with the idea that the speaker has to ‘wrap up’ his content in an effective way to enable the addressee to assimilate it easily. Packaging includes such considerations as distinguishing given from new information, deciding whether a noun should be treated as definite or indefinite, deciding what to make the subject or the topic, and so on. Third, every language imposes various more or less arbitrary syntactic processes that also have to be applied during verbalization. For example, one language might require that a verb agree with its subject with respect to person, number, and gender, while another language might have no such requirement.
I believe that this three-way distinction between content-related processes, packaging processes, and syntactic processes captures some real differences among the kinds of things that happen during verbalization, but in what follows I will focus only on content-related processes. These can in turn be usefully divided into three major subtypes, each of which I will discuss in some detail below. In brief they are the following. First, it is necessary for a speaker to break down larger chunks of content into smaller ones. This kind of process I will call by the somewhat unattractive name ‘subchunking’. Second, when a chunk of an optimal size has been arrived at, it is necessary for the speaker to decide on a propositional structure that will assign various roles to the objects involved in the chunk. I will speak of ‘propositionalizing’. Third, appropriate words or phrases must be found to communicate the ideas of both objects and events. Here it is useful to speak in terms of ‘categorizing’.
Having discussed subchunking, propositionalizing, and categorizing in turn, I will try to make a point that is perhaps best appreciated with reference again to Fig. 1. I mentioned initially that perceptual processes are interpretive, that what enters consciousness, and later memory, by way of perception is not a replica of the original stimulus but an interpretation of it. I believe there is strong evidence that verbalization is interpretive also. The linguistic output of verbalization processes is not a replica of the input to those processes, nor is it algorithmically determined by the input. Rather, as people speak they are constantly making choices about the best way to express what they are thinking of. The conclusion must be that what we remember is of such a nature that further interpretive processes can be applied to it. Not all interpretation takes place during perception; there is much that takes place while we are talking as well. That this is so will first become evident, I think, during the discussion of categorization below. I will then try to show that subchunking and propositionalizing may also be interpretive: that they are not always determined by what is in memory but may be decided on during the course of verbalization, and decided on differently by the same speaker on different occasions.
At several points I will make reference to a source of data that I will call ‘the film’. This is a five-minute film dealing with some activities of a young boy in a rural setting, which was shown to several groups of college students in 1972. After seeing the film they were asked to write descriptions of what was in it, and some of them were asked to do so again eight weeks later. Thus, some idea was obtained of how people verbalized their recall of an (admittedly vicarious) experience shortly after seeing it, and again after it had remained in memory for some time.
Our memories appear to contain representations of discrete chunks of experience that may be ‘large’ or ‘small’ with respect to the spatiotemporal extent of the experience as well as the amount of detail that can be recalled within it. Thus, at a very large level I can recall things that I might label ‘my childhood’ or ‘World War II’. At some intermediate level might be chunks from memory like ‘my trip to Boston last year’ or ‘the attempt to assassinate George Wallace’. But I can also recall quite small events like ‘breaking a shoelace this morning’ or a momentary facial expression of some political figure during a televised news conference. I am not referring to the importance or triviality of these events, but to their ‘size’, possibly measurable in terms of the number of sub-events into which a speaker might conceivably analyze them.
Since we are talking now in terms of a person who has recalled something from memory and is about to verbalize, it is probably best to use an example in which the initial chunk is something resembling the size of ‘my accident last month’ or ‘my visit to Boston’. For more reasons than one, it is less often that I would begin a segment of conversation by saying ‘let me tell you about World War II’ or even ‘about my childhood’. Probably there are characteristic sizes of chunks that are typically introduced into discourse for various purposes, but we know very little about that at the moment. In any case, it is often possible to verbalize this initial, global chunk by labeling it ‘my accident’, ‘my visit to Boston’, or ‘what I heard from Joe yesterday’. Such labels communicate to the hearer some very general idea of what the chunk contains, and presumably set up some framework of expectations in the hearer’s mind. But ordinarily they are promises of more to come, whereby the speaker will try to communicate some of the detail he is able to recall within this initial chunk.
Thus, an important aspect of verbalization is the process whereby larger chunks of recalled experience are broken down into smaller chunks, the verbalization of which will provide the hearer with a level of detail not available if it is only the largest chunk that is verbalized. To a considerable extent this process of ‘subchunking’ is a hierarchical one. The initial, global chunk is first broken down into chunks that are still fairly large, these into somewhat smaller chunks, and so on, until some stopping point is reached. Thus, in telling you about my trip to Boston I might start with a breakdown into (1) getting there, (2) what happened while I was there, and (3) returning home. (I am not suggesting that this is the only way or even the most common way to organize the account of such a trip, but it seems a possible way.) I might then break down ‘getting there’ into (1) going to the airport, (2) the plane trip, and (3) being met at the other end, or something of the sort. The plane trip might be broken down into (1) getting settled on the plane, (2) reading something, (3) eating, (4) watching a bad movie, (5) more reading, or whatever.
There may, of course, be digressions from the hierarchical structure. Any particular chunk, or some element within a chunk, may cause the speaker to think of a chunk that lies outside the main hierarchy, but that he nevertheless decides to verbalize at this point. For example, in describing one episode that he remembered from the film about the boy mentioned above, one person wrote:
He walked down some paths and arrived at an old, chipped, red slide. I liked this scene because it had the multicolored slide, the red fire truck, and the water tower, with a backdrop of the countryside. At any rate, the boy is at the slide. . . .
The writer’s comments on liking this scene lie outside the main hierarchy of the events he was describing. They constitute a parenthetical chunk that cannot be accommodated neatly within a coherent hierarchical tree. (It is interesting to note the use of the phrase ‘at any rate’ as a means of returning to the main hierarchy.) Digressions like that are common. In fact, there are probably significant individual differences to be found in the extent to which speakers follow a consistent hierarchy of subchunking, as opposed to piling digression upon digression. But even the ‘best organized’ verbalizers are likely from time to time to insert digressions.
When we break down a larger chunk into smaller ones, we do not do so in an arbitrary fashion, but rather we are influenced by the prior existence in our minds of certain stereotyped patterns. One particular larger chunk of experience, for example, I might interpret as an instance of a plane trip. Since I have been on a number of plane trips before, this one is not a unique experience to me. In fact, interpreting it as a plane trip means that I already have a pattern in mind: a pattern that includes some kind of breakdown into subchunks (and relations between them). I will call such a pattern a ‘schema’, making use of a term used in approximately the same way by Bartlett (1932). (Terms like ‘frame’ [Goffman, 1974; Minsky, 1974] or ‘script’ [Schank and Abelson, 1975] have also been used in similar ways recently.)
A schema, then, is a stereotyped pattern by which experience is organized, and more specifically a pattern that dictates the way in which a particular larger chunk will be broken down into smaller chunks. I think it likely that most experience is interpreted with reference to schemas. That is, there is little that happens to us that we do not interpret in terms of patterns already existent in our minds. These schemas determine how we will conceptually organize the experience, what attitudes we will have toward it, what expectations we will have concerning it, and also how we will talk about it. Since this last is our most immediate concern, it will be well to look more closely at some evidence for schemas that is provided by the use of language.
What follows is excerpted from a story told to me by Sadie Bedoka Weller of Anadarko, Oklahoma, originally in the Caddo language. Here is a translation of the first part:
It is said that once, long ago, Mr. Wildcat was digging roots in order to make a garden. Presently he heard someone talking. Mr. Turkey was standing there. Mr. Turkey said, “Well, well. You are busy. What are you doing?” “I’m digging roots to make a garden. What are you doing?” “Nothing. I’m just hanging around.” “You’ll be in my way,” said Mr. Wildcat, and he caught him, plucked him, and said, “Go over to my house where my wife is. Tell her to cook you so I can eat you for lunch.” Mr. Turkey went off and came to where Mrs. Wildcat was. She was pushing a cradle and singing. Mr. Turkey said, “Your husband over there sent me to tell you that you should make some parched corn for me. After you’ve made it I’ll go along.” She made the corn quickly and he left.
(Sad to say, in spite of his cleverness, the turkey is eventually eaten by the wildcat, but that is another part of the story.)
At a certain level of subchunking we can say that such a story is composed of ‘scenes’, each scene being something that took place at a coherent place and a coherent time, much like a scene in a play. Between scenes there is either a jump to a new location or a jump to a new segment of time, usually both. The selection above exhibits an organization into two scenes, one at the field and the other at the wildcats’ house.
Each scene appears to be organized according to the same schema. It may be noted that at the beginning of each scene (when the curtain rises) there is someone on stage doing something. In the first one it is Mr. Wildcat digging roots, in the second it is Mrs. Wildcat pushing the cradle and singing. Then there is the arrival of a visitor, in both cases the turkey, followed by a conversation between the visitor and the person already there, whom I will call the host. Then the host does something to or for the visitor, in the first scene plucking him and in the second providing him with the requested parched corn. Finally, the visitor leaves. This schema, then, consists of the following general kinds of chunks:
Arrival of visitor
Conversation between visitor and host
Action by host directed at visitor
Departure of visitor
This same schema seems to be manifested in a number of Caddo stories, where of course the specific events involved are quite diverse. It may even constitute the organizing principle for chunking at a higher level—that is, for the story as a whole rather than for individual scenes within it. For example, there is a story in which some ducks are engaged at the beginning in a game that involves throwing their eyes into a stream, diving in after them, and coming up on the opposite shore with their eyes back in place. Although this chunk is appropriate in ‘size’ (the amount of verbal elaboration) for a whole scene, it corresponds to the background activity that forms the opening segment of each of the two scenes illustrated above. Coyote then arrives, and there follows a conversation between him and the ducks. Eventually Coyote plays the game himself, but the ducks steal his eyes under the water and replace them with a yellow fruit called bone-nettle. This constitutes the action by the host directed at the visitor. Coyote then departs, having been blinded. Thus, the schema that formed the basis of several scenes in the earlier story here forms the basis for the entire tale.
Schemas are not just ways of organizing discourse; they are also ways of interpreting experience of the world around us and of organizing behavior. The Caddo, like other American Indian groups, have traditionally spent a great deal of their time visiting. The visit is a stereotyped pattern of behavior that plays an important role in everyday life. It can be seen that the schema illustrated in the folktales mentioned above is in fact the schema of a visit. While the host is doing something, the visitor arrives, there follows a conversation, and then there is an action directed at the visitor, normally the provision of food, after which the visitor departs. Thus, this schema underlies one of the commonest occurrences in Caddo life and determines the expectations associated with such an occurrence.
The process of breaking down larger chunks into smaller ones does not go on indefinitely, nor does it, by itself, lead directly to a linguistic output. The speaker begins with some large, inclusive chunk recalled from memory, and through subchunking arrives at various smaller, less-inclusive chunks. But these smaller chunks are no different in kind from the larger ones; all of them are memories of situations and events. They are still not language, but ‘ideas’. Several other kinds of processes must be applied before language is finally achieved.
The next step toward language is one that I will call ‘propositionalizing’. It is distinguished from the subchunking process in the fact that it leads to an output that is different in kind from the input. In order to appreciate this point, one needs to differentiate between (ideas of) events and situations, on the one hand, and (ideas of) objects on the other. The difference can be specified in terms of space, time, and particularity. An event (such as might be verbalized as ‘The dog knocked over the chair’), or a situation (‘The dog is under the chair’) is particular with respect to both space and time. It is restricted to a single, unique, limited portion of space, and to a single, unique, limited segment of time. (I am omitting from consideration here generic ideas, such as might be verbalized as ‘The dog is black’, which require a discussion of their own that would lead us too far afield.) An object, on the other hand, appears to be particular in space but not in time. The particular dog, the particular chair that might have been referred to in the examples above, has a continuing existence through time that the particular event or situation does not. For that reason the same object can usually take part in a large and unlimited collection of different events. This particular knocking over of the chair is presumably only one of a large number of events in which this particular dog was involved. Each event occupies a relatively small segment of particular time. The dog extends through them all, and in fact is conceived of as having a continuous existence through time during those many temporal intervals when we are unaware of what he is doing—of the particular events in which he is involved.
Many qualifications have to be added to what I have just said. For example, a particular dog or chair does occupy a particular segment of time in the sense that it comes into existence at some time and goes out of existence at another. There are few objects that do not have a particular ‘lifetime’ in this sense. But such segments of time are, for the most part, related to human experience in quite a different way than are the segments of time occupied by events. Roughly, but I think significantly, the span of time occupied by a typical event falls within the span of conscious attention. Perhaps most of the events that we deal with are possible to comprehend in their entirety within what I have elsewhere called ‘surface memory’ (Chafe, 1973), where the entire segment of time from beginning to end can be held in consciousness without being relegated to deeper levels of memory. Although this is clearly not true of all events, I believe it is usually true of events at the stage of sub-chunking we are now discussing, the stage at which propositionalizing takes place. And probably it has a lot to do with our idea of what an event is, in a prototypical sense. My suggestion is that the typical lifetime of an object is of a different order of magnitude from that of the typical segment of time occupied by an event, and that we conceive of the two as being different sorts of things for that reason. An insect whose lifetime is only a fraction of a second may in immediate conceptual terms be more like an event than an object. If we conceive of this insect as an object, that is only by analogy to the typical form of animal life whose span of existence is significantly longer.
All this discussion has been preparatory to the statement that propositionalizing consists, in part, of the factoring out from an event (or situation) of the objects involved in it. Suppose, for example, that I have a particular event in mind. One way I might verbalize my knowledge of that event would be to say Then I ate a sandwich’, having chosen to verbalize it in terms of two objects, me and the sandwich. Given the same event as a starting point, I might alternatively have verbalized it as Then I ate’, with only one subject, me, factored out, or perhaps as Then I ate some salami in a sandwich’, with the salami included as a third object. There are usually various ways in which objects can be factored from an event, and this constitutes one aspect of a speaker’s interpretation of the event.
To a degree the process here is analogous to that by which larger chunks are broken down into smaller subchunks, but there are two differences. First, the objects factored out from the event are things of a different kind from that of the event itself. Second, the event is not itself lost in the process. In the final verbalization, the event is typically represented by a verb—for example, by the word ate in the sentences given as examples above. Thus, if one thinks in terms of replacing one thing with another, in subchunking a larger chunk is replaced by a constellation of smaller ones, but in propositionalizing a chunk is replaced by a different kind of entity: a structure consisting of an event (or situation) plus, as separate elements, the objects the speaker has chosen to verbalize as participants in the event. Such a structure approximates what has traditionally been called a proposition, and we might think of representing it as, for example, ‘ate (I, sandwich)’, or in other notationally equivalent ways.
But that is not all there is to such a structure. It is important to recognize that each of the objects factored out for separate verbalization plays its own specific role in the event or situation. The major contribution of so-called case grammar has been to recognize the existence and importance of these roles. Thus, in Then I ate a sandwich’ there has been not only a factoring out from the holistic event of the ideas of me and the sandwich but also a decision to treat me as the ‘agent’ of the event and the sandwich as ‘patient’. The names we give to these roles are not at issue here; the only point is that the objects associated with an event do participate in it in different ways. The standard notation for propositions is deficient in not providing a fixed way of specifying these roles. I will not say more here about what specific roles there are, or what the best notation for them might be. I believe that a new way of sorting out case roles is needed. But of the fact that they exist and are important to verbalization there can be no doubt.
In summary, when subchunking has led to a chunk of optimum size for verbalization, the speaker must interpret it in propositional terms. Propositionalizing includes (1) the factoring out of objects from the event or situation that the chunk embraces, and (2) the assignment of roles to these objects within the event or situation. But we still have not arrived at language.
It may be possible to have a certain propositional format in mind without having decided on what specific words to use to convey either the idea of the event or the ideas of the objects that have been interpreted as participants in the event. For example, imagine that the speaker has interpreted some event as an instance of a ‘transfer’; that is, as an event in which possession of a particular object changes from one particular individual to another. At this point he need not have decided whether to convey his idea of this event with the verb ‘give’ or the verb ‘hand’ or perhaps some other verb like pass’. Furthermore, he need not have decided whether to convey his idea of the earlier possessor by calling him ‘Doug’, ‘the tall boy’, ‘your son’, ‘the boy on the left’, or whatever else might be appropriate. Thus the ultimate sentence might be ‘Doug gave the other boy a book’, ‘The tall boy handed a dictionary to Steve’, or a number of other possibilities, none of which would necessarily have been determined by the decision as to how to propositionalize the event. We can imagine any number of sentences with the same propositional structure—benefactive verb, agent, patient, beneficiary—with different words or phrases used to convey the idea of each of these elements.
Some particulars have their own names—proper names, as we call them. This is likely to be true of certain kinds of objects, especially persons and places with which we are familiar and which we are likely to have occasion to talk about repeatedly. But let us focus our attention on what happens when a speaker has in mind a particular object or event that does not have its own name. The important point on which the following discussion will be based is that in order to communicate such an idea, the speaker has to ‘categorize’ it. He must decide to interpret the object or event as an instance of some category, which in the favorable case will then provide him with a word that will more or less satisfactorily convey what he has in mind. Thus, categorizing a particular object as an instance of category X will enable me to call it ‘a dish’ or ‘the dish’. Categorizing a particular event as an instance of category Y will enable me to communicate it as ‘eat’, ‘ate’, and so on.
Perhaps the best way to approach the subject of categorization is by looking briefly at color naming, something that has been studied more thoroughly than any other kind of categorization. It is well known that the color spectrum, at least in physical, pre-interpreted terms, exhibits continuous variation along several dimensions. Most studies of color naming have concentrated on the dimensions of hue and brightness. Given a color chart in the form of a rectangle, where hue varies continuously from left to right and brightness varies in the vertical dimension, there are a few points in the chart that people are able to name very easily. It has been said that these particular colors are ‘highly codable’ (Brown and Lenneberg, 1954). For our purposes it is interesting to note that ease of naming shows up in four different ways. (Brown and Lenneberg mentioned a fifth: number of syllables in the word used.) First, there will be general agreement among different people as to what the color is called; everyone will call it ‘red’, or ‘orange’, or ‘blue’. Second, there will be consistency in naming; the same person will agree with himself on different occasions. Having called it ‘red’ at one time, he will call it ‘red’ another time. Third, he is likely to name it with a single word, like ‘red’, rather than with a longer phrase like ‘dark reddish brown’. Fourth, if he is talking rather than writing, he is likely to utter the name immediately without hesitation.
Suppose we call the points within the spectrum where there is highest codability the ‘foci’ of the various color categories. Thus, a certain point can be regarded as the focus of the category that enables a speaker to call any instance of this same color ‘red’. As we move away from these foci, codability decreases and much of the spectrum is inhabited by colors whose codability is relatively low. For colors, at least, there is no clear boundary between one category and another, but only a gradual fading away from the focus of one category or a gradual approach to the focus of another. The symptoms of low codability are of course the converse of those listed above: (1) there will be less agreement between different persons as to what the color is called, (2) the same person is more likely to call it something different on different occasions, (3) it is more likely that a phrase rather than a single word will be used, and (4) a speaker is more likely to hesitate before verbalizing. These symptoms increase with the distance from the focus of any category.
This same focal property is evidently present in categories other than colors (Rosch, 1973), and it seems clear that relative codability is a quite general phenomenon. Mention was made earlier of an event in which a boy slid down a slide. We can say that the memory of the object he slid down was highly codable, as evidenced by the fact that 99 percent of the people who verbalized this piece of knowledge used the word slide, that there was virtually 100 percent consistency when the same person verbalized his knowledge of this same object eight weeks later, and that the average number of modifiers used in all verbalizations was. 33, a relatively low figure. Since the verbalizations in this case were written, no data are available on hesitations. Ongoing work on tape-recorded narratives, however, shows signs of confirming the value of hesitations (or the lack of them) as additional evidence for codability.
We may contrast this highly codable ‘slide’ object with another object that appeared in the same film. Rather than showing a high percentage of agreement on naming, this object was called many diverse things (see Table 1). Consistency in the naming of this object was 65 percent.
That is, people who called it a particular name immediately after seeing the film used the same name eight weeks later 65 percent of the time. This and other observations suggest that, for items of low codability, consistency tends to be in the general area of somewhat more than 50 percent, while for items of high codability it is close to 100 percent. Presumably the reason that consistency is as high as it is for items of low codability is that many people remember what they called an item the first time. Earlier choices are likely to determine what is said on later occasions. The average number of modifiers used for this object was 1.39, as contrasted with. 33 for the highly codable ‘slide’ mentioned earlier.
Particular events may also be of high or low codability. The categorization of events, in fact, seems to take place in a way that is quite analogous, in these respects at least, to the categorization of objects. Thus, with the event on the slide there was 86 percent agreement on the verb slide. (Other choices included go, come, glide, float, shoot, swoosh, and take off.) Usually a two-word verb was used, where the word just mentioned specified the manner of motion and a second word the direction. In 98 percent of the cases the second word was down. The relatively high codability of this event showed up also in the consistency with which the same person used the same word eight weeks later. Consistency was 83 percent for the word expressing manner of motion, and 91 percent for the word expressing direction.
We may contrast another event in the same film, the one involving the activity of the boy on an old abandoned fire truck. What he did, roughly, was to get on the truck in the rear and move across it to the front. The verbs used are listed in Table 2. Usually a two-word verb was used here also, the second element expressing the direction of motion, but with this second element also there was much less agreement (see Table 3). This time, consistency in expressing the manner of motion was 53 percent after eight weeks, and for the direction of motion it was 50 percent.
What is going on here can be visualized in terms of Figs. 3 and 4. The point labeled X in Fig. 3 is meant to represent the speaker’s knowledge of a particular object, insofar as that knowledge is a unitary thing—a single, discrete piece of knowledge. The small circle above, to which X is connected, is meant to represent what the speaker knows about X: its various properties, perhaps, but perhaps also an analogic kind of knowledge that cannot be accounted for in terms of propositions alone, and that might, for many speakers at least, appear in the speaker’s consciousness as a mental image of X.
Let us suppose that the speaker has propositionalized an event in such a way that he wants to express X as one of the objects that participate in the event. He must, then, find a word or phrase that will make contact with or establish a corresponding X in his listener’s mind. In this example let us assume that the listener has no prior knowledge of X, so that one result of the speaker’s uttering the sentence he is about to produce will be to establish X as a new item of knowledge for the listener. He needs, then, to communicate his knowledge of X in such a way that the listener will have a sufficient idea of what X is like. That is, the speaker would like the listener to have an idea of X whose content will to some degree approximate the content represented by the small circle above X in the diagram. The listener is not likely to end up with a content as rich as the speaker’s, but he needs to end up with a content that is sufficient for the situation. In order to provide the listener with such a content, the speaker must first interpret X as an instance of a category.
The point labeled A is meant to represent a category, from the point of view that a category can be considered a discrete conceptual unit. And the large circle above it, to which A is connected, is meant to represent the content of that category, which I will assume has both discrete and analogic properties. We can imagine that in the speaker’s mind there are a very large number of such categories, and that the speaker’s task, in categorizing his idea of object X, is to match the content of X (the small circle) as closely as possible to the focus of some category. For we may assume that the content of most categories, like the one represented by the large circle here, has a focus and a gradual fading away from that focus, just as do the color categories that have been the best studied in this respect. In other words, the small circle might be regarded as analogous to a color chip that someone is trying to match to a color category. The large circle represents the (often vaguely defined) boundary of a catgeory, and we may assume that somewhere within it lies a focus, analogous to the focus of the red category, for example.
In this case we are assuming that X is highly codable, which is equivalent to saying that X matches closely the focus of some category, in this case category A. X will then be interpreted as an instance of category A with a high degree of agreement, consistency, and lack of hesitation, and no modifiers will be needed. Let us suppose that in this case one major effect of this interpretation is that X can be called ‘a slide’. When the listener hears it called this, he enters into his own mind a new unit, corresponding to X, whose content matches what is defined by the focus of the slide category in his own knowledge. Under the best of circumstances, where X is highly codable and where the content of the category in question is highly similar for both the speaker and the hearer, the speaker will have succeeded in communicating something very similar in content to what he started out with, namely the content of X. In many cases the communication will not be this effective, but very often discrepancies will be unimportant.
Fig. 4 shows what happens when codability is low. The item being communicated is labeled Y. We will assume that in searching his category space for a category whose content has a focus closely matching the content of Y, the speaker has been unsuccessful. The best he has been able to come up with is category B, but the content of Y lies on the periphery of the category’s content, and not near the focus. Let us suppose that category B provides the word tower for particulars that are interpreted as instances of it. The trouble is that by calling Y ‘a tower’ the speaker will communicate something whose content is like the focus of the tower category—a prototypical tower. But in this case the speaker has in mind a content that differs considerably from that focus, as is suggested by the placement of the small circle above Y at the edge of the large circle above B. This is the situation of low codability, where there is less agreement and less consistency in the categorization, where people are likely to hesitate, and where there is a greater use of modifiers.
The function of modifiers is easy to appreciate against this background. Fig. 5 illustrates the principle involved. An adjective or other modifier provides a content that intersects with the content of the original category. The area defined by this intersection is in a sense a new, ad hoc category, produced by the speaker for the specific purpose of communicating more accurately the content he has in mind. In this example we may assume that the speaker’s dissatisfaction with categorizing Y as an instance of the ‘tower’ category has led him to use modifier C, whose name is ‘multicolored’, and modifier D, whose name is ‘wooden’. The intersection of the content of C and D with the content of B defines a content that is closer to the content the speaker has in mind for Y. And the intention is that when the listener hears the phrase ‘a multicolored wooden tower’ he will form in his own mind a content that is closer to what the speaker was thinking of. More complicated modifier constructions can of course be used. If the speaker says ‘a tall red and white platform that looked like a cattle loading platform except too high’ he is creating an intersection, an ad hoc content, which even more narrowly delimits what he has in mind. I believe that the function of adjectives, relative clauses, and other modifiers must be understood in these terms.
It may be noted that the matching of the content of one’s idea of an object or event with the content of a category must be an analogic procedure, like the matching of a color chip against an area of the spectrum. It necessarily involves comparison of one mental representation with another—of the content of a particular with the content of a category. This matching is not a matter of yes or no, but a matter of degree. A particular may be more or less codable, which is to say that its content may match more or less well the content of some category. Such a situation appears to demand that the mental representation correspond either in a continuous way or in some stepwise fashion to what is being represented. That, apparently, is the essence of analogic, as opposed to propositional representation.
As was discussed briefly at the beginning of this chapter, many people, in recalling something they witnessed earlier or were involved in (even something they were told about), have the subjective feeling that they are reliving the original experience in a way that is not identical to that of the original experience but has something important in common with it. This kind of subjective experience is referred to as mental imagery. It need not, of course, be visual only; auditory and kinesthetic imagery, for example, are common enough. There is an old and still unresolved controversy over the nature, role, and importance of mental imagery in thought and language. There seem to be some real individual differences in imaging ability, which may have something to do with the disagreements that still rage in this area. (See Galton, 1907, for an early but still provocative discussion of individual differences, and Pylyshyn, 1973, and Paivio, 1974, for some modern controversy.) In any case, those whose imaginal experiences are strong tend to feel, as they are telling about something they have recalled, that they are describing a kind of replay of the original experience. Others apparently have less of this feeling, or even deny it altogether, claiming that they are conscious of only language.
It is important to realize that what is at issue here is the nature and significance of the conscious experience that results from recall, and that this question may or may not have a direct bearing on the fundamental issue of how knowledge is stored in subconscious memory. In considering the latter issue, one—but only one—of the possibly relevant considerations is the extent to which these subjective experiences of mental imagery that many of us have during recall should be taken seriously as clues to what is in memory. One possible view is that they should be taken seriously, at least to the extent that they show that memory has ‘analogic’ properties. Another possible view is that mental imagery is a secondary and misleading phenomenon that shows very little or nothing about how knowledge is actually stored.
Those who take the latter position (e.g., Pylyshyn, 1973) like to point out that imagery, whatever it is like, is unquestionably very different from the original sensory stimulus. Images, it is said—and in fact everyone who reflects on the matter ought to be willing to agree—are not like pictures’, in the sense that the same interpretive processes that are applied during perception can be applied to them. Imagery, beyond doubt, is at least a partial product of interpretation, not a raw input. What is in memory, and thus necessarily what is recalled from memory into consciousness, must already have been filtered through perceptual processes. There seems to be no way in which imagery could be a direct copy of the original sensory stimylus. An easy way to demonstrate this to oneself is to look at a scene for a short time and then stop looking at it. After a brief interval, try to recall everything that was in the scene—the objects, relations, and events that it contained. You will find this quite a different experience from looking at the scene itself. In particular, I believe, you will not be able to ‘individuate’—to pick out as discrete entities—objects or events that were not individuated while you were looking at the scene itself. Although I am not aware of psychological experiments specifically aimed at this question, it does appear to be at least intuitively plausible that the individuation of objects and events is something that takes place only while the stimulus is present. If it is not accomplished then, it cannot be accomplished later on. Of course it is possible that we will recall or introduce into our verbalization objects that were not present in the original scene. Memory may very well be constructive, in the sense that it adds things that were not there to begin with (cf. Bartlett, 1932; Münsterberg, 1909). But veridical recall—that which is derived from the original stimulus—seems not to allow individuation that was not accomplished when the stimulus was present.
But this is far from being the whole story. Individuation is not the only kind of interpretive process applied to sensory material. It is one thing to identify, for example, the presence of a certain discrete object within a scene. It is quite another to decide that this object is an instance of some category, for example, that it is the kind of thing that can be called an animal, or a dog, or a Dalmatian. Categorization appears to be a special case of a kind of processing that, for lack of a better term, I will call ‘typing’. The basic idea is that we constantly find it necessary to interpret particular objects, situations, and events, each of which is a separate ‘token’, as instances of general types. And although it seems likely that individuation takes place only during perception, it would appear that typing may take place either during perception or during verbalization. Undoubtedly there is much typing that takes place at the same time an object is individuated. For example, at the same time that I pick out something in a scene as a discrete object I may also identify that object as an instance of the type (category) that may lead me to call it a dog. Or at the same time that I isolate a discrete event I may identify it as an instance of running. But, as we have in fact already seen with respect to categorization, a significant amount of typing takes place also during verbalization—while a person is actually talking.
I believe it is important in understanding how language works to realize that a lot of what goes on when we turn thoughts into words is in the nature of interpretation of the thoughts; that we do not simply transform deep structures into surface structures, where the deep structures already contain all the conceptual interpretation there is going to be. Talking about something recalled from memory is a creative activity, in the sense that the talker must at many points make choices that reflect a specific interpretation of his underlying thoughts. At the same time it is necessarily a distorting process, in that these choices are not likely to reflect the speaker’s thoughts in a completely adequate way. He is likely to be left with the feeling that the words he uttered did not convey exactly what he had in mind. And in talking about the same thing on a later occasion he is likely to make different choices—to verbalize in at least a partially different way.
In summary, there appear to be two major kinds of processes involved in people’s processing of the stimuli that come to them from the outside world: individuation and typing. While individuation probably takes place only during perception, typing may take place either during perception or during verbalization. The latter possibility makes of the use of language a more creative process than linguists have generally conceived it as being. So far, we have specifically considered the creative nature of categorization only. But two other kinds of verbalization processes were also discussed above: subchunking and propositionalizing. It is therefore of interest to consider whether those processes may also exhibit creativity in the manner in which they are applied during language use. Is there, for example, any evidence that speakers make choices regarding subchunking or propositionalizing while they are speaking, as appears to be the case with categorizing? If so, we could conclude that, like categorization, these kinds of interpretations are not necessarily a part of knowledge or memory, but may be a part of talking about what one knows and remembers.
Creativity in Subchunking
In talking about the same thing on different occasions, it is quite likely that a speaker will not perform all subchunking operations in the same way each time. The differences may be of several types. For example, given a particular larger chunk the speaker may break it down into one set of smaller chunks on one occasion, and into a partially different set on another. Occasionally he may even make use of different schemas on different occasions, so that the resulting conceptual interpretations are distinct. I will not try to illustrate such cases here, but will discuss only some aspects of the apparent fact that the different subchunks within a larger chunk are not all of equal status.
One of the scenes in the film showed the boy sliding down a slide. The subchunks within this scene that were verbalized at least once by someone included:
1. arriving at the slide
2. climbing up the back
3. reaching the top
4. standing at the top
5. hitching up his pants
6. sitting down
7. sliding down
8. landing at the bottom
9. jumping off
10. standing up
11. walking away
There was no one who mentioned all these events, and in fact only (1), (2), and (7) were mentioned very frequently. Fig. 6 shows the relative frequency of mention. Nearly everyone verbalized (7), the sliding down. Somewhat more than half mentioned (1) and (2), the arrival and climbing up the slide. The other events were mentioned by only a few people each. It is tempting to attribute these differences in frequency of mention to a factor that might be labeled ‘salience’. In so doing we would, of course, like to be doing more than simply giving a name to the observation. In particular, we would like to have independent evidence for relative degrees of salience among events (with respect to a certain population). I hope that sound evidence of this sort will eventually be forthcoming, but at the moment I can only make a few potentially relevant suggestions.
An interesting phenomenon occurred among the persons who described the film shortly after seeing it and were asked to describe it again after an interval of eight weeks. In the later descriptions the sliding-down event was still mentioned by almost everyone, while the arrival and climbing had decreased in mention from somewhat more than half to somewhat less than half. The remaining events, with one exception, failed to be mentioned at all. The exception was the walking away, which was still mentioned by a small minority.
What explanations are there for this attenuation in the verbalization of all the subchunks, with the exception of the sliding down? One possibility is that salience influences memory as well as verbalization, and that it was only the more salient events that were recalled after eight weeks. A variant form of this explanation might be that all that was remembered after eight weeks was that this scene was an instance of the ‘sliding-down’ schema, and that the schema itself then provided a kind of reconstruction of the event. By this view the schema would be most likely to determine mention of arriving, climbing up, sliding down, and perhaps walking away. The other events, being either peripheral to or altogether outside the schema, would not be so likely to be reconstructed. This explanation is in line with Bartlett’s theory of memory (1932:197-214). Still another possibility is that after eight weeks people were less inclined to include minor details in their descriptions, regardless of whether or not they remembered them. Under conditions of lower motivation the tactic would be to verbalize only the more salient items. These various explanations are not mutually exclusive, and at this stage of our knowledge we might suspect that all these factors play a role. The main point to be noted now is that Fig. 6 seems to predict something about what will be mentioned in later verbalizations of this same scene.
The last possibility mentioned above, that one may choose to verbalize only the more salient items even though one may remember more than that, raises the question of the relationship between salience and summarization. Suppose we think of Fig. 6 as the profile of a mountain range, and that we imagine a fog extending through the area such that only the highest peak rises above it. In that condition the verbalizer would say only “He slid down a slide,” and by that single statement embrace the entire scene. Imagine now that the fog partially lowers, so that the two next highest peaks also appear above it. The verbalizer might then say: “He next comes to a slide. He climbs up the ladder of the slide and then slides down.” Or imagine that the fog has more or less dissipated, leaving some or all of the lowest peaks in view as well. Perhaps then the verbalizer might say: “Next the boy comes upon a slide. He climbs up the ladder of the slide, sits down, and then slides down the slide. He reaches the ground, stands up, and again continues walking in the same direction as he started.” (All these examples are from actual verbalizations.) Subchunking, in other words, usually gives the speaker a choice of how much detail to include—how many sub-chunks to verbalize, and one cannot choose randomly but must be governed by priorities in terms of salience. If the situation calls for a minimum of detail, the speaker will verbalize only the most salient chunk. In such a case we are likely to say that he has provided a ‘summary’ of what was embraced by the larger chunk. By including increasingly less salient items he can provide an increasing amount of detail. It would be interesting to know what factors influence the speaker’s judgment as to the amount of detail he finds appropriate to include.
What are the determinants of salience? Ultimately it will not be enough to say that the different chunks defined by a schema, or chosen in the breakdown of a larger chunk, possess different degrees of salience. Nor will it be enough to discover various, at least partially independent methods by which salience can be measured. A fuller understanding of this matter requires that we eventually establish what it is that makes one item more salient than another. Whatever the answer turns out to be, it seems likely to be a complex one; we cannot expect that only one simple factor is involved.
One factor is likely to be the information value of the item in question: its degree of unexpectedness, in the information theory sense. We can imagine that everyday experience has a kind of base line representing things we fully expect to happen. If life never rose above this base line, it would be totally monotonous. But some of life, one hopes, is not that way. There are things that happen, good or bad, that are to some degree unexpected, and to that degree salient. But there must be more to salience than simple unexpectedness. For example, the boy might not have been expected to hitch up his pants at the top of the slide, yet that event seems not to have been overwhelmingly salient. To achieve some high degree of salience, the event must somehow strike the speaker as interesting or relevant. Perhaps this is a matter of how the event reacts with the speaker’s preconceptions as to what is important and what trivial, but it is hard to say much more on this point at present.
One might also suspect that involvement of the speaker’s own ego is a relevant factor. For example, on another occasion a number of people were asked to describe an incident during which, among other things, a certain person was blindfolded. One of the verbalizers was that person himself, and he was the only one who mentioned that the blindfold was at first ineffective and had to be readjusted. Because of his own involvement in at first being able to see and then not being able to see, the readjustment was salient to him. The onlookers, however, had no involvement of their own at that point; none of them mentioned the readjustment, and probably few of them even perceived it. Finally, and this is related to the factor of unexpectedness, salience seems to have something to do with the extent to which one item entails others. Sliding down the slide was the most salient event in that scene because most of the rest was predictable from it. In order to slide down a slide one has to arrive at the slide, climb it, and so on. On that basis it is easy to understand why mention of this one event provides a good summary of the entire scene; it is all that needs to be known in order to know what else happened.
It would seem, then, that the way a speaker decides to break down larger chunks of content into smaller ones can differ from one verbalization to another, and is not algorithmically determined by what is known’ about the material in question. Furthermore, the chunks into which a larger chunk may be broken down are likely to differ significantly in degree of salience. Thus, even if a hierarchical diagram of subchunking were adequate as a first approximation to a representation of how knowledge is stored, its nodes would have to be marked in some way to indicate their degrees of salience. And these degrees would presumably be related in some analogic way to the as yet little understood determinants of salience just discussed.
Creativity in Propositionalizing
I will close with an example that illustrates how, in verbalizing the same item from memory on different occasions, a person may make different decisions not only regarding categorization and subchunking but also regarding propositionalizing. The example I will use was already mentioned in Chafe (1976), but I wish to expand upon several points that were made there in a sketchy fashion.
Soon after seeing the film, one person described one of the scenes as follows:
He picked up some hay and lifted it over the corral fence and into the corral. All of the animals eagerly went after and began eating the hay.
At the outset the speaker must have had in mind a chunk that included this whole incident (but that was, of course, only a subchunk at some lower level within the large chunk that included the whole film). Suppose we call this initial chunk C1. On the basis of the division into two sentences, as well as the shift from one subject (the boy) to another (the animals), we can say that C1 was first broken down into two sub-chunks, C2 and C3, corresponding to the two sentences. Within C2 there were three minimal events, the picking up, the lifting over, and the lifting into. However, it appears that the last two were chunked together as against the first, since they share a common verb and object (‘lifted it’). On that basis we can say that C2 was first subchunked into C4 (the picking up) and C5, and the latter then further subchunked into C6 and C7. C3 (expressed in the second sentence), contains two minimal events that can be labeled C8 (going after the hay) and C9 (eating it). This pattern of subchunking is summarized in Fig. 7.
In addition to subchunking, the speaker had to propositionalize each of the minimal chunks, C1, C6, C7, C8, and C9. Let us focus on just the first of these. Fig. 8 shows one way of representing the propositional structure of the verbalization ‘He picked up some hay’. The speaker chose to interpret this event in terms of an agent-patient structure, assigning the role of agent to the object labeled O1, and the role of patient to the object labeled O2.
Having made this decision, the speaker went on to categorize O1 and O2, as well as C4 itself. The unlabeled arrows pointing upward show the results of these categorizations. O1 was assumed by the speaker to be already in the consciousness of the listener, and thus it is simply represented by the pronoun ‘he’ (Chafe, 1974). C4 was interpreted as an instance of the category, which resulted in its being communicated with the words pick up’; while O2 was interpreted as an instance of the category, which resulted in its being called ‘some hay’.
Of interest is the fact that the speaker who produced the verbalization just described shortly after seeing the film, eight weeks later verbalized the same scene as follows:
He threw some hay over the top rail of the corral fence to the animals inside.
Several points may be noticed in comparing the two verbalizations. First, so far as subchunking is concerned, the second time the speaker immediately propositionalized the entire scene as a single chunk, rather than breaking it down in the manner shown in Fig. 7. Thus, in comparison with the earlier verbalization, he produced a summary. Second, he chose quite a different propositional format, as shown in Fig. 9. This format is more complex than any in the earlier verbalization, as if the lack of subchunking were compensated by the more elaborate propositional structure. The new format contains not only the roles of agent and patient, but also two roles that are indicated here simply with the arrows labeled ‘over’ and ‘to’. The propositional structure, in other words, assigns roles to four different objects, O1, O2, O3, and O4. It should be noted in particular that the role assigned to O4 (the animals), here a beneficiary role, is entirely different from the role assigned to O4 in the earlier verbalization, where the animals were interpreted not as beneficiaries but as agents.
Starting with the same chunk from memory, C1, the speaker verbalized it quite differently on the two occasions, both in terms of the sub-chunking process and in terms of propositionalizing. It is conceivable that his memory representation underwent some change, in propositional terms, during the eight weeks, so that O4 shifted from an agent role to a beneficiary role. What is to my mind more intriguing and more likely, however, is the possibility that the memory representation contained no commitment to either of these roles—that it was of such a nature that either of them could be chosen. It is cases like this that make one doubt the adequacy of arrows like those labeled ‘agent’, ‘patient’, ‘over’, and ‘to’ in Fig. 9 for the representation of what is stored in long-term memory. And in fact this kind of evidence casts doubt on the adequacy of representations like Figs. 7, 8, and 9 to capture a speaker’s underlying knowledge of an event. At best they seem to capture only certain decisions that were made during certain particular verbalizations. They show, not what was known about the event, but only some aspects of how it was talked about.
The speaker also made categorization decisions. As far as the event C1 is concerned, he chose to interpret it in Fig. 9 as an instance of the category that resulted in its being called ‘throw’. Whether he would have done this eight weeks earlier if he had chosen to verbalize this event directly rather than to break it down into smaller chunks, there is no way to know. For the several objects involved in C1 there was little significant difference in their categorization after eight weeks, although a slightly different treatment was given to O3. Whereas earlier it was called ‘the corral fence’, this time it was called ‘the top rail of the corral fence’. In general one can say that the objects were high in codability, judging from the consistency with which the speaker categorized them. Nevertheless, we know from other examples that speakers are frequently inconsistent and hesitant in categorizing, and that many objects have low codability. Thus, the unlabeled arrows representing categorization in Fig. 9 are in principle the kinds of things that speakers may choose differently in different verbalizations of the same event. These arrows, too, need not represent what is stored in memory, but only what is decided on during verbalization. In sum, none of the arrows in Figs. 7, 8, and 9 need in principle represent what is known about this event, but only decisions as to how to verbalize it. What is known must be of such a nature as to allow multiple interpretations of these kinds. The most attractive conclusion, I think, is that what is known is not necessarily in a propositional format at all, but that it is in many cases in some analogic format that allows various propositional interpretations to be given to it.
The implication for linguistics is that there cannot be any such thing as a well-defined or autonomous semantic structure underlying a discourse (or a sentence). Talking is a creative process by which an underlying knowledge, to a large extent analogic in nature, is crystallized into propositional and linguistic structures. It is true that some, perhaps a great deal, of this crystallization will have taken place when the knowledge was first acquired. But much of it will not. And in order to explain things that we find in the use of language—such as inconsistency and hesitation in subchunking, propositionalizing, and categorizing, and even such a basic thing as the use of modifiers—it is necessary to recognize this as-yet-uncrystallized state of much of our knowledge. If we as linguists are interested in understanding how language really works, we will sooner or later have to expand our horizons to include these readily observable aspects of language use.
Bartlett, F. C. 1932. Remembering. Cambridge: Cambridge University Press.
Brown, R. W., and Lenneberg, E. H. 1954. A study in language and cognition. Journal of Abnormal and Social Psychology 49:454-62.
Chafe, W. L. 1973. Language and memory. Language 49:261-81.
———. 1974. Language and consciousness. Language 50:111-33.
———. 1975. Givenness, contrastiveness, definiteness, subject, topic, and point of view. In C. Li, ed., Subject and Topic. New York: Academic Press.
———. 1976. Creativity in verbalization and its implications for the nature of stored knowledge. In R. O. Freedle, ed., Discourse Production and Comprehension. Hillsdale, N.J.: Lawrence Erlbaum.
Galton, F. 1907. Inquiries into Human Faculty and Its Development, 2d ed. London: J. M. Dent and Sons.
Goffman, E. 1974. Frame Analysis: An Essay on the Organization of Experience. Cambridge: Harvard University Press.
Hochberg, J. E. 1964. Perception. Englewood Cliffs, N.J.: Prentice-Hall.
Miller, G. A. 1956. The magical number seven, plus or minus two: some limits on our capacity for processing information. Psychological Review 63:81-97.
Minsky, M. 1974. A framework for representing knowledge. A. I. Memo 306. Massachusetts Institute of Technology Artificial Intelligence Laboratory.
Münsterberg, Hugo. 1909. On the Witness Stand: Essays on Psychology and Crime. New York: Doubleday, Page.
Paivio, A. 1974. Images, propositions, and knowledge. Research Bulletin 309. The University of Western Ontario, Department of Psychology.
Pylyshyn, Z. W. 1973. What the mind’s eye tells the mind’s brain: a critique of mental imagery. Psychological Bulletin 80:1-24.
Rosch, E. H. 1973. On the internal structure of perceptual and semantic categories. In T. E. Moore, ed., Cognitive Development and the Acquisition of Language. New York: Academic Press.
Schank, R., and Abelson, R. 1975. Scripts, plans, and knowledge. Proceedings of the 4th International Joint Conference on Artificial Intelligence.