Schema Theory (drawn from D’Andrade 1995)


Plato elaborates the Greek doctrine of ideal types – such as the perfect circle that exists in the mind but which no one has ever seen. Kant further developed the notion and introduced the word schema. For example, he describes the "dog" schema a mental pattern which "can delineate the figure of a four-footed animal in a general manner, without limitation to any single determinate figure as experience, or any possible image that I can represent in concreto." (Kant 1781).

Since that time, many other terms have been used as well, including "frame," "scene," "scenario," "script" and even "model", "theory". Key theoretical development of schema theory was made in several fields, including linguistics, anthropology, psychology and artificial intelligence. The heyday of schema theory was probably in the ‘70s (although of course in OB it has barely arrived). One of the main engines was artificial intelligence, which was engaged in getting computers to read natural text. It was quickly discovered that most of what is communicated in a newspaper article cannot be understood without reference to a great deal of information that is not included in the article itself. For example, consider this story from D’Andrade (1995):

John wanted to do well on the exam, but his pen ran out of ink and his pencil broke. He tried to find a pencil sharpener, but there wasn't one in the room. Finally he borrowed a pen from another student. By then he was so far behind he had to rush, and the teacher took off points for poor penmanship.

To understand this story, you have to understand the writing schema because the text itself leaves unstated the connection between John running out of ink and his not being able to work on the exam.

The writing schema serves as a good introduction to the idea. Fillmore (1975) contrasts the "text coherent relations" involved in the English verb write and the Japanese term kaku. These two terms are often translated as synonyms but the schemas they name are slightly different. Both schemas include a scene in which somebody guides a pointed trace-leaving implement across a surface. Such a scene invokes a writer, an implement, a surface on which traces are left, and a product. Neither schema specifies the particular implement used – it could be a pencil, a pen, a piece of chalk, a typewriter, a stick, or even sky-writing airplane. Similarly, the surface can be paper, a black-board, or patch of dirt, or the sky. The thing written can be in any script and in any language and vary from a single letter of the alphabet to a massive monograph. The one difference is that in the English schema, what is written is always text (words, numbers, linguistic symbols). In the Japanese schema, however, even that is flexible. Hence what is kaku-ed may be a doodle, or a drawing. Hence the English schema is a special case of the Japanese schema.


Schemas as organizational units

A schema, then, is a set of related place-holders or slots which can be filled in by context or by additional information from the speaker. Often, what is filled in for one slot may affect what can be filled in for other slots. For example, in the writing schema, if the surface being written upon is the sky, then the implement is likely to be an airplane. When no specific information is provided for a particular slot, hearers tend to fill the slots with their normal expectations -- "default values." For example, the default values for the English writing schema probably include a pen as the implement and paper as the surface.

A schema filled in with default values is called a prototype. Whereas a schema is an organized abstract framework of objects and relations, a prototype consists of a specified set of expectations. A prototype is a highly typical instantiation or instance of a schema (Langacker 1987).

One function of schemas is to relate terms drawn different linguistic domains. For example, the writing schema relates pencils, pens, chalk, typewriters, etc., to paper, blackboards, newspapers, manuscripts, etc., to English, French, Arabic, etc., to authors, correspondents, pen-pals, memos, etc.

An important aspect of the organization of schemas is that simpler schemas can be "embedded" within more complex schemas. Or, to put it another way, schemas can be hierarchically structured. The writing schema contains within it a number of sub-schemas; the schema for a writing implement, a writing surface, a language, and an entity that is trying to communicate. Further, each of these sub-schemas is composed of sub-sub schemas; there are schemas for pens, paper, English, authors, etc.

Some schemas are especially fundamental and fertile. For example, the concept of a container is an abstract perceptual object which possesses (contains!) an interior, a boundary, and an exterior. The body is a container. The container schema is implicit in a wide variety of English utterances, including "he walked out of the room", "he fell into a trap". Lakoff’s well-known book Women, Fire, and Dangerous Things discusses these kinds of things.

Languages often have terms which identify ways in which the full schema fails or is only partially satisfactory. The term scrawl refers to a failure to make one's writing on something fully legible, the term illiterate refers to someone who is not able to utilize the writing schema because they cannot read or write effectively. Such terms can only be defined and understood through a comprehension of the full schema and what can go wrong within it.

Dorothy Holland and Debra Skinner (1987) studied the US undergraduate dating schema. They describe the "taken-for-granted world of male/female relations" from the perspective of a female undergraduate as follows:

"… a male earns the admiration and affection of a female by treating her well. Intimacy is a result of this process. The female allows herself to become emotionally closer, perhaps as a friend, perhaps as a lover, perhaps as a fiancee, to those attractive males who make a sufficient effort to win her affection. Besides closeness and intimacy, the process of forming a relationship also has to do with prestige. When a male is attracted to a female and tries to earn her affection by good treatment, her attractiveness is validated and she gains prestige in her social group. For his part, the male gains prestige among his peers when he receives admiration and affection from and gains intimacy with females.

Normally, prestigious males are attracted to and establish close relations with prestigious females, and vice versa. Sometimes, however, a male can succeed in winning the affection of a female whose prestige is higher than his own. However, the more attractive she is, the more he must compensate for his lack of prestige by spectacular efforts to treat her well. correspondingly, females sometimes do form close relationships with males who have higher prestige than they do. When the male is more attractive or has higher prestige than the female, she often must compensate by giving her affection to him without his doing anything to earn it." (1987:101-102)

Within this simplified and idealized world, one set of problematic males is termed jerks, nerds, turkeys, and asses. These are men who are undesirable and don’t know it. They are unattractive (physically or otherwise) and don’t or can’t make up for it with higher cost gifts and other exchange items. Furthermore, they are too dumb to "take a hint," and therefore have to be rejected in such direct ways that the women have to be repeatedly unpleasant, which is stressful for the women. To understand what one of these college women means when she calls a man a jerk we need to understand the (women’s) dating schema.


Schemas as Processors

Schemas clearly affect our recall of events. Russ Bernard, Peter Killworth, David Kronenfeld, and Lee Sailer did a series of studies on informant accuracy in which they convincingly showed that about half of what informants tell us is wrong. This caused a firestorm of argument and replication, including one study by Linton Freeman, Kim Romney, and Sue Freeman (1987), which suggested that the inaccuracies are not random, but rather systematic and predictable, and come from the schemas we develop based on experience and social networks.

The Freeman, Romney, and Freeman study used attendance data for participation in an informal seminar of the Irvine Mathematical Social Science Program during the spring quarter of 1985. The seminar met in a large lounge in which participants sat in a circular pattern, with everyone visible to everyone else. Attendance was recorded for nine consecutive sessions. The ninth and final session was used as the memory target session. All participants were interviewed five days after the final session and asked if they had been at the ninth meeting, and if so, who else had attended. All those who had attended (17 persons) remembered attending. With regard to who else attended, their error rate was 52% (exactly in line with the Bernard et al studies). On average, respondents forgot about 6 of the 16 other people that were there, and recalled the presence of 1.5 persons who, in fact, were not there.

Based on a survey of the literature on memory, Freeman et al. hypothesized that memory recall of an event will depend on two major factors; first, how well organized the person's schema is for that kind of event, and secondly, how typical the event to be remembered is. The better organized the schema, the better over-all memory will be, and the more typical an event, the more likely it is that it will be remembered. However, the more typical an event, the more likely it will be recalled falsely as occurring because it will be "filled-in" by the schema rather than actual perception.

In the analysis, Freeman et al. divided the seminar participants into two groups; an "ingroup" of faculty and graduate students who all have space together on one floor of the Social Science Building, and an "outgroup" of faculty and students whose offices are elsewhere. This spatial division corresponds to a clear social reality in which certain people are face-to-face colleagues, teaching similar courses and involved together in a variety of day-to-day concerns. Freeman et al. expected that the ingroup would have a much better formed schema for the membership of the seminar, and thus to remember more of the participants for the final session. This, in fact, was the case; ingroup members forgot on the average the attendance of 4.7 persons, while outgroup members forgot 8.1 persons. However, the ingroup produced more false recalls, having an average of 2.8 false recalls per informant, while the outgroup only produced 0.4 false recalls per informant. Not surprisingly, those who attended most of the nine sessions were forgotten least often and were most likely to be the subject to false recalls.

Freeman et al. show that, using just the three most productive informants (those who produced the most names), and simply counting the number of times a person's name is recalled, one can accurately identify all of the ten most active participants of the seminar. The correlation between the number of times a name is recalled by the ten most productive informants and the attendance of that person across the nine sessions is .88. What this means is that the memory of those with well-formed cognitive structures reflects long4erm patterns quite well, although they are more likely to distort what happened at an actual event, adding in those who are usually there and forgetting those who are not regular attendees.

The results from the least productive participants are quite different. Pooling the memory data of the five least productive informants predicts all the participants who were actually at the ninth session with only one false recall. Because the low-knowledge informants had little cognitive structure to work with, their memory is not productive but it is not biased. While each non-productive informant provides meager information, pooled together in sufficient numbers they provide a good picture of what actually happened. Thus, while memory is often both scanty and biased, biased memory can be used to uncover the longterm patterning of events, and scanty memories, when pooled, can be used to recover what actually happened at specific moments.

It is not just researchers that believe in the power of schemas. Native speakers do too – they seem to assume that listeners understand through their schemas. For example, Schank & Abelson (1977), present this sentence:

John went to a restaurant. He asked the waitress for coq au vin. He paid the check and left.

Why would the speaker say "the waitress" and "the check", unless he or she was sure that the listener would know that restaurants have waitresses and checks?

Schemas also affect our ability to learn things. A fascinating experiment was performed by DeSoto (1960), who was interested in how people learned social networks. Each subject was presented with a deck of 12 cards that referred to an hypothetical four-person collectivity. The front of each card listed a pair of names, like Jim and Stan. Subjects were told that they were to learn who liked whom in the four-person community. They were instructed to look at the face of the card and report whether the person named first likes the one named second, and then to turn the card over. The back of each card revealed whether the first person liked the second person or not, under the experimental condition being tested. The cards were then shuffled and returned to the subject for another trial. This procedure was repeated until the subject completed two consecutive error-free trials.

DeSoto used eight different social structures as experimental conditions. The important result in the context of the present problem was that the mean number of trials to the criterion was a function of the degree to which the structure being learned approached a form that was both symmetric and transitive. In a second part of the study, the subjects were told that the pairs represented boss-subordinate pairs. This time, subjects learned faster if the structure as anti-symmetric and non-transitive. Overall, it was as though the subjects learned more quickly when the structure in the data conformed to their preconceived notions of how friendship, supervision, and other social relations "work".