Schema Theory (Rumelhart & Norman –1976)

Rumelhart and Norman (1976) described schemata as the means by which generic concepts are represented and then applied in specific situations:

Generic concepts are represented by schemata. These schemata contain variables: references to general classes of concepts that can actually be substituted for the variables in determining the implications of the schema for any particular situation. Particular information is encoded within the memory system when constants—specific values or specific concepts—are substituted for the variables of a general schema. (p. 10)

They further posited that “learning is not a unitary process” (p. 24) and proposed three modes of learning to account for how schemata are acquired and modified: accretion, tuning, and restructuring. Accretion refers to “the normal kind of fact learning, daily accumulation of information in which most of us engage” in which “your knowledge base is merely incremented by a new set of facts” (p. 3):

One basic mode of learning is simply the accumulation of new information. We analyze the sensory events of our current experience, match them with some appropriate set of schemata, form a representation for the experience, and tuck the newly created memory structures away in long-term memory. The newly created data structures are instantiations of the previously existing ones; changed only in that the representations for particular aspects of the current situation have been substituted for the variables of the general schema.

This is learning by accretion: learning by adding new data structures to the existing data base of memory, following the organization already present. Learning by accretion is the natural side effect of the comprehension process. In it, we store some interpretation of the actual experience. If later we retrieve the stored information, we use the instantiated schemata to reconstruct the original experience, thereby “remembering” that experience. The schemata guide reconstruction in much the same way that they guide original comprehension. (Rumelhart & Norman, 1976, p. 13)

Accretion happens when the incoming information is consistent with schemata currently available. When the incoming information does not fit existing schemata, the need for change arises. And “the more discrepant the arriving information from that described by the available schemata, the greater the necessity for change” (Rumelhart & Norman, 1976, p. 21). When the discrepancy is mild, tuning comes into play:

Existing schemata can often serve as the base for the development of new ones by minor changes; by “fine tuning” of their structure. We call this process tuning. We restrict the use of the term “tuning” to those cases where the basic relational structure of the schema remains unchanged, and only the constant and variable terms referred to by the schema are modified. (Rumelhart & Norman, p. 16)

Rumelhart and Norman also explained that the constants and variables of the schema can be changed in four ways (pp. 16-17):

1. Improving the accuracy: The constraints of the variable terms of the schema can be improved to specify the concepts that fit the variables with more accuracy.

2. Generalizing the applicability: The range of a given variable can be generalized to extend its range of applicability. Either the constraints on a variable can be relaxed, or a constant term can be replaced with an appropriately constrained variable term.

3. Specializing the applicability: The range of a given variable can be constrained by adding to the constraints of the variable or, in the extreme, by effectively replacing the variable with a constant term.

4. Determining the default values: The values of the variable that normally apply can be discovered and added to the specification of the schema. Whenever a particular variable is not specified, the default values provide intelligent guesses that can be used in making inferences and guiding further processing.

Restructuring becomes necessary when “existing memory structures are not adequate to account for new knowledge” (Rumelhart & Norman, 1976, p. 14) and the discrepancy is large. Restructuring refers to the creation of brand new schema, through either patterned generation (building a new schema based on an existing schemata, for example by replacing a constant term with a set of variable terms) or schema induction (building a new scheme by combining recurring patterns of existing schemata). Through tuning an existing schema is modified. Through restructuring new schemata are created.

Anderson, Spiro, and Anderson (1978) presented another model that they viewed as an improvement on, and clarification of, both Bartlett’s (1932) ideas on schemata and Ausubel’s (1963; 1968) ideational scaffolding:

Ausubel (1963, 1968) proposed that a reader’s abstract cognitive structures provide the “ideational scaffolding” for the detailed information contained in text. In his words (1968, p. 153), ” . . . [N]ew ideas and information are learned and retained most efficiently when inclusive and specifically relevant ideas are already available in cognitive structure to serve a subsuming role or to furnish ideational anchorage.” Bartlett (1932) suggested a similar notion. However, research in the tradition of Bartlett and Ausubel has proved inconclusive. One reason is that until recently schema notions were hopelessly vague. The purpose of this paper is to provide a clearer formulation of schema theory, and then provide an experimental test of some hypotheses that follow from the theory. (Anderson et al., 1978, pp. 433-434)

One classic example of the role of schema in understanding a reading passage was illustrated by Pichert and Anderson (1977) in a study in which subjects were asked to read the following passage, from the perspective of either a burglar or a homebuyer:

The two boys ran until they came to the driveway. “See, I told you today was good for skipping school,” said Mark. “Mom is never home on Thursday,” he added. Tall hedges hid the house from the road so the pair strolled across the finely landscaped yard. “I never knew your place was so big,” said Pete. “Yeah, but it’s nicer now than it used to be since Dad had the new stone siding put on and added the fireplace.”

There were front and back doors and a side door which led to the garage which was empty except for three parked 10-speed bikes. They went in the side door, Mark explaining that it was always open in case his younger sisters got home earlier than their mother.

Pete wanted to see the house so Mark started with the living room. It, like the rest of the downstairs, was newly painted. Mark turned on the stereo, the noise of which worried Pete. “Don’t worry, the nearest house is a quarter of a mile away,” Mark shouted. Pete felt more comfortable observing that no houses could be seen in any direction beyond the huge yard.

The dining room, with all the china, silver and cut glass, was no place to play so the boys moved into the kitchen where they made sandwiches. Mark said they wouldn’t go to the basement because it had been damp and musty ever since the new plumbing had been installed.

“This is where my Dad keeps his famous paintings and his coin collection,” Mark said as they peered into the den. Mark bragged that he could get spending money whenever he needed it since he’d discovered that his Dad kept a lot in the desk drawer.

There were three upstairs bedrooms. Mark showed Pete his mother’s closet which was filled with furs and the locked box which held her jewels. His sisters’ room was uninteresting except for the color TV which Mark carried to his room. Mark bragged that the bathroom in the hall was his since one had been added to his sisters’ room for their use. The big highlight in his room, though, was a leak in the ceiling where the old roof had finally rotted. (p. 310)

The point of the study was to provide an initial test of the idea that “more significant than the structure that is in some sense contained in a text is the structure the reader imposes on the text” (Pichert & Anderson, 1977, p. 309). The results of the study demonstrated support for two hypotheses:

1. Schemata aid learning by highlighting important elements:

If, for whatever reason, people take divergent perspectives on a text —that is, impose different high-level schemata—the relative significance of text elements will change. Elements that are important on one view may be unimportant on another. By definition, an important element “fits in” to an organized structure of information and is thereby, we hypothesize, more learnable. (p. 309)

2. Schemata aid recall by providing a structure of expectations that serves to cue remembering:

Perspective may have further independent effects on the accessibility of text elements that have been learned. A high-level schema can serve as a retrieval plan, so to speak outlining the questions one should ask of oneself. The schema is bound to provide implicit cues for important elements but is less likely to do so for unimportant ones. Therefore, among those idea units that are stored, the important units will be more accessible and, it is predicted as a consequence, better recalled. (p. 309)

In their general discussion of the results of the study, Anderson et al. (1978) summarized schema theory by saying:

A schema is an abstract description of a thing or event. It characterizes the typical relations among its components and contains a slot or place holder for each component that can be instantiated with particular cases. Interpreting a message is a matter of matching the information in the message to the slots in a schema. The information entered into the slots is said to be subsumed by the schema….Schema theory also provides a reason for supposing that subsuming structures will have effects on memory independent of those on learning. A high-level schema provides a retrieval plan….The idea is that the parent schema from which an instantiated representation of a text originally is constructed can later furnish implicit retrieval cues or specific text information (see Bower, in press).  (p. 314)

A year after publication of the study by Pichert and Anderson (1977), Anderson et al. (1978) presented findings from a similar study that they believed provided “unambiguous confirmation that high-level schemata play a role in the learning and remembering of text information” (p. 437). In the same paper they made the following claims regarding the nature of schemata (p. 434):

  1. A schema will contain slots into which some of the specific information described in a message will fit.
  2. To interpret a particular situation in terms of a schema is to match the elements in the situation with the generic characterizations in the schematic knowledge structure.
  3. The schemata a person already possesses are a principal determiner of what will be learned from a text.
  4. The information that matches slots in the schema would be said to be significant, whereas information that does not would be called unimportant, irrelevant, or—in the limiting case—incongruous.
  5. Information that fits the superordinate schema is more likely to be learned and remembered, perhaps precisely because there is a niche for it.
  6. One schema can provide slots for more of a certain fixed body of information than other schemata.
  7. If the knowledge domain were specified, it should be possible to make qualitative as well as quantitative predictions about just which details will be learned.

Schemata, as described by the foregoing authors, have been assumed to apply to the storage of declarative knowledge. They may also be used to store expectations of procedural knowledge. Procedural knowledge is knowledge concerning how to perform certain activities or processes. A classic example is what to do when ordering food in a restaurant. Through prior experience we learn to expect the waiter or waitress to do and say certain things, and we also know what is expected of us. Rumelhart (1980) compared schemata to plays, which have characters and roles, lines, settings, and actions. Schemata that store procedural knowledge are generally referred to a scripts (see, for example, Schank & Abelson, 1977).

Bower, Black and Turner (1979) published the results of a handful of studies in which they investigated the organization of knowledge of routine activities, such as eating at a restaurant or visiting a dentist, and how that knowledge is used to help people understand and remember narrative texts. In two of the studies they asked subjects to describe the details of what happens during familiar activities. They found that the various accounts collected largely agreed with each other regarding the nature of the characters, objects present, actions, and the order of those actions. They also found agreement on the level of granularity at which the action sequences might be divided into “scenes,” that they interpreted to infer a hierarchical organization in memory of the activity. In other studies they investigated memory based on a script. Their results showed that subjects tended to confuse actions that were stated with actions that were unstated but implied by the script. They also found that subjects preferred to recall script actions in their familiar order. When a text presented actions that were scrambled, subjects would tend to recall them in canonical, rather than presented, order. An additional finding was that subjects reading a text would have to slow down to read statements that deviated from expectations, but that goal-relevant deviations were remembered better than script actions, suggesting that goal-relevant deviations held more significance. Though a later study (Mandler & Murphy, 1983) using the same scripts brought into question the strength of some of the claims made by Bower et al., both studies showed substantial agreement between subjects suggesting the use of schema in organizing knowledge.

Leave a Comment