Warning Concerning Copyright Restrictions The Copyright law of the United States (Title 17, United States Code) governs the making of photocopies or other reproductions of copyright material. Under certain conditions specified in the law, Iibraries and archives are authot- ized to furnish a photocopy or other reproduction. One of these specified conditions is that the photocopy or reproduction not be “used for any purposes other than private study, schohu- ship, or research.” If a user makes a request for, or Iater uses, a photocopy or reproduction for purposes in excess of “fair use,” that user may be liable for copyright infringement.

5 Cognitive Flexibility, Constructivism, and

Hypertext: Random Access Instruction for Advanced Knowledge Acquisition

in Ill- Structured Domains Rand J. Spiro Center for the Study of Reading

University of Illinois Paul J. Feltovich Southern Illinois University

School of Medicine Michael J. Jacobson Center for the Study of Reading

University of Illinois Richard L. Coulson Southern Illinois University

School of Medicine INTRODUCTION: THE COMPLEX CONTEXT OF LEARNING AND THE DESIGN OF INSTRUCTION

A central argument of this chapter is that there is a common basis for the failure of many instructional systems. The claim is that these deficiencies in the outcomes of learning are strongly influenced by underlying biases and assumptions in the design of instruction which represent the instruc- tional domain and its associated performance demands in an unrealisti- cally simplified and well- structured manner. We offer a constructivist theory of learning and instruction that emphasizes the real- world com- plexity and ill- structuredness of many knowledge domains. Any effective approach to instruction must simultaneously consider several highly in- tertwined topics, such as:

l the constructive nature of understanding;

58 Spiro et al. Cognitive Flexibility, Constructivism, and Hypertext 59 5. l the complex and ill- structured features of many, if not most, knowl-

edge domains; l patterns of learning failure; and l a theory of learning that addresses known patterns of learning fail-

ure. Based on a consideration of the interrelationships between these topics, we have developed a set of principled recommendations for the development of instructional hypertext systems to promote successful learning of diffi- cult subject matter (see Spiro, Coulson, Feltovich, & Anderson, 1988; Spiro & Jehng, 1990). This systematic, theory- based approach avoids the ad hoc character of many recent hypertext- based instructional programs, which have too often been driven by intuition and the power of the technology.

In particular, we argue that: l Various forms of conceptual complexity and case- to- case irregularity

in knowledge domains (referred to collectively as wness) pose serious problems for traditional theories of learning and in- struction. l Cognitive and instructional neglect of problems related to content

complexity and irregularity in patterns of knowledge use leads to learning failures that take common, predictable forms. These forms are characterized by concmimplification and the inability to apply knowledge to new cases (failures of transfer). l The remedy for learning deficiencies related to domain complexity

of learning processes that af- This includes the ability to represent

and case perspectives and then, the ability to construct from those different conceptual and case representations a knowledge en-

semble tailored to the needs of the understanding or problem- solv- ing situation at hand. l For learners to develop cognitively flexible processing skills and to

acquire contentive knowledge structures which can support flexible cognitive processing, f&& le learninos are required which permit the same items of knowledge to be presented and learned in a variety of different ways and for a variety of different purposes (commensurate wm complex and irregular nature). l The computer is ideally suited, by virtue of the flexibility it can pro-

vide, for fostering cognitive flexibility. In particular, multidimen- rtext svstems, if appropriately desw to e considerations discussed above, have the power to convey ill- structured aspects of knowledge domains and to

promote features of cognitive flexibility in ways that traditional learning environments (textbooks, lectures, computer- based drill) could not (although such traditional media can be very successful in other contexts or for other purposes). We refer to the principled use of flexible features inherent in computers to produce nonlinear

learning environments as Ii andom Access InstructionifSpiro & Jehng, 690). Following our injunction to consider all crucial issues in the learning and instruction environment jointly, we will develop the following com- pound argument, which integrates the claims just presented:

l Characteristics of ill- structuredness found in most knowledge do- mains (especially when knowledge application is considered) lead to serious obstacles to the attainment of advanced learning goals (such as the mastery of conceptual complexity and the ability to indepen- dently use instructed knowledge in new situations that differ from the conditions of initial instruction). These obstacles can b- e over- come by shifting from a constructive orientation that emphasizes the retrieval from memory of intact preexisting knowledge to an alterna- tive constructivist stance which stresses the flexible reassembly of preexisting knowledge to adaptively fit the needs of a new situation. Instruction based on this new constructivist orientation can promote the development of cognitive flexibilityfising theory- based hypertext e systems that themselves possess characteristics of flexibihty that mir- ror those desired for the learner. In summary, ill- structured aspect f knowledge pose problems for ad- vanced knowledge acquisition that are remedied by the principles of Cog- nitive Flexibility Theory. This cognitive theory of learning is systemati- cally applied to an instructional theory,% andom Access Instruction, which in turn guides the design of nonlinear computer learning environments we refer to as Cognitive Flexibility Hypertexts.

SELECTIVE FOCUS ON ADVANCED KNOWLEDGE ACQUISITION IN ILL- STRUCTURED DOMAINS

The argument developed in this chapter is not intended to cover all as- pects of constructive mental processing. Similarly, instructional technol- ogy is a broad topic that will not be exhaustively addressed in this chapter. Rather we will focus on a set of issues implicated by consideration of some ipecial instructional objectives (Merrill, 1983) and the factors con- tributing to their attainment. In particular, we will be concerned only with learning objectives important to advanced (post- introductory) knowledge acquisition: to attain an understanding of important elements of conceptual complexity, to be able to use acquired concepts for reasoning and inference, and to be able to flexibly apply conceptual knowledge to novel situations. Furthermore, we will consider only complex and ill- structured domains (defined later). This combination of ambitious learn- ing goals and the unobliging nature of characteristics associated with cer- tain knowledge domains will be seen to present special problems for learn-

60 Spiro et al. ing and instruction that call for special responses at the level of cognitive theory and related instructional interventions.

We will argue that one kind of hypertext approach is particularly ap- propriate for this constellation of features associated with the instructional context. The omission of other varieties of computer- based instruction from our discussion does not imply any negative evaluation of their mer- its. Indeed, in other instructional contexts the kinds of hypertexts we will discuss would be inappropriate (e. g., computer- based drill would be better suited to the instructional objective of memorizing the multiplication ta- bles; see Jacobson & Spiro, 1991b, for the presentation of a framework for analyzing instructional contexts to determine the choice of educational technologies).

In what follows, we illustrate how a particular set of factors in the in- structional context (including learning goals and the nature of the knowl- edge domain) and a set of observed learning deficiencies jointly lead to a recommended cognitive theory- based instructional approach.

THE NATURE OF ILL- STRUCTURED KNOWLEDGE DOMAINS AND PATTERNS OF DEFICIENCY IN

ADVANCED KNOWLEDGE ACQUISITION Ill- Structured Knowledge Domains: Conceptual Complexity

and Across- Case Irregularity An ill- structured knowledge domain is one in which the following two properties hold: (a) each case or example of knowledge application typi- cally involves the simultaneous interactive involvement of multiple, wide- application conceptual structures (multiple schemas, perspectives, organizational principles, and so on), each of which is individually com- plex (i. e., the domain involves concept- and case- complexity); and (b) the pattern of conceptual incidence and interaction varies substantially across cases nominally of the same type (i. e., the domain involves across- case ir- regularity). For example, understanding a clinical case of cardiovascular pathology will require appreciating a complex interaction among several central concepts of basic biomedical science; and that case is likely to in- volve differences in clinical features and conceptual involvements from other cases assigned the same name (e. g., other cases of “congestive heart failure”). Examples of ill- structured domains include medicine, history, and literary interpretation. However, it could be argued that even those knowledge domains that are, in the main, more well- structured, have as- pects of ill- structuredness as well, especially at more advanced levels of study (e. g., mathematics). Furthermore, we would argue that all domains which involve the application of knowledge to unconstrained, naturally

5. Cognitive Flexibility, Constructivism, and Hypertext 61 occurring situations (cases) are substantially ill- structured. For example, engineering employs basic physical science principles that are orderly and regular in the abstract and for textbook applications (Chi, Feltovich, & Glaser, 1981). However, the application of these more well- structured con- cepts from physics to “messy” real- world cases is another matter. The na- ture of each engineering case (e. g., features of terrain, climate, available materials, cost, etc.) is so complex and differs so much from other cases that it is difficult to categorize it under any single principle, and any kind of case (e. g., building a bridge) is likely to involve different patterns of principles from instance to instance. Similarly, basic arithmetic is well- structured, while the process of applying arithmetic in solving “word problems” drawn from real situations is more ill- structured. For example, consider the myriad ways that arithmetic principles may be signaled for ac- cess by different problem situations and problem wordings.

Advanced Knowledge Acquisition: Mastery of Complexity and Preparation for Transfer The objectives of learning tend to differ for introductory and more ad- vanced learning. When first introducing a subject, teachers are often satis- fied if students can demonstrate a superficial awareness of key concepts and facts, as indicated by memory tests that require the student only to re- produce what was taught in roughly the way that it was taught. Thus, in introductory learning, ill- structuredness is not a serious problem. Learn- ers are not expected to master complexity or independently transfer their acquired knowledge to new situations. These latter two goals (mastery of complexity and transfer) become prominent only later, when students reach increasingly more advanced treatments of the same subject matter. It is then, when conceptual mastery and flexible knowledge application be- come paramount goals, that the complexity and across- case diversity char- acteristic of ill- structured domains becomes a serious problem for learning and instruction.

Patterns of Advanced Learning Deficiency in Ill- Structured Domains and Remedies in “Cognitive Flexibility Theory”

In this section we briefly review two related bodies of research: the nature of learning failures in advanced knowledge acquisition and new theoretl- Cal approaches to more successful advanced learning and instruction.

Forms of a “Reductive Bias” in Deficient Advanced Knowledge Acquisi- tion. Advanced knowledge acquisition, that very lengthy stage between introductory treatments of subject matter and the attainment of expertise for the subject, has been very little studied (certainly in comparison to the

62 Spiro et al. Iarge number of studies of novices and experts-- e. g., Chase & Simon, 1973; Chi et al., 1981; Feltovich, Johnson, Moller, & Swanson, 1984). However, in our own recent investigations of advanced learning in ill- structured domains, we have found a number of notable results, some of which were somewhat surprising (Coulson, Feltovich, & Spiro, 1989; Feltovich, Spiro, & Coulson, 1989; Myers, Feltovich, Coulson, Adami, & Spiro, 1990; Spiro, Feltovich, Coulson, & Anderson, 1989). These results may be summarized as follows:

l Failure to attain the goals of advanced knowledge acquisition is common. For example, when students are tested on concepts that are consensually judged by teachers to be of central importance and that have been taught, conceptual misunderstanding is prevalent. l A common thread running through the deficiencies in learning is

oversimplification. We call this tendency the reductive bias, and we have observed its occurrence in many forms. Examples include the additivity bias, in which parts of complex entities that have been studied in isolation are assumed to retain their characteristics when the parts are reintegrated into the whole from which they were drawn; the discreteness bias, in which continuously dimensioned at- tributes (like length) are bifurcated to their poles and continuous processes are instead segmented into discrete steps; and the com- partmentalization bias, in which conceptual elements that are in re- ality highly interdependent are instead treated in isolation, missing important aspects of their interaction (see Coulson et al., 1989; Fel- tovich et al., 1989; Myers et al., 1990; Spiro et ai., 1989, for presenta- tions and discussion of the many reductive biases that have been identified). Of course, the employment of strategies of this kind is not a problem if the material is simple in ways consistent with the reductive bias. However, if real complexities exist and their mastery is important, such reduction is an inappropriate oversimplification. l Errors of oversimplification can compound each other, building

larger scale networks of durable and consequential misconception. l The tendency toward oversimplification applies to all elements of the

learning process, including cognitive strategies of learning and men- tal representation, and instructional approaches (from textbooks to teaching styles to testing). These various sources of simplification bias reinforce each other (e. g., one is more likely to oversimplify if an inappropriately easier learning strategy is also employed in textbooks or teaching because it is simple). As we will see in the next section, more appropriate strategies for ad- vanced learning and instruction in ill- structured domains are in many ways the opposite of what works best for introductory learning and in more well- structured domains. For example, compartmentalization of knowledge components is an effective strategy in well- structured do-

mains, but blocks effective learning in more intertwined, ill- structured domains that require high degrees of knowledge interconnectedness. In-

5. Cognitive Flexibility, Constructivism, and Hypertext 63 structional focus on general principles with wide scope of application across cases or examples works well in well- structured domains (this is one thing that makes these domains well- structured), but leads to seduc- tive misunderstandings in ill- structured domains, where across- case vari- ability and case- sensitive interaction of principles vitiates their force. Well- structured domains can be integrated within a single unifying repre- sentational basis, but ill- structured domains require multiple representa- tions for full coverage. For example, consider one kind of single unifying representation, an analogy to a familiar concept or experience. We have found that a single analogy may help at early stages of learning, but actu- ally interfere with more advanced treatments of the same concept later on (Spiro et al., 1989; see also Burstein & Adelson, 1990). Any single analogy for a complex concept will always be limited in its aptness, and misconcep- tions that will develop when the concept is treated more fully can be pre- dicted by knowing the ways in which the introductory analogy is mislead- ing about or under represents the material to be learned. To summarize, we have found that the very things that produce initial success for the more modest goals of introductory learning may later impede the attain- ment of more ambitious learning objectives.

There is much that appears to be going wrong in advanced learning and instruction (see also GPEP, 1984; Perkins & Simmons, 1989). The cognitive theories and instructional practices that work well for introductory learning and in well- structured domains not only prove inadequate for later, more advanced treatments of the same topics, but adherence to those theories and practices may produce impediments to further progress. Our conclusion is that a reconceptualization of learning and instruction is required for advanced knowledge acquisition in ill- structured domains (see also Feltovich, Spiro, & Coulson, in press; Spiro & Jehng, 1990; Spiro et al., 1987, 1988, 1989). Such a reconceptualization, taking into account the problems posed by domain ill- structuredness and the patterns of advanced learning deficiency observed in our studies, is presented next, in our discussion of constructivism and a new constructive orientation, Cognitive Flexibility Theory. After a brief survey of the tenets of that theory, we show its implications for the design of computer hypertext learning environments that are targeted to. the features of difficulty faced by advanced learners in ill- structured domains.

CONSTRUCTIVISM, OLD AND NEW: COGNITIVE FLEXIBILITY THEORY AND THE PROMOTION . OF ADVANCED KNOWLEDGE ACQUISITION

Our interpretation of constructivism, as it is applied to learning and in- struction, is complex. We argue that there are different points in cognitive

64 Spiro et al. acts where constructive mental processes occur. First, we take it as an ac- cepted cognitive principle that understanding involves going beyond the presented information. For example, what is needed to comprehend a text is not solely contained in the linguistic and logical information coded in that text. Rather, comprehension involves the construction of meaning: The text is a preliminary blueprint for constructing an understanding. The information contained in the text must be combined with informa- tion outside of the text, including most prominently the prior knowledge of the learner, to form a complete and adequate representation of the text’s meaning (see Spiro, 1980, for a review; also see Ausubel, 1968; Bartlett, 1932; Bransford & Johnson, 1972; Bruner, 1963).

However, our approach to constructivist cognition goes beyond many of the key features of this generally accepted view (see Spiro et al., 1987). The interpretation of constructivism that has dominated much of cogni- tive and educational psychology for the last 20 years or so has frequently stressed the retrieval of organized packets of knowledge, or schemas, from memory to augment any presented information that is to be understood or any statement of a problem that is to be solved. We argue that concep- tual complexities and across- case inconsistencies in ill- structured knowl- edge domains often render the employment of prepackaged (“ precompiled”) schemas inadequate and inappropriate. Rather, because knowledge will have to be used in too many different ways for them all to be anticipated in advance, emphasis must be shifted from the retrieval of intact knowledge structures to support the construction of new under- standings, to the novel and situation- specific assembly of prior knowledge drawn from diverse organizational loci in preexisting mental representa- tions. That is, instead of retrieving from memory a previously packaged “prescription” for how to think and act, one must bring together, from various knowledge sources, an appropriate ensemble of information suited to the particular understanding or problem- solving needs of the situation at hand. Again, this is because many areas of knowledge have too diverse a pattern of use for single prescriptions, stored in advance to cover enough of the cases that will need to be addressed. (For other dis- cussions of issues related to cognitive flexibility and “inert knowledge,” see Bereiter & Scardamalia, 1985; Bransford, Franks, Vye, & Sherwood, 1989; Brown, 1989; Whitehead, 1929.)

Thus, in Cognitive Flexibility Theory, a new element of (necessarily) constructive processing is added to those already in general acceptance, an element concerned primarily with the flexible use of preexisting knowl- edge (and, obviously, with the acquisition and representation of knowl- edge in a form amenable to flexible use). (However, also see Bartlett’s 1932, notion of “turning round upon one’s schema.“) This “new construe: tivism” is doubly constructive: (a) understandings are constructed by using prior knowledge to go beyond the information given; and (b) the prior knowledge that is brought to bear is itself constructed, rather than re- trieved intact from memory, on a case- by- case basis (as required by the

5. Cognitive Flexibility, Constructivism, and Hypertext 65 across- case variability of ill- structured domains). (Also see Bereiter, 1985.) Cognitive Flexibility Theory is a “new constructivist” response to the diffi- culties of advanced knowledge acquisition in ill- structured domains. It is an integrated theory of learning, mental representation, and instruction. We now turn our attention to that theory. (Having discussed the relation- ship of Cognitive Flexibility Theory to constructivism, the latter term will not be used explicitly very often in the remainder of the chapter- but it should be understood that when we talk about Cognitive Flexibility The- ory, we are referring to a particular constructivist theory.)

Cognitive Flexibility Theory: A Constructivist Approach to Promoting Complex Conceptual Understanding

and Adaptive Knowledge Use for Transfer Limitations of space will not permit a detailed treatment of the key fea- tures of Cognitive Flexibility Theory in this section. Let it suffice to say that the tenets of the theory are direct responses to the special require- ments for attaining advanced learning goals, given the impediments asso- ciated with ill- structured features of knowledge domains and our findings regarding specific deficiencies in advanced learning- knowing what is go- ing wrong provides a strong clue for how to fix it. In lieu of any compre- hensive treatment, we will discuss here one central aspect of the theory. Then, we will show how that aspect creates implications for the design ( and use of hypertext learning environments. For more detailed treat- * ments of Cognitive Flexibility Theory, see Spiro et al. (1987, 1988), Spiro and Jehng (1990), and Feltovich et al. (in press).

The aspect of Cognitive Flexibility Theory that we will briefly discuss here and use for illustrative purposes involves the importance of multi- ple juxtapositions of instructional content. Some other aspects. of the the- ory will be referred to in passing in the context of that discussion. (Many key tenets of Cognitive Flexibility Theory will not be mentioned at all; e. g., the vital importance of students’ active participation in learning.) A cen- tral claim of Cognitive Flexibility Theory is that revisiting the same mate- rial at different times, in rearranged contexts, for different purposes, and fro6 different conceptual perspectives is essential for attaining the goals of advanced knowledge acquisition (mastery of complexity in understanding and preparation for transfer). Content must be covered more than once for full understanding because of psychological demands resulting from the complexity of case and concept entities in ill- structured domains, com- bined with the importance of contextually induced variability and the need for multiple knowledge representations and multiple interconnect- edness of knowledge components (see Spiro et al., 1988, for justifications of all these requirements). Any single explanation of a complex Concept or case will miss important knowledge facets that would be more salient in a different context or from a different intentional point of view. Some Of

Spiro et al. the representational perspectives necessary for understanding will be grasped on a first or second exploration, while others will be missed until further explorations are undertaken. Some useful connections to other instructed material will be noticed and others missed on a single pass

(with connections to nonadjacently presented information particularly likely to be missed). And so on. Revisiting material in an ill- structured domain is not a simple repetitive process useful only for forming more durable memories for what one already knows. For example, re- examin- ing a case in the context of comparison with a case different from the comparison context (i. e., the first time the case was investigated) will lead to new insights (especially if the new “reading” is appropriately guided); this is because partially nonoverlapping aspects of the case are highlighted in the two different contexts. The more complex and ill- structured the domain, the more there is to be understood for any instructional topic; and therefore, the more that is unfortunately hidden in any single pass, in any single context, for any restricted set of purposes, or from the perspec- tive of any single conceptual model.

For example, consider the importance of multiple knowledge represen- tations, which is one thing made possible by multiple passes through the same material. A key feature of ill- structured domains is that they em- body knowledge that will have to be used in many different ways, ways that cannot all be anticipated in advance. Knowledge that is complex and ill- structured has many aspects that must be mastered and many varieties of uses that it must be put to. The common denominator in the majority of advanced learning failures that we have observed is oversimplification, and one serious kind of oversimplification is looking at a concept or phe- nomenon or case from just one perspective. In an ill- structured domain, that single perspective will miss important aspects of conceptual under- standing, may actually mislead with regard to some of the fuller aspects of understanding, and will account for too little of the variability in the way knowledge must be applied to new cases (Spiro et al., 1989). Instead, one must approach all elements of advanced learning and instruction with the tenet of multiple representations at the center of consideration.

Cognitive Flexibility Theory makes specific recommendations about multiple approaches that range from multiple organizational schemes for presenting subject matter in instruction to multiple representations of knowledge (e. g., multiple classification schemes for knowledge representa- tion). Knowledge that will have to be used in a large number of ways has to be organized, taught, and mentally represented in many different ways. The alternative is knowledge that is usable only for situations like those of initial learning; and in an ill- structured domain that will constitute just a small portion of the situations to which the knowledge may have to be applied.

Given all of this, it should not be surprising that the main metaphor we employ in the instructional model derived from Cognitive Flexibility Theory (and in our related hypertext instructional systems) is that of the

Cognitive Flexibility, Conshuctivism, and Hypertext 67 5. c& s- crossed landscape (Spiro et al., 1987; Wittgenstein, 1953), with its sug- gestion of a nonlinear and multidimensional traversal of complex subject matter, returning to the same place in the conceptual landscape on differ- ent occasions, coming from different directions. Instruction prepares stu- dents for the diversity of uses of ill- structured knowledge, while also demonstrating patterns of multiple interconnectedness and context de- pendency of knowledge, by criss- crossing the knowledge domain in many ways (thereby also teaching students the importance of considering com- plex knowledge from many different intellectual perspectives, tailored to the context of its occurrence). This should instill an epistemological belief structure appropriate for ill- structured domains and provide a repertoire of flexible knowledge representations that can be used in constructing as- semblages of knowledge, taken from here and from there, to fit the diverse future cases of knowledge application in that domain.

CONSTRAINTS ON THE DESIGN OF HYPERTEXT LEARNING ENVIRONMENTS DRAWN FROM IMPLICATIONS OF COGNITIVE FLEXIBILITY THEORY

Thus far we have discussed the relationship between the nature of ill- structureh knowledge domains and difficulties in the attainment of ad- vanced learning goals (mastery of complexity and transfer to new situa- tions). A principle of Cognitive Flexibility Theory was then introduced as one antidote to the problems of advanced knowledge acquisition in ill- structured domains. Now, we will briefly point to some of the ways that these cumulative considerations impinge on the design and use of hyper- text learning environments.

First the preceding discussion should make it reasonably clear that hy- pertext invironments are good candidates for promoting cognitive flexi- bility in ill- structured domains. We have referred to the need for rear- ranged instructional sequences, for multiple dimensions of knowledge representation, for multiple interconnections across knowledge compo- nents, and so on. Features like these correspond nicely to well known properties of hypertext systems, which facilitate flexible restructuring Of instructional presentation sequences, multiple data codmgs, and multiple linkages among content elements. It appears straightforward that a non- linear medium like hypertext would be very well suited for the kinds of “landscape criss- crossings” recommended by Cognitive Flexibility Theory (and needed in ill- structured knowledge domains; see also Bednar, Cun-

P’ ningham, Duffy, & Perry, chapter 2). However, it is not that easy. Implementing Cognitive Flexibility The-

ory is not a simple matter of just using the power of the computer to “connect everything with everything else.” There are many ways that hy-

68 Spiro et al. pertext systems can be designed, and there is good reason to believe that a large number of those do not produce successful learning outcomes (e. g., because they lead the learner to become lost in a confusing labyrinth of in- cidental or ad hoc connections). What is needed is the discipline of grounding hypertext design in a suitable theory of learning and instruc-

tion. That is what we have done in several prototype hypertext systems derived from Cognitive Flexibility Theory and tailored to the known ob- stacles to advanced learning in difficult and ill- structured domains (Spiro et al., 1988; Spiro & Jehng, 1990). To provide some idea of how theory in- forms design, consider just one very simple example of a hypertext design decision that responds to an aspect of Cognitive Flexibility Theory- based logic discussed in the last section: rearrangement of the presentation se- quence of content (that has been investigated previously), in order to pro- duce different understandings when that content is “re- read.”

Illustrating the Theory- and Context- Based Logic of Hypertext Design Because of the feature of conceptual instability in ill- structured domains (i. e., the same conceptual structure takes on many more meanings across instances of its use than in well- structured domains), Cognitive Flexibility Theory dictates, as discussed in the last section, that one kind of instruc- tional revisiting should produce an appreciation in the learner of the va- rieties of meaning “shades” associated with the diversity of uses. As Wittgenstein argued (1953), the meaning of ill- structured concepts is in their range of uses, rather than in generally applicable definitions- there is no simple “core meaning.” We extend Wittgenstein’s claim to larger units than the individual concept (e. g., complex conceptual structures such as a theme of a literary work). So, a feature built into our hypertexts is conceptual structure search: Content is automatically re- edited to pro- duce a particular kind of “criss- crossing” of the conceptual landscape that visits a large set of case examples of a given conceptual structure in use. The learner then has the option of viewing different example cases in the application of a concept he or she chooses to explore. That is, the instruc- tional content is re- edited upon demand to present just those cases and parts of cases that illustrate a focal conceptual structure (or set of concep- tual structures). Rather than having to rely on sporadic encounters with real cases that instantiate different uses of the concept, the learner sees a range of conceptual applications close together, so conceptual variability can easily be examined. Learning a complex concept from erratic expo- sures to complex instances, with long periods of time separating each en- counter, as in natural learning from experience, is not very efficient. When ill- structuredness prevents telling in the abstract how a concept should be used in general, it becomes much more important to show to- gether the many concrete examples of uses. In sum, a hypertext design feature is incorporated as a response to a learning difficulty caused by a

a 5. Cognitive Flexibility, Constroctivism, and Hypertext 69

characteristic of ill- structured knowledge domains. (Of course, the issue of example selection and sequencing in concept instruction has been dealt with before, e. g., Tennyson & Park, 1980. What is novel about the present approach is the particular way that this issue is addressed and the kinds of higher- order conceptual structures that are studied. Even more important is the fact that that single issue is addressed within a larger, integrative framework. That is, the treatment of conceptual variability is just one as- pect of a complete approach in which the diverse aspects are theoretically united.)

Following this same kind of logic, we will sketch briefly some of the other ways that hypertext design features can be made to match the goals of advanced learning- under the constraints of domain ill- structuredness and according to the tenets of Cognitive Flexibility Theory. For this pur- pose we will use one of our Cognitive Flexibility Hypertext prototypes, “Exploring Thematic Structure in Citizen Kane ” (“ KANE,” for short- Knowledge Acquisition in Nonlinear Environments; see Spiro and Jehng, 1990, for details), which teaches processes of literary interpretation in a post- structuralist mode (e. g., Barthes, 1967).

KANE is a learning environment that goes beyond typical instructional approaches to literary interpretation that too often settle on a single, inte- grative understanding (“ The theme of Citizen Kane is X”). Instead, stu- dents are shown that literary texts (in this case a videodisc of a literary film) support multiple interpretations, the interpretations combine and interact, they take on varying senses in different contexts, and so on. For example, the issue of conceptual variability that was discussed earlier is addressed by providing an option that causes the film to be re- edited to show just those scenes that illustrate any selected conceptual theme of the film (e. g., ‘Wealth Corrupts,” ” Hollow, Soulless Man,” etc.). Using this option, the learner could, for example, see five scenes in a row, taken from various places in the film, that illustrate different varieties or “flavors” of the ‘Wealth Corrupts” theme. Each scene essentially forms a miniature case of the Kane character’s behavior that illustrates the targeted theme. (Although the student is assumed to have already seen the film one or more times- this is advanced knowledge acquisition for Citizen Kane- the nonlinear presentation may still occasionally confuse. Therefore,. to deal with this and other kinds of out- of- sequence criss- crossings, a design feature of Cognitive Flexibility Hypertexts is the provision of optional background information on the contexts immediately preceding the one being explored.) Because of the inability of abstract definitions (as might be construed for a theme such as “Soulless Man”) to cover Conceptual mean- ings- in- use in ill- structured domains, supplementary guidance about the way meaning is used in a particular situation (Brown, Collins, & Duguid, 1989) is required. This is provided for in KANE by giving the learner the option of reading an expert commentary on the special shade of meaning associated with the conceptual theme, as applied to a scene, immediately after the scene is viewed. These functional and context- sensitive

Spiro et al. (particularized) definitions explain why the scene is considered to be a case of a theme, such as “Wealth Corrupts.” Note that a particularized repre- sentation of meaning is not the same as a dictionary sense of a word: The

latter refers to different subtypes of a word’s meaning, but with an im- plied similarity or overlap across instances of the same type- so there is less need to tailor to the individual case; in contrast, particularizing, as we mean it, implies a representation of a concept that is necessarily expressed in terms of an instance of usage (case, example, scene, occasion of use), as required in an ill- structured domain. Commentaries also include infor- mation about knowledge access: what cues in the case context should pro- vide a “tip- off” that a particular concept might be relevant for analyzing a case- if one cannot access relevant conceptual information in memory, this knowledge will not be useful on subsequent occasions.

The commentaries also provide cross- references to other instantiations of the conceptual structure that constitute an instructionally efficacious set of comparisons (e. g., other cases/ scenes in which either a roughly similar or saliently different particularized sense of that conceptual theme occurs). The guiding commentaries also include another important kind of cross- reference, namely to other conceptual themes that have interpretive rele- vance in accounting for the same case of Kane’s behavior, concepts that in- teract with and influence the meaning of each other in that scene. (Note that these different kinds of cross- references counter the reductive tenden- cies toward compartmentalization of concepts and their cases of applica- tion that we have found to be harmful in advanced learning.) Thus there is a double particularization in Cognitive Flexibility Hypertexts: The generic conceptual structure is particularized not only to the context of a specific case, but also to the other concepts simultaneously applicable for analyzing that case. That is, each case or example is shown to be a complex entity requiring for its understanding multiple conceptual representations, with the role of non- additive conceptual interdependencies highlighted.

Each of the conceptual themes used in KANE is itself a wide- scope in- terpretive schema that has been argued for in the secondary literature on the film as being the most important theme for understanding the charac- ter of Kane. In reality, however, an ill- structured domain has no single schema that is likely to cover everything of interest for an individual case, nor is any schema/ theme/ concept likely to dominate across a wide range of cases. Therefore, the greater the number of such broad- gauge schemata that are available (and KANE provides ten), the greater the utility for un- derstanding in two senses. First, there will be adequate coverage of the complexity of an individual case by an appropriately diverse set of schemata (something which is also modeled in KANE by the simultane- ous display of all the relevant conceptual themes in each scene). Second, the likelihood is increased that the most apt set of conceptual schemata will be cognitively available for understanding any one of the highly di- verse new cases that will be encountered in an ill- structured domain- the more conceptual structures there are to choose from, each a powerful

5, Cognitive Flexibility, Constructivism, and Hypertext 71 schema itself and each taught in its complex diversity of patterns of use, the greater the chance that you will find a good fit to a given case. A re- lated virtue is that configurations of combinations of conceptual structures are thereby demonstrated; since multiple conceptual representations will be required for each instance of knowledge application, the ability to corn- ;, bine conceptual entities and to recognize common patterns of their com- bination is crucial. The process of situation- specific knowledge construe- + tion, so important for transfer in ill- structured domains, is thus supported in at least two important ways: The processes of adaptive knowledge as- sembly are demonstrated, and the flexible knowledge structures required for this assembly are acquired. Furthermore, as users of the program shift over time into more of a “free exploration” mode, where they indepen- dently traverse the themes of the film in trying to answer questions of in- terpretation (posed by teachers or themselves), their active participation in learning the processes of knowledge assembly increases.

Flexible tools for covering content diversity and for teaching knowl- edge assembly combine to increase the resources available for future trans- fer/ application of knowledge (e. g., interpreting a scene that has not yet been viewed or assembling prior knowledge to facilitate comprehension of a critique written about the film). By making many potential combina- tions of knowledge cognitively available- either by retrieval from mem- ory or by context- sensitive generation- the learner develops a rich palette to paint a knowledge structure well fit to helping understand and act upon a particular case at hand. This is especially important in an ill- structured domain because there will be great variety in the demands on background knowledge from case to case (and with each case individually rich in the knowledge blend required).

This discussion could continue for many other features of hypertext learning environments that are specifically derived from Cognitive Flexi- bility Theory. What would be in common across any such discussion is that each feature could be shown to have the following purpose: to counter an advanced learning difficulty endemic to ill- structured do- mains.

CONCLUDING REMARKS We have just discussed a few of the many kinds of revisitings of instruc- tional content in rearranged contexts that are implied by Cognitive Flexi- bility Theory and embodied in our hypertext systems. However, our goals in this chapter were necessarily limited. Our purpose was merely to begin to illustrate the way design features of a particular kind of computer learn- ing environment are related to cognitive and instructional theories that are themselves based on the problems posed by the interaction of learning objectives and characteristics of ill- structured knowledge domains. That is, our intention was to illustrate a way of thinking about the design of hy-

72 Spiro et al. pertext learning environments that is sensitive to and dependent upon the cognitive characteristics necessary for advanced knowledge acquisition in ill- structured domains. In particular, these are the characteristics of the “new constructivism” that we discussed earlier and that are properties of Cognitive Flexibility Theory. The realm of constructive processes must be taken beyond the retrieval of knowledge structures from memory (for the purpose of “going beyond the information given” in some learning situa- tion), to also include the independent, flexible, situation- specific assembly of the background knowledge structures themselves.

In sum, we consider our work to be moving toward a systematic the- ory of hypertext design to provide flexible instruction appropriate for de- veloping cognitive flexibility. We have called the instructional theory that is derived from Cognitive Flexibility Theory and applied in flexible

computer learning environments Random Access Instruction. It, and the developing hypertext theory, is laid out in considerable detail in Spiro and Jehng (1990). We are encouraged so far about the robustness, systematicity, and generality of our hypertext design principles, in that they have been applied in very similar ways to develop hypertext prototypes in domains as diverse as cardiovascular medicine, literary interpretation, and military strategy. Preliminary data on the effectiveness of these Cognitive Flexibil- ity Hypertexts is also encouraging. For example, Jacobson and Spiro (1991a) investigated two different design approaches for structuring a hy- pertext learning environment to provide instruction in a complex and ill- structured domain (the social impact of technology). The results of this experiment revealed that while the design which emphasized the mastery of declarative knowledge led to higher performance on measures of memory for presented facts, the design based on Cognitive Flexibility The- ory (which highlighted different facets of the material by explicitly demon- strating critical interrelationships between abstract and case- centered knowledge components, in multiple contexts on different passes through the same content) promoted superior transfer to a new problem- solving situation. More empirical testing is clearly required, and numerous other issues of hypertext design remain to be discussed. However, those are sto- ries for another time.

ACKNOWLEDGMENTS The research reported in this chapter was supported in part by the Basic Research Office of the Army Research Institute (MDA903- 86- K- 0443) and the Office of Educational Research and Improvement (OEG0087- ClOOl). Some of the background research on the learning and understanding of complex conceptual material was supported in part by the Office of Naval Research, Cognitive Science Division (N00014- 87- G 0165, N00014- 88- K- 0077). The chapter does not necessarily reflect the views of these agencies.

5. Cognitive Flexibility, Constructivism, and Hypertext 73 We would like to express our gratitude to Susan Ravlin for her first- rate programming work on the Cognitive Flexibility Hypertexts, and to Tom Duffy and Jane Adami for several very helpful comments on an earlier draft of the chapter.

REFERENCES Ausubel, D. P. 1968). Educational psychology: A cognitive view. New York:

Holt, Rinehart & Winston. Barthes, R. (1967). S/ Z. New York: Hill, Wang. Bartlett, F. C. (1932). Remembering. Cambridge, England: Cambridge Uni-

versity Press. Bereiter, C. (1985). Toward a solution of the learning paradox. Reviao of

Educ& ional Research, 55, 201- 226. Bereiter, C., & Scardamalia, M. (1985). Cognitive coping strategies and the

problem of “inert knowledge. ” In S. F. Chipman, J. W. Segal, & R. Glaser (Eds.), Thinking and learning skills: Current research and 0pe. n questions (Vol. 2, pp. 65- 80). Hillsdale, NJ: Lawrence Erlbaum AssOCl-

BraiFfkrd, J. D., Franks, J. J., Vye, N. J., & Sherwood, R. D. (1989). New ap- proaches to instruction: Because wisdom can’t be told. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 470- 497). Cambridge, England: Cambridge University Press. Bransford, J. D., & Johnson, M. K. (1972). Contextual prerequisites for un-

derstanding: Some investigations of comprehension and recall. lournal of Verbal Learning and Verbal Behavior, 12, 717- 726. Brown, A. (1989). Analogical learning and transfer: What develops? In S.

Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 369- 412). Cambridge, England: Cambridge University Press. Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the

culture of learning. Educational Researcher, 18, 32- 42. Bruner, J. S. (1963). The process of education. Cambridge, MA: Harvard

University Press. Burstein, M. H., & Adelson, B. (1990). Issues for a theory of analogical

learning. In R. Freedle (Ed.), Artificial intelligence and the future Of testing. Hillsdale, NJ: Lawrence Erlbaum Associates. Chase, W. C., & Simon H. (1973). Perception in chess. Cognitive Psychol-

ogy, 4, 55- 81. Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and repre-

sentation of physics problems by experts and novices. Cognitive Scr- ence, 5, 121- 152. COulson R. L., Feltovich, P. J., & Spiro, R. J. (1989). Foundations of a mis-

undeistanding of the ultrastructural basis of myocardial failure: A re- ciprocation network of oversimplifications. The Journal of Medicine

74 Spiro et al. and Philosophy (special issue on “The Structure of Clinical Knowl- edge”), 24,109- 146. Feltovich, I’., Johnson, P. E., Moller, J. H., & Swanson, D. B. (1984). LCS:

The role and development of medical knowledge in diagnostic exper- tise. In W. J. Clancey & E. H. Shortliffe (Eds.), Readings in medical arti- ficial intelligence: The first decade. Reading, MA: Addison Wesley. Feltovich, P. J., Spiro, R. J., & Coulson, R. L. (1989). The nature of concep-

tual understanding in biomedicine: The deep structure of complex ideas and the development of misconceptions. In D. Evans & V. Pate1 (Eds.), Cognitive science in medicine: Biomedical modeling. Cam- bridge, MA: MIT (Bradford) Press. Feltovich, P. J., Spiro, R. J., & Coulson, R. L. (in press). Learning, teaching,

and testing for complex conceptual understanding. In N. Frederiksen, R. Mislevy, & I. Bejar (Eds.), Test theory for a new generation of tests. Hillsdale, NJ: Lawrence Erlbaum Associates. [GPEP] Panel on the General Professional Education of the Physician.

(1984). Physicians for the twenty- first century: The GPEP report. Wash- ington, DC: American Association of Medical Colleges. Jacobson, M. J., & Spiro, R. J. (1991a). Cognitive flexibility, hypertext learn-

ing environments, and the acquisition of complex knowledge: An em- pirical investigation. Manuscript submitted for publication. Jacobson, M. J., & Spiro, R. J. (1991b). A framework for the contextual anal-

ysis of computer- enhanced learning environments. Manuscript sub- mitted for publication. Merrill, M. D. (1983). Component display theory. In C. M. Reigeluth (Ed.),

Instructional- design theories and models: An overview of their cur- rent status (pp. 282- 333). Hillsdale, NJ: Lawrence Erlbaum Associates. Myers, A. C., Feltovich, P. J., Coulson, R. L., Adami, J. F., & Spiro, R. J.

(1990). Reductive biases in the reasoning of medical students: An in- vestigation in the domain of acid- base balance. In B. Bender, R. J. Hiemstra, A. J. J. A. Scherbier, & R. P. Zwierstra (Eds.), Teaching and as- sessing clinical competence. Groningen, the Netherlands: BoekWerk Publications. Perkins, D. N., & Simmons, R. (1989). Patterns of misunderstanding: An

integrative model for science, math, and programming. Review of Ed- ucational Research, 58, 303- 326. Spiro. R. J. (1980). Constructive processes in prose comprehension and re-

call. In R. J. Spiro, B. C. Bruce, & W. F. Brewer (Eds.), Theoretical issues in reading comprehension (pp. 245- 278). Hillsdale, NJ: Lawrence Erl- baum Associates. Spiro, R. J., Coulson, R. L., Feltovich, P. J., & Anderson, D. K. (1988). Cogni-

tive flexibility theory: Advanced knowledge acquisition in ill- struc- tured domains. In The tenth annual conference of the cognitive science society. Hillsdale, NJ: Lawrence Erlbaum Associates. Spiro, R. J, Feltovich, P. J., Coulson, R. L., & Anderson, D. K. (1989). Multi-

ple analogies for complex concepts: Antidotes for analogy- induced mis- 5. Cognitive Flexibility, Constroctivism, and Hypertext 75

conception in advanced knowledge acquisition. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 498- 531). Cam- bridge, England: Cambridge University Press. Spiro, R. J., & Jehng, J. C. (1990). Cognitive flexibility and hypertext: Theory

and technology for the nonlinear and multidimensional traversal of complex subject matter. In D. Nix & R. J. Spiro (Eds.), Cognition, educa- tion, and multimedia: Exploring ideas in high lechnology (pp. 163- 205). Hillsdale, NJ: Lawrence Erlbaum Associates. Spiro, R. J., Vispoel, W., Schmitz, J., Samarapungavan, A., & Boerger, A.

(1987). Knowledge acquisition for application: Cognitive flexibility and transfer in complex content domains. In B. C. Britton (Ed.), Executive control processes (pp. 177- 200). Hillsdale, NJ: Lawrence Erlbaum Asso- cia tes. Tennyson, R. D., & Park, 0. (1980). The teaching of concepts: A review of

instructional design research literature. Review of Educational Re- search, 50, 55- 70. Whitehead, A. N. (1929). The aims of education. New York: Macmillan. Wittgenstein, L. (1953). Philosophical investigations. New York: Macmil-

lan.