«3 Three Types of Conceptual Change: Belief Revision, Mental Model Transformation, and Categorical Shift Michelene T. H. Chi University of Pittsburgh ...»
Chi, M.T.H. (2008). Three types of conceptual change: Belief revision,
mental model transformation, and categorical shift. In S.
Vosniadou (Ed.), Handbook of research on conceptual change
(pp. 61-82). Hillsdale, NJ: Erlbaum.
Three Types of Conceptual Change:
Belief Revision, Mental Model
Transformation, and Categorical Shift
Michelene T. H. Chi
University of Pittsburgh
CONCEPTUAL CHANGE KIND OF LEARNING
Although this definition of conceptual change appears straightforward, conceptual change kind of learning entails several complex, non-transparent, and interleaved issues. Some of the key non-transparent ideas are: (a) In what ways is knowledge misconceived? (b) Why is such misconceived knowledge often resistant to change? (c) What constitutes a change in prior knowledge? and (d) Ho should instruction be designed to promote conceptual change? The existence of decades of research on conceptual change speaks to the complexity of these issues.
This chapter hopes to add clarity to some of these issues by laying out three different grain sizes in which knowledge can be “in conflict with” the to-be-learned materials, postulating for each grain size the processes by which such “in conflict with knowledge” can be changed, and
“CONCEPTS” AND “CATEGORIES” IN CONCEPTUAL CHANGE In this section, we elaborate on (1) the scope of the term “concepts” in conceptual change research, (2) the assumptions about the role of categorization in learning and conceptual change, and (3) the relationships among different levels and kinds of “categories.”
Scope of Concepts
Several decades of psychological literature (see Medin & Rips, 2005, for a recent review; and see Jackendoff along with the Forum published in Mind & Language, 1989, for a broader view) have dealt with determining how concepts can be identified and defined. That classic literature has typically been devoted to defining isolated and static concepts and categories such as robins and birds. From that literature, we adopt the common assumptions that a concept has several perceptual features and conceptual attributes, and a concept can be viewed as belonging to some category. For example, a robin has a red breast (a perceptual feature), lives in a temperate climate (more of a conceptual attribute), and belongs to the category of birds. (Throughout this chapter, we will use the term features to refer to perceptual properties, “attributes” to refer to conceptual properties, and italicize category terms and scientific concepts.) Although prior conflicting ideas are often referred to as misconceptions, and learning that involves altering such incorrect ideas is referred to as conceptual change, the grain size of that prior knowledge does not have to be at the level of a concept, in the traditional sense of static concepts typically studied by psychologists, such as chairs and furniture. Even though psychologists have begun to expand the notion of a category beyond concrete static types to include explanation-based categories such as food items for a diet (e.g., popcorn, diet soda, lean turkey, Barsalou, 1983) or principle-based categories (such as physics problems that share the same principle, Chi, Feltovich, & Glaser, 1981), the kind of misconceived knowledge in subject matter domains taught in schools (especially science domains) are at a much larger grain size, more complex and inter-related. For example, students are expected to learn about systems (such as the circulatory system) consisting of many inter-related components (such as blood, organs, etc). Students are also expected to learn not only about static concepts, but also about dynamic concepts, such as the processes of heat transfer and natural selection. In short, the term “concepts” in conceptual change research often refers to a broader scope than isolated and static concepts.
ROLE OF HIERARCHICAL CATEGORIZATION IN LEARNING
Categorizing is the process of identifying or assigning a concept to a category to which it belongs. One of the most important assumptions about categorizing that we also adopt is its role in learning (Bransford, Brown, & Cocking, 1999). Categorization is an important learning mechanism because a concept, once categorized, can “inherit” features and attributes from its category membership. For example, we can infer that robins lay eggs even if we were never told that fact, as long as we know that robins are birds and birds lay eggs. By knowing that robins are a kind of bird allows us to infer that robins inherit the properties of birds. Thus, categorizing, or assigning a concept to a correct category, is powerful because a learner can use knowledge of the category to make many inferences and attributions about a novel concept/phenomenon (Medin & Rips, 2005). Even young children can do this. For example, 4to 7-year-old dinosaur aficionados can generate many appropriate inferences about an unfamiliar dinosaur once they have categorized it on the basis of surface features (Chi & Koeske, 1983;
Gobbo & Chi, 1986).
3. THREE TYPES OF CONCEPTUAL CHANGE 63 Besides the common assumption that categorization allows new concepts to inherit categorical properties, we propose two additional assumptions about the role of categorization in learning. The first new assumption is that when learners have no obvious basic category to assign a new concept or phenomenon — they will assign it to the next higher level of category that is appropriate. For example, suppose an observer in a museum sees a strange large creature (a gavial) with four short legs, scaly skin and a flat bill-like snout. Not knowing that it’s a kind of reptile, like a crocodile, the observer would categorize it at the next level up, as a kind of animal, since it appears to have the properties of animals, can move on its own, eat, and so forth. (The second new assumption will be described in the next section.) As illustrated above, the type of relationships cognitive psychologists have explored about inheritance of properties are hierarchical ones. Hierarchical relationships among categories are primarily inclusive in nature. For example, living beings include animals, and animals include reptiles and birds, and birds include robins. (See Figure 3.1, left-most hierarchical tree.) Living beings, in turn, can be subsumed under an even higher category, such as objects; and objects can be subsumed under yet an even higher category such as Entities. The classic psychological research that dealt mostly with hierarchical relationships among categories asked questions such as: What level within this hierarchy is the most “basic” and useful? How does correct categorization support reasoning and inferencing? Can priming the correct super-ordinate category enhance recognition?
Little research has focused on incorrect hierarchical categorization, perhaps because it is not wrong but merely too specific or overly general. As in the preceding gavial example, the overly general hierarchical categorization of gavial as an animal is not that damaging, since the observer can still benefit from correct inferences and attributions inherited from the animal category. For example, the observer can understand new instruction about gavials, such as that they breathe air through their snouts. The observer can assimilate this new piece of information because it is compatible with what s/he knows about animals in general. Therefore, categorizing a concept at a higher categorical level is not damaging to learning.
Lateral and Ontological Categories Research in cognitive psychology has paid much less attention to the role of “lateral” (rather than hierarchical) categories. For example, artifacts can be considered a lateral category more-orless “parallel” to living beings (see Figure 3.1) Artifacts does not include the subcategories of living beings, such as animals, reptiles, birds, or robins. Instead, artifacts includes a different set of subcategories, such as furniture and toys, and furniture includes subcategories such as tables and chairs (see Figure 3.1) In short, artifacts and living beings can be thought of as occupying different branches of the same hierarchical tree (Thagard, 1990), in this case the Entities tree. We will refer to categories on different branches as “lateral” (vs. hierarchical) categories and, when lateral categories occur at about the same level within a tree, we will refer to them as “parallel.” Although artifacts and living beings can both be subsumed under the higher-level category of objects and therefore share higher-level properties of objects such as “has shape” and “can be thrown,” the properties of artifacts and living beings tend to be distinct and mutually exclusive.
For example, living beings “can move” on their own volition, whereas artifacts cannot; living beings "can reproduce” whereas artifacts cannot. (Examples of properties of each category are shown in quotes in Figure 3.1.). Gelman (1988) and Schwartz (1977) might have referred to these categories as different in “kind..
Having mutually exclusive properties means that it does not make sense to talk about a concept of one category as having a property from a lateral category. Conversely, a concept can be 64 CHI
described as having a property of its own category whether or not it is true. For example, living beings can reproduce whereas artifacts cannot. This means that Fido (a dog), being a living being, has the potential to reproduce even though Fido (a specific dog), having been neutered, cannot. On the other hand, a toy dog (an instance of an artifact) does not have this potential. Thus, it makes sense to say that Fido will have grey puppies even though Fido cannot have puppies, but it does not make sense to say that the mechanical toy dog will have puppies. Thus, a property of a category can be applied to members of that category or its subcategories, whether or not it is true, whereas it cannot be applied to a member of a lateral category. Thus, having mutually exclusive properties means that it does not make sense to talk about a concept as having a property of a lateral category, whereas it does make sense to talk about a concept as having a property of its own category even if it is false.
To take another example, a object such as a piece of clear glass, being an Entity, can have the property of “color,” even though a specific piece of glass is colorless. That is, it is acceptable and sensible to say “the glass is green” even though it is not, whereas it makes no sense to
3. THREE TYPES OF CONCEPTUAL CHANGE 65 say “the baseball game is green.” This is because a baseball game, being a direct process, which is a category on an alternative tree, cannot take on the property of “color,” so that it does not make sense to say that “the baseball game is green.” Thus, one way to determine that two categories are laterally distinct (either within the same tree or between trees) is to use such a sensibility judgment task (Keil, 1979). Although in the past, we and others have called such lateral categories ontological (Chi, 1997; Keil, 1981), we reserve the term ontologies to refer to categories between different trees (as shown in Figure 3.1), since categories on different trees never share any properties, given that they do not share any super-ordinate level categories. For example, Entities have properties such as “can be contained” and “has volume,” whereas Processes have properties such as “occurring over time.” Thus, no process, whether it’s an event such as a baseball game, a procedure such as baking a cake, or a state change such as melting, can have the property of “has volume,” “has color,” or “can be contained,” whereas no entity, such as a cake or a ball, can have the property of “lasting two hours.” Thus, each tree might be considered an “ontology,” (and its name will be capitalized) (Chi, 1997, 2005), in that the trees refer to a system of taxonomic categories for certain existences in the world, as defined by philosophers (Sommers, 1971). Thus, in this chapter, we will refer to categories that occupy different trees as different “ontologically”, and categories that occupy parallel branches within a tree as different laterally or in “kind.” The goal of our research is not to lay out the exact ordering and structure of hierarchical and lateral categories and trees, nor to decide which categories deserve the name ontology, or how many kinds or ontologies there are. The nature of categorical structure is an epistemological issue. Our goal instead is to focus on the role of lateral and ontological categories in conceptual change kind of learning. Thus, Figure 3.1 is offered merely as an example of a crude and intuitive rendition of categorical structures. It is by no means the absolute or the correct one.