The Importance of Categorization 11 way that does not stifle the empirical study of the mind.We will suggest such a way in chapter 17. The PC view of rigor leads to rigor mortis in the study of categoriza- tion.It leads to a view of the sort proposed by Osherson and Smith(1981) and Armstrong,Gleitman,and Gleitman(1983)and discussed in chapter 9 below,namely,that the classical view of categorization is correct and the enormous number of phenomena that do not accord with it are either due to an"identification"mechanism that has nothing to do with reason or are minor"recalcitrant"phenomena.As we go through this book,we will see that there seem to be more so-called recalcitrant phenomena than there are phenomena that work by the classical view. This book surveys a wide variety of rigorous empirical studies of the nature of human categorization.In concluding that categorization is not classical,the book implicitly suggests that the PC view of scientific rigor is itself not scientifically valid.The result is not chaos,but an expanded per- spective on human reason,one which by no means requires imprecision or vagueness in scientific inquiry.The studies cited,for example,those by Berlin,Kay,Ekman,Rosch,Tversky,Dixon,and many others,more than meet the prevailing standards of scientific rigor and accuracy,while challenging the conception of categories presupposed by the PC view of rigor.In addition,the case studies presented below in Book II are intended as examples of empirical research that meet or exceed the pre- vailing standards.In correcting the classical view of categorization,such studies serve to raise the general standards of scientific accuracy in the cognitive sciences. The view of categorization that I will be presenting has not arisen all at once.It has developed through a number of intermediate stages that lead up to the cognitive model approach.An account of those intermediate steps begins with the later philosophy of Ludwig Wittgenstein and goes up through the psychological research of Eleanor Rosch and her associ- ates
The Importance of Categorization 11 way that does not stifle the empirical study of the mind. We will suggest such a way in chapter 17. The PC view of rigor leads to rigor mortis in the study of categorization. It leads to a view of the sort proposed by Osherson and Smith (1981) and Armstrong, Gleitman, and Gleitman (1983) and discussed in chapter 9 below, namely, that the classical view of categorization is correct and the enormous number of phenomena that do not accord with it are either due to an "identification" mechanism that has nothing to do with reason or are minor "recalcitrant" phenomena. As we go through this book, we will see that there seem to be more so-called recalcitrant phenomena than there are phenomena that work by the classical view. This book surveys a wide variety of rigorous empirical studies of the ncture of human categorization. In concluding that categorization is not classical, the book implicitly suggests that the PC view of scientific rigor is itself not scientifically valid. The result is not chaos, but an expanded perspective on human reason, one which by no means requires imprecision or vagueness in scientific inquiry. The studies cited, for example, those by Berlin, Kay, Ekman, Rosch, Tversky, Dixon, and many others, more than meet the prevailing standards of scientific rigor and accuracy, while challenging the conception of categories presupposed by the PC view of rigor. In addition, the case studies presented below in Book II are intended as examples of empirical research that meet or exceed the prevailing standards. In correcting the classical view of categorization, such studies serve to raise the general standards of scientific accuracy in the cognitive sciences. The view of categorization that I will be presenting has not arisen all at once. It has developed through a number of intermediate stages that lead up to the cognitive model approach. An account of those intermediate steps begins with the later philosophy of Ludwig Wittgenstein and goes up through the psychological research of Eleanor Rosch and her associates
CHAPTER 2 From Wittgenstein to Rosch The short history I am about to give is not intended to be exhaustive.Its purpose,instead,is to give some sense of the development of the major themes I will be discussing.Here are some of those themes. Family resemblances:The idea that members of a category may be re- lated to one another without all members having any properties in common that define the category. Centrality:The idea that some members of a category may be "better examples"of that category than others. Polysemy as categorization:The idea that related meanings of words form categories and that the meanings bear family resemblances to one another. Generativity as a prototype phenomenon:This idea ccncerns categories that are defined by a generator(a particular member or subcategory) plus rules(or a general principle such as similarity).In such cases, the generator has the status of a central,or"prototypical,"category member. Membership gradience:The idea that at least some categories have de- grees of membership and no clear boundaries. Centrality gradience:Theidea that members(or subcategories)which are clearly within the category boundaries may still be more or less central. Conceptual embodiment:The idea that the properties of certain catego- ries are a consequence of the nature of human biological capacities and of the experience of functioning in a physical and social environ- ment.It is contrasted with the idea that concepts exist independent of the bodily nature of any thinking beings and independent of their experience. Functional embodiment:The idea that certain concepts are not merely understood intellectually;rather,they are used automatically,un- consciously,and without noticeable effort as part of normal func- 12
CHAPTER 2 From Wittgenstein to Rosch The short history I am about to give is not intended to be exhaustive. Its purpose, instead, is to give some sense of the development of the major themes I will be discussing. Here are some of those themes. Family resemblances: The idea that members of a category may be related to one another without all members having any properties in common that define the category. Centrality· The idea that some members of a category may be "better examples" of that category than others. Polysemy as categorization: The idea that related meanings of words form categories and that the meanings bear family resemblances to one another. Generativity as a prototype phenomenon: This idea ccncerns categories that are defined by a generator (a particular member or subcategory) plus rules (or a general principle such as similarity). In such cases, the generator has the status of a central, or "prototypical," category member. Membership gradience: The idea that at least some categories have degrees of membership and no clear boundaries. Centrality gradience: The idea that members(orsubcategories) which are clearly within the category boundaries may still be more orless central. Conceptual embodiment: The idea that the properties of certain categories are a consequence of the nature of human biological capacities and of the experience of functioning in a physical and social environment. It is contrasted with the idea that concepts exist independent of the bodily nature of any thinking beings and independent of their experience. Functional embodiment: The idea that certain concepts are not merely understood intellectually; rather, they are used automatically, unconsciously, and without noticeable effort as part of normal func- 12
From Wittgenstein to Rosch 13 tioning.Concepts used in this way have a different,and more impor- tant,psychological status than those that are only thought about consciously. Basic-level categorization:The idea that categories are not merely organized in a hierarchy from the most general to the most specific, but are also organized so that the categories that are cognitively basic are"in the middle"of a general-to-specific hierarchy.Generalization proceeds"upward"from the basic level and specialization proceeds “downward.” Basic-level primacy:The idea that basic-level categories are functionally and epistemologically primary with respect to the fol- lowing factors:gestalt perception,image formation,motor move- ment,knowledge organization,ease of cognitive processing (learn- ing,recognition,memory,etc.),and ease of linguistic expression. Reference-point,or "metonymic,"reasoning:The idea that a part of a category (that is,a member or subcategory)can stand for the whole category in certain reasoning processes. What unites these themes is the idea of a cognitive model: -Cognitive models are directly embodied with respect to their content. or else they are systematically linked to directly embodied models. Cognitive models structure thought and are used in forming catego- ries and in reasoning.Concepts characterized by cognitive models are understood via the embodiment of the models. -Most cognitive models are embodied with respect to use.Those that are not are only used consciously and with noticeable effort. -The nature of conceptual embodiment leads to basic-level categoriza- tion and basic-level primacy. -Cognitive models are used in reference-point,or "metonymic,"rea- soning. -Membership gradience arises when the cognitive model characterizing a concept contains a scale. -Centrality gradience arises through the interaction of cognitive mod- els. Family resemblances involve resemblances among models. Polysemy arises from the fact that there are systematic relationships between different cognitive models and between elements of the same model.The same word is often used for elements that stand in such cognitive relations to one another. Thus it is the concept of a cognitive model,which we will discuss in the re- mainder of the book,that ties together the themes of this section
From Wittgenstein to Rosch 13 tioning. Concepts used in this way have a different, and more important, psychological status than those that are only thought about consciously. Basic-level categorization: The idea that categories are not merely organized in a hierarchy from the most general to the most specific, but are also organized so that the categories that are cognitively basic are "in the middle" of a general-to-specific hierarchy. Generalization proceeds "upward" from the basic level and specialization proceeds "downward ." Basic-level primacy: The idea that basic-level categories are functionally and epistemologically primary with respect to the following factors: gestalt perception, image formation, motor movement, knowledge organization, ease of cognitive processing (learning, recognition, memory, etc.), and ease of linguistic expression. Reference-point, or "metonymic," reasoning: The idea that a part of a category (that is, a member or subcategory) can stand for the whole category in certain reasoning processes. What unites these themes is the idea of a cognitive model: - Cognitive models are directly embodied with respect to their content, or else they are systematically linked to directly embodied models. Cognitive models structure thought and are used in forming categories and in reasoning. Concepts characterized by cognitive models are understood via the embodiment of the models. - Most cogniti\e models are embodied with respect to use. Those that are not are only used consciously and with noticeable effort. - The nature of conceptual embodiment leads to basic-level categorization and basic-level primacy. - Cognitive models are used in reference-point, or "metonymic," reasoning. • - Membership gradience arises when the cognitive model characterizing a concept contains a scale. - Centrality gradience arises through the interaction of cognitive models. - Family resemblances involve resemblances among models. - Polysemy arises from the fact that there are systematic relationships between different cognitive models and between elements of the same model. The same word is often used for elements that stand in such cognitive relations to one another. Thus it is the concept of a cognitive model, which we will discuss in the remainder of the book, that ties together the themes of this section
14 Chapter 2 The scholars we will be discussing in this section are those I take to be most representative of the development of these themes: -Ludwig Wittgenstein is associated with the ideas of family resem- blance,centrality,and gradience. -J.L.Austin's views on the relationships among meanings of words are both a crystalization of earlier ideas in lexicography and historical semantics and a precursor of the contemporary view of polysemy as involving family resemblances among meanings. -Lotfi Zadeh began the technical study of categories with fuzzy boundaries by conceiving of a theory of fuzzy sets as a generalization of standard set theory. -Floyd Lounsbury's generative analysis of kinship categories is an im- portant link between the idea that a category can be generated by a generator plus rules and the idea that a category has central members (and subcategories).. -Brent Berlin and Paul Kay are perhaps best known for their research on color categories,which empirically established the ideas of centrality and gradience. -Paul Kay and Chad McDaniel put together color research from anthropology and neurophysiology and established the importance of the embodiment of concepts and the role that embodiment plays in determining centrality. Roger Brown began the study of what later became known as"basic- level categories."He observed that there is a "first level"at which children learn object categories and name objects,which is neither the most general nor most specific level.This level is characterized by distinctive actions,as well as by shorter and more frequently used names.He saw this level of categorization as"natural,"whereas he viewed higher-level and lower-level categorization as"achievements of the imagination.” -Brent Berlin and his associates,in research on plant and animal nam- ing,empirically established for these domains many of the fundamen- tal ideas associated with basic-level categorization and basic-level pri- macy.They thereby demonstrated that embodiment determines some of the most significant properties of human categories. Paul Ekman and his co-workers have shown that there are universal basic human emotions that have physical correlates in facial expres- sions and the autonomic nervous system.He thereby confirmed such ideas as basic-level concepts,basic-level primacy,and centrality while demonstrating that emotional concepts are embodied
• 14 Chapter 2 The scholars we will be discussing in this section are those I take to be most representative of the development of these themes: - Ludwig Wittgenstein is associated with the ideas of family resemblance, centrality, and gradience. - J. L. Austin's views on the relationships among meanings of words are both a crystalization of earlier ideas in lexicography and historical semantics and a precursor of the contemporary view of polysemy as involving family resemblances among meanings. - Lotfi Zadeh began the technical study of categories with fuzzy boundaries by conceiving of a theory of fuzzy sets as a generalization of standard set theory. - Floyd Lounsbury's generative analysis of kinship categories is an important link between the idea that a category can be generated by a generator plus rules and the idea that a category has central members (and subcategories).. - Brent Berlin and Paul Kay are perhaps best known for their research on color categories, which empirically established the ideas of centrality and gradience. - Paul Kay and Chad McDaniel put together color research from anthropology and neurophysiology and established the importance of the embodiment of concepts and the role that embodiment plays in determining centrality. - Roger Brown began the study of what later became known as "basiclevel categories." He observed that there is a "first level" at which children learn object categories and name objects, which is neither the most general nor most specific level. This level is characterized by distinctive actions, as well as by shorter and more frequently used names. He saw this level of categorization as "natural," whereas he viewed higher-level and lower-level categorization as "achievements of the imagination." - Brent Berlin and his associates, in research on plant and animal naming, empirically established for these domains many of the fundamental ideas associated with basic-level categorization and basic-level primacy. They thereby demonstrated that embodiment determines some of the most significant properties of human categories. - Paul Ekman and his co-workers have shown that there are universal basic human emotions that have physical correlates in facial expressions and the autonomic nervous system. He thereby confirmed such ideas as basic-level concepts, basic-level primacy, and centrality while demonstrating that emotional concepts are embodied
From Wittgenstein to Rosch 15 -Eleanor Rosch saw the generalizations behind such studies of particu- lar cases and proposed that thought in general is organized in terms of prototypes and basic-level structures.It was Rosch who saw categori- zation itself as one of the most important issues in cognition.Together with Carolyn Mervis and other co-workers,Rosch established re- search paradigms in cognitive psychology for demonstrating centra- lity,family resemblance,basic-level categorization,basic-level pri- macy,and reference-point reasoning,as well as certain kinds of embodiment.Rosch is perhaps best known for developing experi- mental paradigms for determining subjects'ratings of how good an example of a category a member is judged to be.Rosch ultimately realized that these ratings do not in themselves constitute models for representing category structure.They are effects that are inconsistent with the classical theory and that place significant constraints on what an adequate account of categorization must be These scholars all played a significant role in the history of the paradigm we will be presenting.The theory of cognitive models,which we will dis- cuss later,attempts to bring their contributions into a coherent paradigm. There are some notable omissions from our short survey.Since graded categories will be of only passing interest to us,I will not be mentioning much of the excellent work in that area.Graded categories are real.To my knowledge,the most detailed empirical study of graded categories is Kempton's thoroughly documented book on cognitive prototypes with graded extensions(Kempton 1981).It is based on field research in Mexico on the categorization of pottery.I refer the interested reader to that su- perb work,as well as to Labov's classic 1973 paper.I will also have rela- tively little to say about fuzzy set theory,since it is also tangential to our concerns here.Readers interested in the extensive literature that has de- veloped on the theory of fuzzy sets and systems should consult(Dubois and Prade 1980).There is also a tradition of research in cognitive psychol- ogy that will not be surveyed here.Despite Rosch's ultimate refusal to in- terpret her goodness-of-example ratings as constituting a representation of category structure,other psychologists have taken that path and have given what I call an EFFECTS STRUCTURE INTERPRETATION to Rosch's re- sults.Smith and Medin(1980)have done an excellent survey of research in cognitive psychology that is based on this interpretation.In chapter 9 below,I will argue that the EFFECTS STRUCTURE INTERPRETATION is in gen- eral inadequate. Let us now turn to our survey
From Wittgenstein to Rosch 15 - Eleanor Rosch saw the generalizations behind such studies of particular cases and proposed that thought in general is organized in terms of prototypes and basic-level structures. It was Rosch who saw categorization itself as one of the most important issues in cognition. Together with Carolyn Mervis and other co-workers, Rosch established research paradigms in cognitive psychology for demonstrating centrality, family resemblance, basic-level categorization, basic-level primacy, and reference-point reasoning, as well as certain kinds of embodiment. Rosch is perhaps best known for developing experimental paradigms for determining subjects' ratings of how good an example of a category a member is judged to be. Rosch ultimately realized that these ratings do not in themselves constitute models for representing category structure. They are effects that are inconsistent with the classical theory and that place significant constraints on what an adequate account of categorization must be. These scholars all played a significant role in the history of the paradigm we will be presenting. The theory of cognitive models, which we will discuss later, attempts to bring their contributions into a coherent paradigm. There are some notable omissions from our short survey. Since graded categories will be of only passing interest to us, I will not be mentioning much of the excellent work in that area. Graded categories are real. To my knowledge, the most detailed empirical study of graded categories is Kempton's thoroughly documented book on cognitive prototypes with graded extensions (Kempton 1981). It is based on field research in Mexico on the categorization of pottery. I refer the interested reader to that superb work, as well as to Labov's classic 1973 paper. I will also have relatively little to say about fuzzy set theory, since it is also tangential to our concerns here. Readers interested in the extensive literature that has developed on the theory of fuzzy sets and systems should consult (Dubois and Prade 1980). There is also a tradition of research in cognitive psychology that will not be surveyed here. Despite Rosch's ultimate refusal to interpret her goodness-of-example ratings as constituting a representation of category structure, other psychologists have taken that path and have given what I call an EFFECTS = STRUCTURE INTERPRETATION to Rosch's results. Smith and Medin (1980) have done an excellent survey of research in cognitive psychology that is based 011 this interpretation. In chapter 9 below, I will argue that the EFFECTS = STRUCTURE INTERPRETATION is in general inadequate. Let us now turn to our survey