Concepts
People classify objects into classes, samples into substances, and events into event types. This capacity, typically called “categorization” is a basic capacity of human cognition: without it, we would be unable to acquire any general knowledge.
Psychologists assume that when we categorize an object into a class (often called “a category” in psychology), we rely (maybe unconsciously) on some knowledge about this class.[125] Thus, when I classify an object as a table, I use some knowledge about tables. Psychologists call “concepts” those bodies of knowledge that are used by default to categorize (for an overview of the psychology of concepts, see Murphy 2002, Machery 2009; for a history of the psychology of concepts, see Machery 2007a). Importantly, in addition to categorization, concepts are also used by the psychological processes underlying other capacities, such as induction. Thus, a concept of water is a body of knowledge about water that is used by default to categorize samples as being samples of water, to reason inductively about water, and so on. Psychologists interested in concepts attempt to describe the properties of these bodies of knowledge.Philosophers have long paid attention to the psychology of concepts.[126] Famously, Fodor has argued that all the theories of concepts developed in psychology were incorrect (Fodor 1994, 1998). His favorite target has been the prototype theory of concepts (Rosch and Mervis 1975, Hampton 1979). Prototype theorists argue that a concept is a body of statistical knowledge about a class (or a substance, etc.). The simplest versions propose that a prototype is a body of knowledge about the typical properties of the members of a class. Thus, a prototype of a dog is a body of knowledge about the typical properties of dogs. Fodor's main objection against the prototype theory can be put simply.
Concepts compose. Thus, anybody who can think about dogs and about blue things can ipso facto think about blue dogs. But prototypes do not compose. Thus, concepts cannot be prototypes. To support the second premise of this argument, Fodor has put forward several considerations. The pet fish argument is the best known of these. Fodor notes that a poodle might be a prototypical pet and that a shark might be a prototypical fish, while a prototypical pet fish is a golden fish. Thus, our prototype of a pet fish is not derived from our prototype of a pet and our prototype of a fish. Rather, it is derived from our experience with pet fish. Thus, prototypes do not compose.Fodor's pet fish argument is unconvincing. The fact that the prototype of a pet fish does not result from the combination of the prototype of a pet and the prototype of a fish does not show that prototypes do not compose, when we have no experience with members of the extension of the complex concepts. That is, we might combine a prototype of an x (say, of a spy) and the prototype of a y (say, of a grandmother), when we have no experience with objects that are x and y (spy grandmothers). Experimental evidence does suggest that people produce complex prototypes in these conditions (see Murphy 2002, ch. 12 for a review). Particularly, people seem to assume that the typical properties of an x and of a y tend to be also typical of an object that is both an x and a y (Hampton 1987). Thus, a grandmother spy has the typical properties of a grandmother: she might have grey hair and wear out-of-fashion clothes (see, however, Connolly, Fodor, Gleitman, and Gleitman 2007).
Recent philosophical work on concepts has focused on two main issues: whether an empiricist theory of concepts is viable (Prinz 2002) and whether concepts form a natural kind (Machery 2005, 2009). I consider these two issues in turn. Following psychologist Lawrence Barsalou (1999), Jesse Prinz (2002) has argued that recent developments in the psychology of concepts support a view of concepts that has many affinities with David Hume's empiricist theory of ideas.
Although there are several differences between Barsalou's, Prinz's, and others' neo-empiricist theories of concepts, they all endorse the two following theses (Machery 2006b):1. The knowledge that is stored in a concept is encoded in several perceptual representational formats.
2. Conceptual processing involves reenacting some perceptual states and manipulating these perceptual states.
Thesis (1) is about how we encode our conceptual knowledge (Prinz 2002, 109). Neo-empiricists assume that each perceptual system involves a distinct representational format. Thesis (1) asserts that our conceptual knowledge is encoded in these perceptual representational formats. By contrast, amodal theorists argue that our conceptual knowledge is encoded in a representational format that is distinct from our perceptual representational formats. That is, for amodal theorists, we possess a distinct, sui generis representational format, which is used to encode our conceptual knowledge, in addition to our perceptual representational formats. This distinct representational format is usually thought of as being language-like. To illustrate this distinction, according to neo-empiricists, Marie's conceptual knowledge of apples consists of the visual, olfactive, tactile, somatosensory, and gustative representations of apples that are stored in her long-term memory. These representations are a subset of the perceptual representations of apples Marie has had in her life. According to amodal theorists, Marie's conceptual knowledge of apples consists of representations encoded in a single, distinct representational format.
Thesis (2) concerns the nature of the psychological processes underlying categorization, induction, deduction, analogy-making, linguistic comprehension, and so forth. The central insight is that retrieving a concept from long-term memory during reasoning, categorization, etc., consists in producing some perceptual representations. For instance, retrieving the concept of dog when we reason about dogs consists in producing some visual, auditory, etc., representations of dogs.
This process is called “simulation” or “reenactment.” Thinking about dogs during reasoning, thus, consists of simulating seeing, hearing, and smelling dogs. Our psychological processes consist in manipulating these reenacted percepts. Thus, according to Barsalou, when we decide whether some object has a given part, for example whether lions have a mane, we produce a visual representation of a lion and another of a mane and we match these two representations; if they do match, we decide that lions have a mane (Solomon and Barsalou 2001, 135-6).Two main lines of reply to these neo-empiricist theories of concepts have been developed in the philosophical literature. Some philosophers and psychologists have argued that neo-empiricist theories of concepts suffer from the very same problems that plagued David Hume's theory of ideas. For instance, John Sarnecki (2004) and Arthur Markman and Hunt Stilwell (2004) argue that Prinz's empiricist account of concepts cannot be applied to abstract concepts (see also Machery 2006b), while Markman and Stilwell argue that it cannot be applied to relational concepts, such as the concept of an uncle (see Prinz 2004b for a rejoinder to Sarnecki and Markman and Stilwell).
Instead of discussing the theoretical problems of the neo-empiricist theories of concepts, I have focused on the evidence for these theories, for it is supposed to favor neo-empiricism over amodal theories of concepts (Machery 2007b). Thus, Barsalou and colleagues write: “Amodal theories have been attractive theoretically because they implement important conceptual functions, such as the type-token distinction, categorical inference, productivity, and propositions. [...] Conversely, indirect empirical evidence has accumulated for modality-specific representations in working memory, long-term memory, language, and thought” (Barsalou, Simmons, Barbey, and Wilson 2003, 85-6).
The evidence for neo-empiricism is not as strong as might appear at first. Three problems plague most of the current empirical research inspired by neo-empiricism.
First, neo-empiricists erroneously assume that for a given experimental task, a single prediction can be derived on behalf of amodal theorists and can be tested experimentally. In fact, however, there are numerous competing amodal models of the psychological process involved in a given task and these competing models make different predictions about subjects' performance in this task. Because different amodal models make different predictions, it is not the case that Barsalou's and others' experimental findings are inconsistent with an amodal view of concepts in general. Rather, they are inconsistent with specific amodal models.Second, neo-empiricists have not acknowledged that amodal theorists, such as Fodor or Zenon Pylyshyn, recognize that we can use imagery to solve various problems. Thus, when one is asked to count the number of windows in one's own house, one typically visualizes one's house and counts the number of visualized windows. Because amodal theorists admit the existence and importance of imagery, they do predict that in some tasks people will simulate having perceptual states. But neo-empiricist researchers have often failed to focus on tasks for which amodal theorists would not expect people to use perceptual imagery.
Finally, neo-empiricists have not acknowledged the possibility that in some domains, or for some tasks, or, maybe, in some contexts, people might use perceptual representations, while using amodal representations in other domains, or in other tasks, or in other contexts.
The second debate among philosophers of psychology interested in concepts concerns whether the class of concepts is a natural kind. Most
psychologists working on concepts assume that concepts share many scientifically important properties and attempt to describe those properties. They assume thereby that concepts constitute a natural kind, that is, roughly, a class of entities about which numerous non-accidental, scientifically important generalizations can be made - an assumption I have called “the Natural Kind Assumption” (Machery 2005, 2009).
Against the Natural Kind Assumption, I have argued that most classes are represented by several concepts that belong to kinds that have little in common. For instance, I propose that the class of dogs is typically represented by several concepts of dog and that these concepts have few (scientifically relevant) properties. To support this proposal, I have shown that for each relevant cognitive capacity (categorization, induction, concept combination), some phenomena are best explained if one posits a first kind of concepts (namely, prototypes), other phenomena are best explained if one posits a second kind of concepts (what psychologists call “exemplars”), and yet other phenomena are best explained if one posits a third kind of concepts (what psychologists call “theories”).[127] Because these three kinds of concepts have little in common, I have concluded that concepts are not a natural kind. Furthermore, I have proposed that the notion of concept is ill-suited for a scientific psychology and that the term “concept” should be eliminated from its theoretical vocabulary, exactly as the notion of superlunear objects was eliminated from astronomy.The claims that concepts are not a natural kind and that the notion of concept should be eliminated from psychology have come under criticism. While agreeing that there are different kinds of concepts, Gualtiero Piccinini and Sam Scott (2006; see Machery 2006a for a reply) have argued that if most classes were represented by several concepts (for instance, if dogs were represented by a prototype, by a set of exemplars, and by a theory), then, contrary to the conclusion I drew, concepts would be a natural kind. For, each concept consists of several parts. For instance, the concept of dog would have three parts (one corresponding to the prototype of a dog, one corresponding to exemplars of particular dogs, one corresponding to a theory about dogs). Dan Weiskopf (Weiskopf 2009) has developed a different criticism. While highlighting the diversity of concepts, he argues that numerous generalizations can in fact be made about concepts. He concludes that eliminating “concept” from the theoretical vocabulary of psychology would prevent the formulation of numerous important generalizations.
5.