Semantic cognition 1 Running head: SEMANTIC COGNITION Précis of Semantic Cognition: A Parallel Distributed Processing Approach

In this précis of our recent book, we present a parallel distributed processing theory of the acquisition, representation and use of human semantic knowledge. The theory proposes that semantic abilities arise from the flow of activation amongst simple, neuron-like processing units, as governed by the strengths of interconnecting weights; and that acquisition of new semantic information involves the gradual adjustment of weights in the system in response to experience. These simple ideas explain a wide range of empirical phenomena from studies of categorization, lexical acquisition, and disordered semantic cognition. In this précis we focus on phenomena central to the reaction against similarity-based theories that arose in the 1980’s and that subsequently motivated the “theory-theory” approach to semantic knowledge. Specifically, we consider i) how concepts differentiate in early development, ii) why some groupings of items seem to form “good” or coherent categories while others do not, iii) why different properties seem central or important to different concepts, iv) why children and adults sometimes attest to beliefs that seem to contradict their direct experience, v) how concepts reorganize between the ages of 4 and 10, and vi) the relationship between causal knowledge and semantic knowledge. The explanations for these phenomena are illustrated with reference to a simple feed-forward connectionist model; and the relationship between this simple model, the broader theory, and more general issues in cognitive science are discussed. Semantic cognition 3 Précis of Semantic Cognition: A Parallel Distributed Processing Approach When we open our eyes and look around us we observe a host of objects — people, animals, plants, cars, buildings, and other artifacts of many different kinds — most of which are quite familiar. We have tacit expectations about the unseen properties of these objects — for example, what we would find underneath the skin of an orange or banana — and how the objects would react or what effects they would have if we interacted with them in various ways. Would a furry animal bite if we tried to stroke it? Would a particular artifact hold a hot liquid? We can usually name these objects, describe their visible and invisible properties to others, and make inferences about them, such as whether they would likely die if deprived of oxygen, or whether they would break if dropped onto a concrete floor. Understanding the basis of these abilities—to recognize, comprehend, and make inferences about objects and events in the world, and to comprehend and produce statements about them—is the goal of research in semantic cognition. Since antiquity, philosophers have considered how we make semantic judgments, and the investigation of semantic processing was a focal point for both experimental and computational investigations in the early phases of the cognitive revolution. Yet the mechanistic basis of semantic cognition remains very much open to question. While explicit computational theories were offered in the 1960’s and into the 1970’s (Collins & Quillian, 1969; Collins & Loftus, 1975), the mid-70’s saw the introduction of findings on the gradedness of category membership and on the privileged status of some categories that these theories did not encompass (Rosch and Mervis, 1975; Rosch et al., 1976). Eleanor Rosch, who introduced most of these phenomena, eschewed any sort of explicit mechanistic theorizing (e.g. Rosch, 1978). Subsequently a new thrust of research has emerged within a framework often called ”theory Semantic cognition 4 theory” (Carey, 1985; Murphy & Medin, 1985; Gopnik & Meltzoff, 1997; Keil, 1989), which proposes that semantic knowledge is rooted in a system of implicit beliefs about the causal forces that give rise to the observable properties of objects and events. On this view, implicit and informal causal theories determine which sets of items should be treated as similar for purposes of induction and generalization; which properties are important for determining category membership; which properties will be easy to learn and which difficult; and so on. Conceptual development is viewed as arising (at least in part) from change to the implicit causal theories that structure concepts. This framework has been very useful as a springboard for powerful experimental demonstrations of the subtlety and sophistication of the semantic judgments adults and even children can make, but it has left the field without an explicit mechanistic theory of the representation and use of semantic knowledge, since the fundamental tenets of the theory theory are general principles whose main use has been to guide the design of ingenious experiments rather than the explicit formulation of computational mechanisms. In what follows we provide a précis of our recent book Semantic Cognition, which puts forward a theory about the cognitive mechanisms that support semantic abilities based on the domain general principles of the connectionist or parallel distributed processing framework. Our approach captures many of the appealing aspects of spreading-activation and prototype theories while resolving some of the apparent paradoxes they face; and it provides a mechanistic means of understanding the phenomena that have motivated theory-theory and related approaches. The book illustrates how a simple model implementation of the theory addresses, among other things, classic findings from studies of semantic cognition in infancy and childhood; the role of frequency, typicality, and expertise on semantic cognition in adulthood; and the progressive disintegration of conceptual knowledge observed in some forms of dementia. In this précis, however, we focus on phenomena that were central to the critical reaction against Semantic cognition 5 “similarity-based” theories that arose in the 1980’s and that subsequently motivated the appeal to theory-based approaches as the basis for semantic abilities. These phenomena are briefly summarized in Table 1, and are explained in further detail below. We emphasize these particular phenomena because they are often thought to challenge the notion that semantic abilities might arise from general-purpose learning mechanisms, and to support the view that such abilities must arise from initial domain-specific knowledge, via domain-specific learning systems. These issues are central to questions about what makes us uniquely human. Do we possess, at birth, and by virtue of evolution, a set of highly specialized cognitive modules tailored to support knowledge about particular domains? Or do our advanced semantic abilities reflect the operation of a powerful learning mechanism capable of acquiring, through experience, knowledge about all semantic domains alike? A key point of our book is that the learning mechanisms adopted within the connectionist approach to cognition are quite different from classical associationist learning; that the capabilities of connectionist models have been under-appreciated in this respect; and that such models can provide an intuitive explanation of how domain-general learning supports the emergence of semantic and conceptual knowledge over the course of development. The models we describe employ domain-general learning mechanisms, without initial knowledge or domain-specific constraints. Thus, if they adequately capture the phenomena listed in Table 1, this calls into question the necessity of invoking initial domain-specific knowledge to explain semantic cognition. The material below is largely exurpted from our book, with some restructuring, condensation and minor corrections. In the interest of providing a relatively succinct overview of the theory, we have omitted substantial detail, both in the range of phenomena to which the model has been applied and in the descriptions of the simulations themselves. Where we feel these details may prove especially useful, we refer the reader to the corresponding section of the Semantic cognition 6 book. Also, we have avoided adding new material reviewing more recent work by ourselves or others; we hope to address such findings in our our response to open peer commentary.