We present the results of a study designed to show that dissociations between lexical and similaritybased boundary partitions for a set of items can be produced in the laboratory. This is achieved by an incremental process of learning to assign a category label to items increasingly far removed (in similarity space) from the center of that category and

During the learning of speech sounds and other perceptual

lso have direct implications for the study of the origins and evolution of language. The first issue is to establish exactly what a symbol is, by giving a clear and unambiguous definition of it. Subsequently, the process of how symbols take their meanings needs to be understood, for example by studying the symbol grounding problem. Finally, questions regarding the evolution of symbols and symbol manipulation abilities need to be addressed. Definition of a symbol The definition of a symbol is a yet open and highly debatable issue. Although it is possible to give a precise definition of a symbol in a computational symbol system, it is more difficult when we use this term in the context of language and communication systems. Historically, a semiotic distinction was made between the different constituents of a communication system: icons, indices, and symbols. This distinction, originally introduced by Peirce (1978), is based on the type of reference existing between objects and com

Studies of the categorical perception (CP) of sensory continua have a long and rich history in psychophysics. In 1977, Macmillan et al. introduced the use of signal detection theory to CP studies. Anderson et al. simultaneously proposed the first neural model for CP, yet this line of research has been less well explored. In this paper, we assess the ability of neural-network models of CP to predict the psychophysical performance of real observers with speech sounds and artificial/novel stimuli. We show that a variety of neural mechanisms is capable of generating the characteristics of categorical perception. Hence, CP may not be a special mode of perception but an emergent property of any sufficiently powerful general learning system. Neural Models of CP 3 Neural Network Models of Categorical Perception Studies of the categorical perception (CP) of sensory continua have a long and rich history. For a comprehensive review up until a decade ago, see the volume edited by Harnad (1987). ...

Neural network models of categorical perception (compression of within-category similarity and dilation of between-category differences) are applied to the symbol-grounding problem (of how to connect symbols with meanings) by connecting analog sensorimotor projections to arbitrary symbolic representations via learned category-invariance detectors in a hybrid symbolic/nonsymbolic system. Our nets are trained to categorize and name 50x50 pixel images (e.g., circles, ellipses, squares and rectangles) projected onto the receptive field of a 7x7 retina. They first learn to do prototype matching and then entry-level naming for the four kinds of stimuli, grounding their names directly in the input patterns via hidden-unit representations ("sensorimotor toil"). We show that a higher-level categorization (e.g., "symmetric" vs. "asymmetric") can learned in two very different ways: either (1) directly from the input, just as with the entry-level categories (i.e., by toil), or (2) indirectly, from boolean combinations of the grounded category names in the form of propositions describing the higher-order category ("symbolic theft"). We analyze the architectures and input conditions that allow grounding (in the form of compression/separation in internal similarity space) to be "transferred" in this second way from directly grounded entry-level category names to higher-order category names. Such hybrid models have implications for the evolution and learning of language.

We report simulations with backpropagation networks trained to discriminate and then categorize a set of stimuli.

This paper presents two experiments on categorical rhythm perception. It investigates how listeners perceive discrete rhythmic categories while listening to rhythms performed on a continuous time scale. This is studied by considering the space of all temporal patterns (all possible rhythms made up of three intervals) and how they, in perception, are partitioned into categories, i.e. where the boundaries of these categories are located. This process of categorization is formalized as the mapping from the continuous space of a series of time intervals to a discrete, symbolic domain of integer ratio sequences. The methodological frame work uses concepts from mathematics and psychics (e.g., convexity and entropy) that allow for precise characterizations of the empirical results. In the first experiment 29 participants performed an identification task with 66 rhythmic stimuli (a systematic sampling of the performance space). The results show that listeners do not just perceive the time intervals between onsets of sounds as placed in a homogeneous continuum. Instead, they can reliably identify rhythmic categories, as a chronotopic time clumping map reveals. In a second experiment the effect of metric priming was studied by presenting the same stimuli but preceded with a duple or triple meter subdivision. It is shown that presenting patterns in the context of a meter has a large effect on rhythmic categorization: the presence of a specific musical meter primes the perception of specific rhythmic patterns. 1.

This research uses grounded adaptive agents for investigating the evolutionary origins of syntactic categories, such as nouns and verbs. To analyze the sensorimotor bases of linguistic structure, the techniques of categorical perception and of synthetic brain imaging are employed. The simulation uses two different architectures for the adaptive agent's neural controller. Analyses show that the neural processing of verbs is consistently localized in the regions of the networks that perform sensorimotor integration, while nouns are associated with sensory processing areas. The general implications of such model and of the analysis techniques for adaptive behavior and language evolution research are discussed.

Neural network models of categorical perception (compression of within-category similarity and dilation of between-category differences) are applied to the symbol-grounding problem (of how to connect symbols with meanings) by connecting analog sensorimotor projections to arbitrary symbolic representations via learned category-invariance detectors in a hybrid symbolic/nonsymbolic system. Our nets are trained to categorize and name 50x50 pixel images (e.g., circles, ellipses, squares and rectangles) projected onto the receptive field of a 7x7 retina. They first learn to do prototype matching and then entry-level naming for the four kinds of stimuli, grounding their names directly in the input patterns via hidden-unit representations ("sensorimotor toil"). We show that a higher-level categorization (e.g., "symmetric" vs. "asymmetric") can learned in two very different ways: either (1) directly from the input, just as with the entry-level categories (i.e., by toil), or (2) indirectly, from boolean combinations of the grounded category names in the form of propositions describing the higher-order category ("symbolic theft"). We analyze the architectures and input conditions that allow grounding (in the form of compression/separation in internal similarity space) to be "transferred" in this second way from directly grounded entry-level category names to higher-order category names. Such hybrid models have implications for the evolution and learning of language.

ovides us with the fundamental representations that we subsequently combine and tweak. In assessing the contribution of developmental research on concepts and categories to our general understanding of human concepts, we will ask four questions. What are concepts? What is the relation between perception and concepts? What are the constraints on concept learning? What are promising future directions for research on concepts? What Are Concepts? A good starting place is Edward Smith's (1989) characterization of a concept as "a mental representation of a class or individual and deals with what is being represented and how that information is typically used during the categorization" (p. 502). It is common to distinguish between a concept and a category (e.g., Hampton & Dubois, 1993). A concept refers to a mentally possessed idea or notion, whereas a category refers to a set of entities that are grouped together. The concept dog is whatever psychological state signifies thoughts of dogs.