Abstract not found

Children learn a variety of verbs for hand actions starting in their second year of life. The semantic distinctions can be subtle, and they vary across languages, yet they are learned quickly. Howis this possible? This dissertation explores the hypothesis that to explain the acquisition and use of action verbs, motor control must be taken into account. It presents a model of embodied semantics|based on the principles of neural computation in general and on the human motor system in particular|which takes a set of labelled actions and learns both to label novel actions and to obey verbal commands. Akey feature of the model is the executing schema, anactivecontroller mechanism which, by actually driving behavior, allows the model to carry out verbal commands. A hard-wired mechanism links the activity of executing schemas to a set of linguistically important features including hand posture, joint motions, force, aspect and goals. The feature set is relatively small and is xed, helping to make learning tractable. Moreover, the use of traditional feature structures facilitates the use of model merging, a Bayesian probabilistic learning algorithm which rapidly learns plausible word meanings, automatically determines an appropriate number of senses for each verb, and can plausibly be mapped to a connectionist recruitment

Work in philosophy and psychology has argued for a dissociation between perceptuallybased similarity and higher-level rules in conceptual thought. Although such a dissociation may be justified at times, our goal is to illustrate ways in which conceptual processing is grounded in perception, both for perceptual similarity and abstract rules. We discuss the advantages, power and influences of perceptually-based representations. First, many of the properties associated with amodal symbol systems can be achieved with perceptually-based systems as well (e.g. productivity). Second, relatively raw perceptual representations are powerful because they can implicitly represent properties in an analog fashion. Third, perception naturally provides impressions of overall similarity, exactly the type of similarity useful for establishing many common categories. Fourth, perceptual similarity is not static but becomes tuned over time to conceptual demands. Fifth, the original motivation or basis for sophisticated cognition is often less sophisticated perceptual similarity. Sixth, perceptual simulation occurs even in conceptual tasks that have no explicit perceptual demands. Parallels between perceptual and conceptual processes suggest that many mechanisms typically associated

Abstract not found

Autonomous agents make frequent use of knowledge in the form of categories --- categories of objects, human gestures, web pages, and so on. This paper describes a way for agents to learn such categories for themselves through interaction with the environment. In particular, the learning algorithm transforms raw sensor readings into clusters of time series that have predictive value to the agent. We address several issues related to the use of an uninterpreted sensory apparatus and show specific examples where a Pioneer 1 mobile robot interacts with objects in a cluttered laboratory setting.

The ability to combine words into novel sentences has been used to argue that humans have symbolic language production abilities. Critiques of connectionist models of language often center on the inability of these models to generalize symbolically (Fodor & Pylyshyn, 1988; Marcus, 1998). To address these issues, a connectionist model of sentence production was developed. The model had variables (role-concept bindings) that were inspired by spatial representations (Landau & Jackendoff, 1993). In order to take advantage of these variables, a novel dual-pathway architecture with event semantics is proposed and shown to be better at symbolic generalization than several variants. This architecture has one pathway for mapping message content to words and a separate pathway that enforces sequencing constraints. Analysis of the model’s hidden units demonstrated that the model learned different types of information in each pathway, and that the model’s compositional behavior arose from the combination of these two pathways. The model’s ability to balance symbolic and statistical behavior in syntax acquisition and to model aphasic double dissociations provided independent support for the dual-pathway architecture.

The way people think about the World-Wide Web (WWW) has implications for the way that they navigate it. In this paper, we discuss the nature of people's metaphorical conception of the WWW, as gathered from interviews with beginning and experienced web users. Based on linguistic data, we argue that people naturally think of the web as a kind of physical space in which they move, although information on the web is not physical, and web users do not actually move. Nevertheless, such metaphorical thought is motivated by the same basic image schemata that people rely on to mentally structure everyday life.

We present a new account of the fine-grained structure of semantic categories derived from neuropsychological, behavioral, and developmental data. The account places theoretical emphasis on the functions of the referents of concepts. We claim (i) that the distinctiveness of functional features correlated with perceptual features varies across semantic domains; and (ii) that category structure emerges from the complex interaction of these variables. The representational assumptions that follow from these claims make strong predictions about what types of semantic information are preserved in patients showing category-specific deficits following brain damage. These claims are illustrated with a connectionist simulation which, when damaged, shows patterns of preservation of distinctive and shared functional and perceptual information which varies across semantic domains. The data model both dissociations between knowledge for artifacts and for living things and recent neuropsychological evidence concerning the robustness of functional information in the representation of concepts. © 2000 Academic Press

Please do not cite or quote without permission. Preparation of the chapter and much of the research reported in it were supported by Grant

Recent developments in philosophy, linguistics, developmental psychology and artificial intelligence make it possible to envision a developmental path for an artificial agent, grounded in activity-based sensorimotor representations. This paper describes how Neo, an artificial agent, learns concepts by interacting with its simulated environment. Relatively little prior structure is required to learn fairly accurate representations of objects, activities, locations and other aspects of Neo's experience. We show how classes (categories) can be abstracted from these representations, and discuss how our representation might be extended to express physical schemas, general, domain-independent activities that could be the building blocks of concept formation.