Symbol Grounding Revisited 1 the Origin of Symbol Grounding 1.1 Overview 1.2 the Symbol Grounding Problem

Symbol Grounding has originated in the domain of cognitive connectionism as an approach to a model of language acquisition.It has, however, transcended this restricted domain and its most prominent proponents now regard it as a technique which is of general importance to the connectionist-symbolic debate, especially for their integration into a common framework. This paper revisits the claims made by symbol grounders and summarizes diierent models which have been presented to date. We try to answer the question what Symbol Grounding really is all about and point to its parallels with classical logic and reasoning which have not received suucient attention so far. This paper is intended to evaluate recent developments within a speciic subarea of symbolic-connectionist integration. This subarea consists of several connectionist approaches to the problem of grounding meaning in perception through somehow connecting sensory experience to symbolic forms. It contains a description and comparison of today's approaches as well as a fundamental analysis of Symbol Grounding (SG). The aim of this paper is not, however, to present yet another model of connectionist SG and the reasons for this should become clear soon. Section 1 provides an overview of the history of Symbol Grounding and introduces the reader to areas of research within which Symbol Grounding is pursued. Section 2 presents a critical overview of more technical approaches to this problem, compares them and presents a generic SG architecture. Implications of these models which concern the question what SG actually is are presented in section 3. One of these consequences is a strong connection to symbolic (and transcendental) logic which we try to motivate in section 4. Based on a fundamental criticism of symbolic models, Stevan Harnad introduces \The Symbol Grounding Problem" in his 1990 paper. This nowadays prominent article presents the problem of how formal symbol systems can acquire a semantics which is not based on other symbol structures but on a system's own sensory experience. The problem begins with the observed fact that conventional rule based AI systems consist of speciic representational structures (symbols) which only possess meaning with respect to other symbols in the program or because a program designer assigns meaning to them. In this last case Harnad would say they are \parasitic" on the meaning in our head. Such symbol structures have no meaning in themselves, i.e. no intrinsic, only derived meaning.