Overview of Contributions

he first papers address the epistemological sentationalism”). Evidence for the first comes e.g. status of representation and possible catego- from which has demonstrated that T rizations. These rather general consider- brain-like structures can function without explicit ations about representation are followed by contri- representations. Arguments for the second are of a butions that focus on computational approaches, philosophic nature denying the possibility of speak- especially on how meaning can be simulated in arti- ing of fixed entities of the world, and, instead, pro- ficial devices, and how meaning can be grounded. posing that the “things” are defined only with Solutions are offered on various levels, among them respect to their user and thus cannot be “mapped” symbol grounding and system-theoretical consider- onto the brain of that user. Matthias Scheutz inves- ations to representation. Various disciplines inde- tigates the ontological status of representations, and pendently arrive at emphasizing the importance of questions whether they are entities on their own; he actions for representations and the necessity to argues that whether something counts as represen- close the sensory–motor loop; examples from the tation of something else is dependent on the level of perspective of computational approaches (“embod- description. By talking about a mental structure rep- iment” into an environment), the neuroscientific resenting something in the world, one implies that perspective (“top-down” processes), the psycholog- these two things exist on two different ontological ical perspective and others are presented. Several levels. This is not true, however, since the mind and papers address the role of social interactions—in the things of the world are both entities within our particular language—as a means of stabilizing sys- cognitive experience. Thus, the problem of repre- tems, and in the of meaning. The deep sentation arises only as artifact of our describing the epistemological implications connected with the phenomenon of cognition. Daniel Hutto explores above considerations are discussed in the final the difference between non-conceptual representa- chapter of the volume dedicated to constructivist tions and conceptual representations and asks the approaches. Following the idea of a true interdisci- question whether it makes sense to define represen- plinary approach, the contributions to the various tation on the lowest level of abstraction or on the sections are organized exclusively with respect to highest level. thematic content rather than to scientific discipline. Is it possible to simulate meaning or cognition? The remainder of this overview presents the papers The issue of computation to representation is pre- and their mutual relationship in more detail. sented in the chapter “Computational Approaches”. In the first chapter, dedicated to theoretical con- The paper by Georg Schwarz draws conclusions siderations, Georg Dorffner provides an introduc- for the functioning of the brain from a computa- tion to the problem of representation by defining tional perspective. From a philosophical point of three types of representation. He shows that connec- view, William Robinson tries to address a contro- tionism solves several problems of representation versy in computational approaches to cognition by and actually helps to abandon one of these types of introducing a distinction between cognition and representation. Similarily, Alfredo Pereira investi- cognitive abilities. Robert French makes the argu- gates two types of the representations used in the ment that creating representations cannot be sepa- neurosciences, perceptual and executive processes. rated from manipulating them. Moreover, he Anthony Chemero tries to classify possible cri- emphasizes the context-dependency of linguistic tiques regarding representations. He presents two terms (one of the reasons that computational lin- types of anti-representationalism: either the repre- guistics is non-trivial matter) and, as a consequence, senting structure can be questioned (“empirical the distractive character of context-laden represen- anti-representationalism”), or the represented struc- tations. Andrew Coward proposes that natural ture can be questioned (“metaphysical anti-repre- pressures have resulted in biological brains having

Understanding Representation in the Cognitive Sciences Edited by A. Riegler, M. Peschl, and A.von Stein. Kluwer Academic/Plenum Publishers, New York, 1999 19 20 Understanding Representation in the Cognitive Sciences simple functional architectures. The type of archi- large scale cortical interactions, he derives a tecture constrains the type of representations which dynamical view of cortical representation based on are possible. interareal pattern constraints and large-scale relax- While connectionism overcame some of the prob- ation. lems derived from considering the brain as a von In a related realm, away from representations Neuman Computer, serious new questions came understood as projections of environmental entities into focus. With simulating representations in neu- onto mental structures, several disciplines indepen- ronal networks it became even more obvious that dently discovered the importance of actions for rep- the problem to be solved was that the meaning was resentations. They propose to close the sensory– put into the system by the engineer. Whether the motor loop. (For this topic of parallelisms in scien- state of the network would stand for something is tific/philosophical developments: see the comment defined by an external observer but is not inherent by Wolf Singer in the discussions, page 293). In to the system. This problem of the foundation of modeling, the grounding problem described above meaning is referred to as the grounding problem, introduced the relevance of interaction for represen- which is treated in the chapter on “Symbol Ground- tations; embodied systems that interact with their ing and Language”. Tom Ziemke provides a review environment in a sensory–motor loop are created. about the grounding problem and it’s two proposed Philosophically, constructivism and the theory of solutions. The first solution of grounding is ground- autopoietic systems demonstrated that what we per- ing a symbol to an input, which was first proposed ceive as representations of a predefined external by Stevan Harnad. The papers by Nathan Chan- world is better described as constructions that serve dler et al. and Christian Balkenius & Simon Win- to successfully generate behavior. In neuroscience, ter present such models on grounded language sys- the old concept that environmental objects are tems. As described in the paper by Ziemke, this, mapped onto neurons in a feed-forward manner is however, is only a partial solution; the real break- challenged by anatomical and physiological evi- through is only obtained when a robot is really dence. It has been shown that information does not embedded into an environment and interacting with exclusively flow from the sensors to the higher brain it. Examples of this kind of grounding are given by areas, where the representations are then thought to Mark Wexler and Ralf Möller (see below). be used for “thinking” and the final activation of Not only in classical AI but also in connection- motor programs (“bottom up”). Rather, massive ism, meaning and representations are mostly of a activity is transported the other way around (“top- static nature. Thus, although distributed, representa- down”). Thus motor activity, i.e., planned actions, tions are still very much treated like “boxes”. Cog- can be viewed as supervising the incoming signals nition, however, is dynamic and it needs time (see and thus structuring cortical representations them- also discussion, page 290). Systems theory provides selves. Psychophysical evidence, such as the fig- a framework that encompasses these issues (chapter ure–vase ambiguity, show that perception is ambig- “Cognition as a Dynamical System”). From a psy- uous, dependent on priming and expectancy. Thus, chological point of view, Pim Haselager provides a several approaches come to the conclusion that per- review about the relevance of dynamical systems ception is not a passive, process of theory to the topic of representation and its (philo- mapping but an active construction, where the need sophical) classification in the context of behavior- for action creates an hypothesis about the environ- ism. Marius Usher & Ernst Niebur present neuro- ment which is then compared to the incoming sen- scientific evidence and a model demonstrating that sory signals. The chapter “Relevance of Action for neural representations are active processes which Representation” is dedicated to these problems. can mediate contextual computation and bind rela- The idea of “the inversion of sensory processing tional properties. Ken Mogi discusses the concept by feedback pathways” was introduced by Erich of time and causality in the relation of perception Harth. In his paper, he describes neurophysiologi- and neural firing. His principle of interaction simul- cal evidence for a scenario where processing is not taneity provides an explanation for the origin of from the periphery to the higher brain areas but subjective time. Steve Bressler presents a complex- rather the other way around. He presents a model systems theory approach to representation and cog- where a cortical hypothesis (“internal sketchpad”) nition. Inspired by empirical findings of local and might be controlled by the incoming signals in a Overview of Contributions 21 hill-climbing algorithm. From the perspective of over classical AI is only a partial one, and that lan- neurobiology, Astrid von Stein interprets the corti- guage has to be included. He introduces the ideas of cal architecture with its feed-forward and feedback Bourdieu, and McDowell, who take language not as connections as a medium of interwoven sensory– a static set of words and grammatical rules but as a motor loops interacting on different levels of a hier- “way of life”, a Habitus. Habitus (Bourdieu) is “a archy. Representations do not “sit” on either end of system of dispositions which serves to set the indi- the connections, but are stable states in these sen- vidual’s understanding and actions in agreement sory–motor loops. This is consistent with Piaget’s with its social and physical environment”. Since finding that an object is defined through the per- Habitus is installed by socialization, it will be the ceived change in sensation that an action performed same within a group, i.e., the experiencing will be on the object induces. According to this approach, similar and a common world-view will develop. what we perceive as entities is neither in the world Thus, language as a Habitat has to be included into nor in the brain but in the interaction between both. models of cognitive agents. A prototypical example Based on the idea of “perception through anticipa- of representation, language and semantics is given tion”, Ralf Möller performs a simulation with a by Peter Gärdenfors. In contrast to realist theories similar approach: he presents a system where the of semantics he suggests that it is not reasonable to representation is not a building block within the sys- speak about the relation between signs and things in tem, with perception being a projection from out- the outer world. Instead, he argues that it makes side, but where perception is a process of anticipat- sense to speak about the relation between signs (i.e., ing the sensory consequences of actions. Tom “linguistic expressions”) and concepts, i.e., meaning Wexler gives empirical evidence for the importance in the head of the individuals. That linguistic expres- of actions in representations using both psycho- sions mean the same for everybody is not because physical examples and simulation (representation they map a fixed world but because they emerge dur- as strategy for acting rather than as a mapping of the ing interaction from the concepts in the heads of the environment). He demonstrates examples where language users. Examples of neurophysiological previously unresolved problems (e.g. the “n-parity correlates of language processing are given by Sab- problem“) can be solved if the learning system is ine Weiss et al. The problem of transferring mean- directly interacting with the environment in a sen- ing between two languages—the problem of transla- sory–motor relation, i.e., if the environment is tion—is considered by Hanna Risku. included into the representation. Till now we referred to representation as men- When a first step in defining representations on a tal–environmental stabilities that have evolved to new ground is to consider the sensory–motor cou- successfully guide the behavior of organisms; pling with an organisms’s environment, a second objects would in this sense be considered as adap- step is to consider the coupling not only with a static tive constructs in this organism–environment rela- environment but with other representing systems. tion. This, however, does not address the question The role of social interactions—in particular lan- of the subjective experience of representations, guage—for stabilities within a socio-cultural net- i.e., the awareness that there are things in the out- work, and the emergence of representations and side world, or the awareness of ones self as active “meaning” is a subject which is further treated by part in the interaction with the world. The chapter several authors in the chapter “Communication and “Qualitative Aspects of Representation and Con- Social Coupling”. From a social-psychological sciousness” deals with this problem of conscious- point of view, Richard Eiser asks the question how ness. Wolf Singer addresses the issue of conscious individuals acquire knowledge or representation of experience from a neuroscientific perspective, social reality that adapt with new experience, and ending with the conclusion that properties such as also how groups can coordinate their behavior in a self awareness and consciousness cannot be manner adaptive to their environment. He presents a explained in terms of neuronal correlates alone, theory about the self, where the self is seen as the but only by additionally considering the (social) product of the cognitive system at work, i.e., as interactions among brains. His position is, thus, explanatory model for already performed or to-be- both dualistic and materialistic, saying that con- performed behavior. Tom Routen demonstrates that sciousness is housed by a neuronal substrate but the advantage of the adaptive behavior approach can never be explained by neuroscience itself since 22 Understanding Representation in the Cognitive Sciences it considers a phenomenon which cannot be found mental content cannot refer to anything but to fur- in single brains (See also discussions, page 291). ther mental content. He suggests to use the word Michael Pauen stresses the importance of struc- “presentation” instead, to imply that it is a proposal tural similarities in representations: differences in the brain offers to the outside world in adapting to the world have to scale to differences in the neu- that world; a world, however, that in principal can ronal activities. Qualia is proposed to serve as the not be known. Empirical evidence for the construc- aspect organizing these structures. The problem of tivist nature of representation is presented by qualia dissolves if one abandons the artificial dis- Sverre Sjölander. He investigates the subject from tinction between subjective experience and objec- an evolutionary perspective and describes how tive outside world. object-centered (invariant) representations develop Finally, the chapter “Constructivism” concludes from simple input–output processing in lower ani- with the deep epistemological implications the mals. He shows that the evolutionary purpose of above considerations present. If we agree that what sensory processing is not a mapping of the world is represented in the brain is not just a projection but producing adaptive behavior. Annika Wallin from entities in the outside world but rather a reper- shows that the fact that we have two types of repre- toire of hypothesis about future actions, grouped sentations—immediate perceptions and re-presen- and structured in an efficient way to successfully tations such as during imagery—creates a serious act—what, then, is that reality out there? problem for constructivism: according to construc- Ernst von Glasersfeld has a radical approach to tivist theories, the concepts constructed to sub-serve that question: he argues that what we perceive as adaptive behavior are adjusted according to the con- outside world is nothing but constructions of our straints presented to the system. cognitive apparatus. In his article he proposes to not use the word re-presentation at all, since it is com- As this overview shows, our goal was to bring monly understood as a “copy” or “reproduction” together interdisciplinarily working scientists in suggesting the reference to an original. While this order to contribute to this volume. We believe that implies that we can compare a mental experience to our selection of papers will help foster a better something outside our mental experience—the out- understanding and fruitful application of the ubiqui- side world—he argues that this is not possible since tous nature of representation.