A commentary on 'Perceptual Symbol Systems' by Lawrence W. Barsalou, and accepted for publication in Behavioral and Brain Sciences.

Liane Gabora
Center Leo Apostel, Brussels Free University (VUB)
Krijgskundestraat 33,
B-1160 Brussels, Belgium

lgabora@vub.ac.be

The symbol grounding problem is important, and Barsalou's paper does an excellent job of directing our attention to it. However, to be grounded in perceptual experience does not mean to spend eternity underground. Just as a seed transforms into a sprout which emerges from the ground as a plant, an abstract concept can originate through, or be grounded in, perceptual experience, yet turn into something quite different from anything ever directly perceived. To insist that abstractions are just arrangements of perceptual symbols is like insisting that a plant is just seed + water + sunlight.

As Barsalou considers increasingly abstract concepts, he increases the complexity of the perceptual symbol arrangements. To me it seems more parsimonious to say that what was once a constellation of memories of similar experiences has organized itself into an entity whose structure and pattern reside primarily at a level that was not present in the constituents from which it was derived. This does not mean that abstractions can't retain something of their "perceptual character" [p. 5].

Barsalou assumes that viewing symbols as amodal is incompatible with their being derived from sensory experience, because of "their failure to provide a satisfactory account of the transduction process that maps perceptual states into amodal symbols" [p. 5]. However he admits that perceptual symbols are analogical [p. 3, S1.1P2], that they are integrated "combinatorially and recursively" [p. 9, S1.5P2], and speaks of "filtering out the specific circumstances" [p. 13, S2.3.1P1]. Aren't these the kind of processes that could transduce perceptual symbols? (Incidentally, it is strange that Barsalou cites Gentner, who works on structure mapping in analogy, to support the notion that "Mental models are roughly equivalent to only the surface level" and "tend not to address underlying generative mechanisms that produce a family of related simulations" [p. 15; 2.4.2P2].)

Barsalou pays surprisingly little attention to the highly relevant work of connectionists. The only rationale he provides is to say that because 'the starting weights between connections are set to small random values' ... 'the relation between a conceptual representation and its perceptual input is arbitrary' [p. 5, S1.2P8]. This is misleading. Just because there are many paths to an attractor doesn't mean the attractor is not attracting. Surely there are analogous small random differences in real brains. This quick dismissal of connectionism is unfortunate because it has addressed many of the issues Barsalou addresses, but its more rigorous approach leads to greater clarity.

Consider, for example, what happens when a neural network abstracts a prototype such as the concept 'depth', which is used in senses ranging from 'deep blue sea' to 'deep-freezed vegetables' to 'deeply moving book'. The various context-specific interpretations of the word cancel one another out; thus the concept is, for all intents and purposes, amodal, though acquired through specific instances. There is no reason to believe this does not happen in brains as well. The organizational role Barsalou ascribes to 'simulators' emerges implictly in the dynamics of the neural network. Moreover, since Barsalou claims that "a concept is equivalent to a simulator" [p. 15, S2.4.3P2], it is questionable whether the new jargon earns its keep. Why not just stick with the word 'concept'?

Harder to counter is Barsalou's critique that an amodal system necessitates "evolving a radically new system of representation" [p. 40, S4.2P3]. Barsalou repeatedly hints at but does not explicitly claim that the origin of abstract thought presents a sort of chicken-and-egg problem. That is, it is difficult to see how an abstraction could come into existence before discrete perceptual memories have been woven into an interconnected worldview that can guide representational redescription down potentially fruitful paths. Yet it is just as difficult to see how the interconnected worldview could exist prior to the existence of abstractions; one would expect them to be the glue that holds the structure together. In [Gabora, 1998; Gabora, in press] I outline a speculative model of how this might happen, drawing on Kauffman's [1993] theory of how an information-evolving system emerges through the self-organization of an autocatalytic network. Self-organizing processes are rampant in natural systems [Kauffman 1993], and could quite conceivably produce a phase transition the catapults the kind of change in representational strategy that Barsalou rightly claims is necessary.

It also seems misleading to cite Donald [1993] as supporting the statement "Where human intelligence may diverge [from animals] is in the use of language" [p. 40, S4.2P3]. Donald writes: "Speech provided humans with a rapid, efficient means of constructing and transmitting verbal symbols; but what good would such an ability have done if there was not even the most rudimentary form of representation already in place? There had to be some sort of semantic foundation for speech to have proven useful, and mimetic culture would have provided it." Thus Donald argues that human intelligence diverged from animals before the appearance of language, through of the advent of mimetic skill.

Despite these reservations, I found the article interesting and insightful.

Donald, M. (1991) Origins of the modern mind, Harvard University Press, p. 199.

Gabora, L. (1998) Autocatalytic Closure in a Cognitive System: A tentative scenario for the evolution of culture. Psycoloquy, Volume 9.

Gabora, L. Weaving, bending, patching, mending the fabric of reality: A cognitive science perspective on worldview inconsistency. To appear in Foundations of Science.

Kauffman, S. A. (1993) Origins of order, Oxford University Press.