Concept-Driven Suppositional Scenarios

Work on concepts has concentrated on categorization. Categorization is a process that starts with perceptual representations (and other domain-specific resources) and transitions to a conceptual representation. Just as important, however, are inferences that, rather than being driven by current stimuli, start with conceptual representations. Often we engage in ‘offline’ thinking to formulate plans and take decisions. We derive consequences from facts that we remember and episodes that we have experienced. The usual paradigm for conceptually-driven thinking is reasoning. Reasoning moves from some concept-involving thoughts to others. Reasoning deploys a domain-general computational process, rather like theorem-proving in logic, and uses a domain-general representational system, that of conceptual thought.

Often overlooked between these two paradigms is the way that offline processing driven by conceptual thought draws on special-purpose resources: sensory, motoric, affective, or supra-modal (e.g. the cognitive map of space). The phenomenon itself if familiar: we imagine a situation or think through counterfactual scenarios using representations from special-purpose systems. It encompasses simulation, prospection and some forms of imagination. What is under-theorised is how suppositional thinking driven by conceptual representations works computationally.

In this paper I ask: what information do these kinds of inferences draw on, how is it encoded, and what kinds of computational process operate on it? I will argue that information is encoded in three quite distinct ways. As well as being represented explicitly in conceptually structured thoughts, suppositional thinking draws on information represented implicitly in the computational dispositions of domain-specific systems, and on information represented explicitly but in a domain-specific way in structural representations in special-purpose systems. There is also a fundamental difference in how the computations work. Computations in domain-specific systems are usually ‘content-specific’, whereas computations over explicit representations in conceptual thought call for a quite different capacity, the capacity for ‘non-content-specific’ computation. The paper will define, illustrate and defend this new distinction.