Two Proposals for Causal Grammars

In the previous chapter (Tenenbaum, Griffiths, & Niyogi, this volume), we introduced a framework for thinking about the structure, function, and acquisition of intuitive theories inspired by an analogy to the research program of generative grammar in linguistics. We argued that a principal function for intuitive theories, just as for grammars for natural languages, is to generate a constrained space of hypotheses that people consider in carrying out a class of cognitively central and otherwise severely underconstrained inductive inference tasks. Linguistic grammars generate a hypothesis space of syntactic structures considered in sentence comprehension; intuitive theories generate a hypothesis space of causal network structures considered in causal induction. Both linguistic grammars and intuitive causal theories must also be reliably learnable from primary data available to people. In our view, these functional characteristics of intuitive theories should strongly constrain the content and form of the knowledge they represent, leading to representations somewhat like those used in generative grammars for language. However, until now we have not presented any specific proposals for formalizing the knowledge content or representational form of “causal grammars.” That is our goal here. Just as linguistic grammars encode the principles that implicitly underlie all grammatical utterances in a language, so do causal grammars express knowledge more abstract than any one causal network in a domain. Consequently, existing approaches for representing causal knowledge based on Bayesian networks defined over observable events, properties or variables, are not sufficient to characterize causal grammars. Causal grammars are in some sense analogous to the “framework theories” for core domains that have been studied in cognitive development (Wellman & Gelman, 1992): the domain-specific concepts and principles that allow learners to construct appropriate causal networks for reasoning about