Filtering Semantics for Counterfactuals: Bridging Causal Models and Premise Semantics

<p>I argue that classical counterfactual semantics in the style of Stalnaker, Lewis, and Kratzer validates an inference pattern that is disconfirmed in natural language. The solution is to alter the algorithm we use to handle inconsistency in premise sets: rather than checking all maximally consistent fragments of a premise sets, as in Krazter's semantics, we selectively remove some of the premises. The proposed implementation starts from standard premise semantics and involves a new 'filtering' operation that achieves just this removal. The resulting semantics is interestingly related to the semantics for counterfactuals emerging from Judea Pearl's causal models framework in computer science: in particular, filtering is a possible worlds semantics counterpart of Pearl's interventions.</p>

EXPRESSIBLE SEMANTICS FOR EXPRESSIBLE COUNTERFACTUALS

Lewis (1981) showed the equivalence between two dominant semantic frameworks for counterfactuals: ordering semantics, which relies on orders between possible worlds, and premise semantics, which relies on sets of propositions (so-called ordering sources). I define a natural, restricted version of premise semantics, expressible premise semantics, which is based on ordering sources containing only expressible propositions. First, I extend Lewis’ (1981) equivalence result to expressible premise semantics and some corresponding expressible version of ordering semantics. Second, I show that expressible semantics are strictly less powerful than their nonexpressible counterparts, even when attention is restricted to the truth values of expressible counterfactuals. Assuming that the expressibility constraint is natural for premise semantics, this result breaks the equivalence between ordering semantics and (expressible) premise semantics. Finally, I show that these results cast doubt on various desirable conjectures, and in particular on a particular defense of the so-called limit assumption.