West Coast Conference on Formal Linguistics 28.
Based on the previously unnoticed contrast between standard counterfactuals and the non-canonical counterfactual construction "if not for P, Q" (hereafter NC, for `Not' Counterfactuals), this paper argues for the emerging proposal that two distinct routes to counterfactuality are available in natural language (see e.g., Schulz 2007): (i) global revision of a belief state, and (ii) local revision of a world. Schulz's (2007) observation is that the classic epistemic inferences are precisely those that disappear under local update in a causal model, allowing for a treatment of the ontic-epistemic distinction in counterfactuals as an ambiguity between local and global belief revision. We show that NCs systematically reject epistemic readings, supporting the position that different ways of evaluating the antecedent permit different types of counterfactual inferences, where NCs only allow local revision due to their distinct morphology. The core proposal is that models for interpreting counterfactuals must be enriched with causal laws, and that the antecedents of NCs make local changes to a world to remove a fact contributed by the antecedent, potentially violating the model's causal structure. This accounts for the fact that NCs systematically reject non-causal epistemic inferences, while otherwise retaining their paraphrasability with standard counterfactuals.