Counterfactuality and causal structure
The following is an example of a counterfactual conditional in English:
(1) If I had thrown a six, I would have won the game.
One normally infers from (1) that the antecedent is counterfactual (i.e. I did not throw a six; I write this as CF-p), and that the consequent is counterfactual (i.e. I did not win the game; written CF-q). Whereas most previous literature focuses on the counterfactuality of the antecedent exclusively (perhaps assuming that an analysis of CF-p extends to CF-q), this work provides an analysis for how the counterfactual inference of the consequent (CF-q) is generated, and explains its empirical distribution.
I identify several contexts in which the CF-q inference gets cancelled. In some of these, cancellation is the result of the presence of a specific lexical item (such as “also”). In other cases, it is the intonation contour of the conditional that leads to cancellation. By analyzing the topic-focus structure of conditionals, I argue that the various contexts in which CF-q gets cancelled have a pragmatic property in common: they are multiple cause contexts. This means that they make more than one cause for the same consequent salient.
The next step in my analysis is adopting an idea going back to Karttunen (1971), who suggests that conditional perfection (the pragmatic strengthening of conditionals to biconditionals) is a necessary ingredient for the CF-q inference to arise. The key prediction, which has not been explored before, is that if for some reason conditional perfection is not triggered, the CF-q inference is not drawn. I derive the independent result that multiple cause contexts do not trigger conditional perfection. This provides the desired explanation of why in multiple cause contexts the CF-q inference is not drawn. This analysis opens a new way to investigate counterfactuality, namely by using tools from the study of discourse (questions and answers, topic and focus, exhaustivity). Finally, I sketch some directions of future work on how using causal networks to represent multiple causation can be applied to the pragmatics of counterfactual conditionals.