What If: The Semantics, Pragmatics, & Psychology of Counterfactuals

Abstracts of Talks


Patricia Ganea, "What if reality matters? Children’s representation of reality affects their ability to reason counterfactually" (joint work with Angela Nyhout)

Previous work has suggested that children are not able to reason counterfactually about causally-overdetermined events until adolescence. For example, if children hear a story about Max and Susie who both wear their muddy shoes into the house and make the floor all dirty, and children are then asked "what would have happened if Susie had taken her shoes off?" they often disregard the second cause of the muddy floor (Max) and respond that the floor would be clean. We argue that one reason that children fail this and similar tasks is due not to an inability to reason counterfactually, but due to differences in the their representation of reality off which their counterfactual inferences are based.

In two studies using short stories, we varied causal event structure and found that when the causal structure was less ambiguous - thereby reducing the likelihood of unwarranted causal inferences - children demonstrated adult-like counterfactual reasoning by the age of 8. Before age 8, children's performance was at chance. In a third study, also using short stories, we clarified the causal event structure more by using unambiguously causally-related or unrelated pairs of events that involved a character finding out a new piece of information in two different ways (e.g., learning about a new planet on the TV and by being told by someone). Even with a clearer causal structure, 5- and 6-year- olds have difficulty on this task. Finally, in a study using a simple physical causal system with doubly- or singly-determined outcomes, we are finding that 5-year-olds can reason counterfactually when asked about the removal of one of the causes. These results suggest that given a sufficiently simple task with a clear causal structure, children as young as 5 can engage in counterfactual reasoning in overdetermined contexts. Together, these studies suggest a developmental trajectory whereby children's counterfactual reasoning abilities interact with their developing causal reasoning abilities and domain-specific knowledge (e.g., knowledge of physics vs. psychology). Children may in some cases arrive at different counterfactual conclusions because of differences in how they represent the causal structure of reality.


Karen Lewis, "Counterfactual Contextualism and Embedded Counterfactuals"

In Lewis (2016, 2017) I defend a contextualist theory of counterfactual conditionals, in which counterfactuals are sensitive to conversational relevance. One of the central upshots of the view is that it can account for clashes between counterfactuals like the following:

    1. #If I had dropped my mug, I might have swiftly caught it before it fell to the floor; but of course if I had dropped my mug, it would have fallen to the floor.
    2. #If I had dropped my mug and I had swiftly caught it, it wouldn’t have fallen to the floor; but of course if I had dropped my mug, it would have fallen to the floor.

On my view, the second counterfactual in each discourse is false in the context in which it is uttered, though the same sentence can express a true proposition in other contexts (such as discourse initially in an ordinary conversation). The embedding data for counterfactuals raised in Moss (2013) challenges the idea that these counterfactuals are false (in context). They don’t embed like falsehoods: they don’t embed neatly under negation, but do embed neatly under attitudes like ‘I wonder’ and operators like ‘probably’, for instance. I argue that the embedding data is not as problematic for my view as one might think. I explore the potential of a reinterpretation strategy, i.e., that ‘Probably, If P then would Q’ actually means something of the form ‘If P, the probably would Q’ and defend it against some of Moss’s objections. I also argue that there are good metaphysical and semantic reasons for maintaining the falsehood of these counterfactuals and others in the face of the admittedly recalcitrant embedding data.


Kyle Rawlins, "Asking What-Ifs" (joint work with Justin Bledin)

What-if questions in discourse can be used in extremely diverse ways, from hypotheticals and counterfactuals such as (1), to much more ordinary suggestions, enticements, and resistance moves such as (2-4).  

    1.       What if I tried to re-enter the atmospere in my car? (extreme hypothetical; from https://what-if.xkcd.com/142/)
      1. Who should we invite to give a talk?
      2. What if we invite Joanna? (suggestion)
      1. I'm not going to the party.
      2. What if Joanna's there? (enticement)
      1. Open the window.
      2. What if it's still raining? (resistance)

The puzzle I address is how to give an analysis of the semantics and pragmatics of "what if" questions that unifies all of these cases while capturing the core intuition that "what if"s introduce hypothetical possibilities into the flow of discourse.  The proposal is that "what if"s are anaphoric to an existing "Question Under Discussion" in discourse, re-raising whatever question is currently under discussion, but in conditionalized form.  However, in some cases interpreting a "what if" question requires going beyond immediate questions in discourse to reason about "decision problems" — choices among actions that agents are currently considering for some salient purpose.  The analysis therefore requires a model of discourse that integrates inquisitivity and actions.  Along the way I show that "what if" questions have all the properties of conditionals, including accepting subjunctive/counterfactual morphology, but without a consequent.