In choosing to write a book primarily about counterfactuals, Ruth Byrne has been able to integrate the literatures on reasoning and social cognitive aspects of decision making and judgment. In so doing, she has performed an invaluable service to cognitive and social psychologists alike. Thus, there is much to applaud in The Rational Imagination (Byrne Reference Byrne2005) and there are several of its aspects upon which we could comment. Because we have commented elsewhere on Byrne's application of mental model theory to judgmental phenomena such as the action effect (see Feeney & Handley Reference Feeney and Handley2006), in this commentary we focus on the mental model account that she outlines of how people reason about semifactual conditionals.
According to Byrne, a subjunctive semifactual conditional such as
(1) Even if Pete had studied hard he would have failed the exam
usually conveys the conjecture that its antecedent is false and its consequent true, and is often used to assert that the antecedent could not have prevented the consequent from occurring. Even-if conditionals call for the representation of two possibilities: one where the antecedent occurs and the consequent occurs, and one where the antecedent does not occur and the consequent occurs. If we consider the foregoing example, people represent the conjecture, “He studied hard and failed the exam,” and they represent the presupposed facts, “He didn't study hard and failed the exam.” This mental representation is said to explain people's tendency not to affirm the consequent (i.e., that he studied hard, from being told that he failed the exam) and to infer the opposite to the standard conclusion (i.e., that he failed rather than passed the exam) from a denial of the antecedent (see Handley & Feeney Reference Handley, Feeney, Noveck and Sperber2004; Reference Handley, Feeney, Schaeken, Vandierendonck, Schroyens and d'Ydewalle2007; Moreno-Rios et al. 2004). It also provides an explanation for one of the most intriguing characteristics of concessive conditionals – their compelling invitation to the listener to infer the consequent, a characteristic much commented on in linguistics and philosophy (Konig Reference Konig, Traugott, Meulen, Reilly and Ferguson1986).
It has been recognised in these literatures that the study of less common conditional forms can provide significant insights into the way in which the ordinary conditional is represented and processed. Consequently, the study of how people reason and think about even-if is a valuable enterprise and Byrne's analysis provides an interesting, though in our view incomplete, framework. The principal problem is that the account as presented is purely descriptive, and lacks clear principles or a detailed mechanism that can explain how people arrive at a representation corresponding to the possibilities that Byrne describes. In this commentary we present an alternative account of even-if, which is grounded in linguistic pragmatics, and we contrast this with Byrne's model. The account details how even serves to modify the representation of the conditional connective if, and draws upon recent work on the suppositional conditional (Evans et al. Reference Evans, Handley and Over2003; Handley et al. Reference Handley, Evans and Thompson2006).
Understanding even-if requires a consideration of the function of even in everyday natural language. Consider the assertion in conditional 2:
(2) Even Tony distrusts George
Several philosophers (Jackson Reference Jackson1987; Sanford Reference Sanford1989) have suggested that even serves to deny an available presupposition, for example, that we might expect Tony to trust George. It serves to pick out an extreme position, and calls up a range of contextually determined alternatives that are less surprising; for example, that Gordon distrusts George, Hilary distrusts George, or Jacques distrusts George. In so doing, the utterance invites the listener to infer that George is a man not to be trusted.
In the study of different conditional constructions it is important to consider how the linguistic terms interact in determining meaning. Understanding how even interacts with if also requires an account of the conditional connective. According to the suppositional account, conditionals cue a mental simulation (often referred to as the Ramsey test) in which the listener imagines that the antecedent condition holds and evaluates their degree of belief in the consequent in that context (Evans et al. Reference Evans, Over and Handley2005). For example, consider the following conditional:
(3) If the United States cuts fuel emissions then global warming will be reduced
This assertion cues us to suppose that the United States cuts their emissions, and on the basis of this supposition, together with background beliefs, we can evaluate our belief that global warming will in fact be reduced. The suppositional account predicts that belief in a conditional is closely related to conditional probability (P[q/p]), a prediction confirmed in numerous recent studies (Evans et al. Reference Evans, Handley and Over2003; Over et al. Reference Over, Hadjchristidis, Evans, Handley and Sloman2007). Of course one might not believe that there is any sort of relationship between U.S. fuel emissions and global warming and imagine that global warming will increase irrespective of U.S. policy, which creates a perfect opportunity for asserting a concessive conditional:
(4) Even if the United States cuts fuel emissions global warming will increase
Combining the analysis of even with our account of if, the concessive in conditional 4 denies the presupposition in conditional 3, and calls up a range of alternative conditionals on a probability scale that are less surprising or unexpected, and where, in probabilistic terms, P(q/p) is higher. Often, as Jackson (Reference Jackson1987) has pointed out, the scale will consist of the conditional as in (4) above, with its antecedent negated:
(5) If the United States doesn't cut fuel emissions global warming will increase
The combination of the conditional in (4) with the conditional in (5) leads directly to the inference that global warming will increase, irrespective of U.S. policy. It is important to note here that this inference is both logically (through constructive dilemma) and probabilistically valid, but that it is not the result of representing discrete possibilities, consistent with the initial conditional, as Byrne's account claims. What is important about our account, and contrasts clearly with Byrne's, is that we specify how even and if combine in cueing the activation and representation of alternative associations that can then be integrated with a representation of the original assertion in order to make inferences. Unfortunately, the details of such a mechanism are absent in Byrne's account.
A second important point of contrast between our account and Byrne's is that, because ours is rooted in linguistic pragmatics, it is naturally extended to inferences about speaker intentions. For example, we have presented people with even-if conditionals and asked them to make immediate inferences about whether the speaker intends to carry out the action specified in the antecedent (Handley & Feeney Reference Handley, Feeney, Noveck and Sperber2004; Reference Handley, Feeney, Schaeken, Vandierendonck, Schroyens and d'Ydewalle2007). In general, people seem to infer from these conditionals that the speaker does not intend to carry out the antecedent action. So when presented with an even-if assertion of the following kind:
(6) Even if I study hard I will fail the exam
participants infer that the speaker does not intend to study. This finding suggests first that people's representations of the utterance might include information about the cost of the antecedent action, and second that even-if conditionals serve an important rhetorical function. Our intuitions here are, no doubt, related to Byrne's claims about the role that semifactual conditionals play in denying a causal link between antecedent and consequent. One can justify a decision not to study by implying that in this instance studying will not cause one to pass an exam.
In conclusion, although we welcome Ruth Byrne's highly integrative book and her original analysis of even-if and the semifactual conditional, we disagree with some of the details of her account. We very much look forward to debating these and other issues with her in the coming years.
