Syntactic and Causal Constraints on the Necessity of Conditional Inferences by Readers

The data of three experiments (Campion, in press) confirm that the readers of texts including conditional arguments process the conditional syntax as an asymmetric constraint which warrants the Modus Ponens, a logically valid inference. However, causal knowledge can raise doubt about that inference and warrant the validity of the reciprocal inference (Affirmation of the Consequent). Thus, according to the theory of natural logic, the readers can make formal deductions, such as the modus Ponens. However, the readers also reason on causal bases and a formal deduction must not contradict knowledge about causality in the world. This provides further evidence that readers can represent their inferences as hypotheses (Campion, 2004). Logically valid deductions are made during text comprehension (Campion, 2004; Rader and Sloutsky, 2002; Lea, O’Brien, Fisch, Braine and Noveck, 1990; Lea, 1995). What is controversial is whether these inferences reflect the readers’ ability to draw formal deductions from text propositions. According to the mental or natural logic theory (Braine, 1978; Braine & O'Brien, 1991), the adults possess a set of mental rules that correspond to inference schemas and lead to make logically correct deductions. According to Rader and Sloutsky (2002), the mental model theory (Johnson-Laird, 1983. Johnson-Laird et Byrne, 2002) gives a better account of the deductions made when reading conditional arguments in texts. In support of that view, they used short stories with a conditional premise of the form: if antecedent then consequent (e.g., “If it is night, then it is cold outside”). A subsequent premise either affirmed the antecedent (e.g., “it is night”), and therefore elicited a Modus Ponens (MP) inference (e.g., “it is cold outside”), or affirmed the consequent (e.g., “it is cold outside”), and therefore elicited an affirmation of the consequent (AC) inference (e.g., “it is night”). AC inferences do not correspond to logically-valid deductions. However, Rader & Sloutsky (2002) found that the AC and MP inferences were both activated by the readers of their texts and were then both recognized with a same frequency as paraphrases of the text information. As pointed out by Rader & Sloutsky, these findings do not support the natural logic theory because no inference schema corresponds to AC inferences in the theory, in contrast to MP inferences that should be favored. In contrast, the findings support the mental model theory which stipulates that only one model is normally constructed to represent the meaning of a conditional premise. Building more models would be made when the conditional meanings need to be exhaustively represented, and reading a text for comprehension does not require these additional models. Consequently, readers would not build the additional model that falsifies the AC inferences, and the AC and MP inferences of readers would be represented as identical. The first aim of this study was to investigate whether readers represent their AC inferences as hypothetical facts, and therefore distinguish them from their MP inferences, that they represent as facts that are certain. A previous study (Campion, 2004) supported a similar claim: the readers would represent their deductions from categorical syllogisms as facts that are certain, while they represent the predictions they make about what should happen next in the story as hypotheses. In agreement with that view, I suggest that AC inferences might be hypothetical inferences. Readers would draw the AC inferences, because they are the relevant conclusions of AC arguments, but would represent them as hypotheses, because they are not constrained by the conditional syntax. A complementary assumption is that the readers would represent the MP inferences as certain facts, because they are syntactically constrained by the text. This study also examined the impact of readers' worldknowledge on the AC and MP inferences that involve causal relations between events. The existence of that impact has already been demonstrated in studies of reasoning. Thus, alternative causes prevent certainty responses to AC inferences from a causal conditional: “If cause then effect”. For example, the conditional premise: “If a dog has a skin disease, then it will scratch constantly”, leads to few certainty responses to AC inferences, presumably because the alternative cause “ a dog has fleas” can be easily retrieved from knowledge about dogs. (Barrouillet, Markovits and Quinn, 2001; Cummins, 1995; Markovits and Quinn, 2002). Similarly, reading conditional arguments would automatically activate the corresponding causal knowledge in readers Long-term memory (Campion & Rossi, 2001; Kintsch, 1998) and should invite the readers to check the compatibility of the conditional meaning with the underlying causal knowledge about the events in conditional relation (Singer, Halldorson, Lear, Andrusiak, 1992; van den Broek, 1990). However, if we assume that the conditional syntax constrains the making of MP inferences that are certain, world-knowledge should have a limited impact on the representation of MP inferences by readers. In contrast, AC inferences are not syntactically constrained, and their hypothetical or certain status will be heavily dependent on world-knowledge.