The Role of Negation in Conditional Inference

This paper addresses the issue of how negative components affect people's ability to draw conditional inferences. The study was motivated by an attempt to resolve a difficulty for the mental models theory of Johnson-Laird and Byrne, whose account of matching bias in the selection task is apparently inconsistent with Johnson-Laird's explanation of the double negation effects in conditional inference reported by Evans, Clibbens, and Rood (1995). Two experiments are reported, which investigate frequencies of conditional inferences with task presentation similar to that of the selection task in two respects: the presence of a picture of four cards and the use of implicit negations in the premises. The latter variable was shown to be critical and demonstrated a new phenomenon: Conditional inferences of all kinds are substantially suppressed when based on implicitly negative premises. This phenomenon was shown to operate independently of and in addition to the double negation effect. A third experiment showed that the implicit negation effect could be extended to the paradigm in which people are asked to produce their own conclusions. It is argued that these two effects can be explained within either the mental models theory or the inference rule theory, of propositional reasoning, but that each will require some revision in order to offer a convincing account.     

The Influence of Linguistic Form on Reasoning: The Case of Matching Bias

A well-established phenomenon in reasoning research is matching bias : a tendency to select information that matches the lexical content of propositional statements, regardless of the logically critical presence of negations. Previous research suggested, however, that the effect might be restricted to reasoning with conditional statements. This paper reports two experiments in which participants were required to construct or identify true and false cases of propositional rules of several kinds, including universal statements, disjunctions, and negated conjunctions. Matching bias was observed across all rule types but largely restricted to problems where participants were required to falsify rather than to verify the rules. A third experiment showed a similar generalization across linguistic forms in the Wason selection task with only if conditionals substituted for universals. The results are discussed with reference to contemporary theories of propositional reasoning.     

Some steps towards a general theory of relevance

The classical analysis of relevance in probabilistic terms does not fit legal, moral or conversational relevance, and, though analysis in terms of a psychological model may fit conversational relevance, it certainly does not fit legal, moral or evidential relevance. It is important to notice here that some sentences are ambiguous between conversational and non-conversational relevance. But, if and only ifR is relevant to a questionQ, R is a reason, though not necessarily a complete or conclusive reason, for accepting or rejecting something as an answer toQ. Reasons of this kind are governed by appropriate covering laws or principled probabilities and a number of questions thus arise about the relationship between relevance and certain formal-logical properties.     

A critique of van Fraassen's voluntaristic epistemology

Van Fraassen's epistemology is forged from two commitments, one to a type of Bayesianism and the other to what he terms voluntarism. Van Fraassen holds that if one is going to follow a rule in belief-revision, it must be a Bayesian rule, but that one does not need to follow a rule in order to be rational. It is argued that van Fraassen's arguments for rejecting non-Bayesian rules is unsound, and that his voluntarism is subject to a fatal dilemma arising from the non-monotonic character of reasoning.