Linguistic determinants of performance on formal problems

Abstract

The research described investigates why subjects frequently give logically wrong answers to problems requiring deductive reasoning. In Experiment 1 a pattern of erroneous responses to Mason's selection task which has commonly been attributed to verification bias is shown to be due to a for a of matching bias (higher-order matching) which takes account of negation. This form of matching is shown to be associated with binary problem content.
In Experiments 2 and 3 it is shown that responses are sometimes made on the basis of salient problem features when a logical task is of only moderate difficulty. This suggests that, contrary to a widely held view, matching nay not be a fall-back strategy in response to logical difficulty.
Experiment 4 demonstrates that comprehension nay be based on pragmatic features of sentences, rather than their grammatical structure. It is argued that subjects night comprehend the selection task in this way, and be unaware of its logical structure and difficulty. Independent evidence using conditional reasoning problems is described in support of this conclusion.
In Experiment 3 subjects frequently formulate a potentially conditional truth-function in a conjunctive fora to which higher-order matching yields correct responses. The reason for previous experimenters' failure to observe facilitation with conjunctive formulations of the selection task is demonstrated in Experiments 6, 7 and 6.
It is argued that errors on the THOG problem are due to a fora of matching bias. In Experiment 9 some facilitation is observed on a plausible version of the THOG problem, possibly because the disjunctive rule was not stated. In Experiment 10 conjunctive formulation is also shown to considerably improve performance. It is argued that conjunctive formulation cues a more complete representation of the premises and an effective solution strategy of a kind which has not previously been suggested.