Consistency Checking in Hypothesis Generation

Abstract Three experiments were performed to study an aspect of a hypothesis-generation model. This model assumes a candidate hypothesis may be retrieved or activated in memory in response to only part of the available data. This candidate hypothesis may then be checked for consistency against the remaining data. The latter process is called “consistency checking.” Experiment 1 was performed to provide evidence that consistency checking occurs during hypothesis generation. Subjects were able to recognize hypotheses which were retrieved during a hypotheses-generation problem but not emitted as hypothesis responses, suggesting that consistency checking was responsible for the rejected hypotheses. Experiment 2 indicated that the amount of time needed to process an additional datum is a consistency-checking task was less than an estimate of the time needed to process an additional datum in hypothesis retrieval. The results suggest that consistency checking is a high-speed verification process rather than a slower search process. Experiment 3 was performed to provide evidence that consistency checking is a self-terminating process. Subjects' latencies depended upon the position of a disconfirming datum within a data set, supporting this conjecture. The results generally confirmed the existence of a high-speed verification process in hypothesis generation and also suggest that the generation of hypotheses in response to multiple data occurs as a result of dual processes.