A GNU implementation of the Mental

1 Abstract An outline of Johnson-Laird's theory of Mental Models J-L 83] and J-L&Byr 91] is presented along with a short explanation of its importance. Implementational issues are discussed, and a possible GNU based architecture for a syllogistic reasoner is proposed. 2 Introduction Mental models J-L 83, J-L&Byr 91] are an attempt to explain human cognition, and speciically reasoning which appears to be rule-based, in terms of mental manipulations of \representative" images, known as mental models. Hence, it is asserted, a subject who is asked to make deductions from a syllogism of the form All A are B. Some B are C. rst forms a model comprising a number of As, Bs and Cs and then adds annotations or performs manipulations of the image to come to a conclusion. The way in which the models are formed, the way in which they are manipulated, the way in which they are tested for correctness and appropriateness and the way in which a conclusion is drawn from the model all aaect the apparent ease of the problem, and lead to testable hypotheses about the relative ease or diiculty of particular syllogisms. This explicitly model based method of representing information seems particularly suited to a Neural State Machine A&M, 93] treatment. However the complexity of the task is such that many analytical issues the govern the behviour of GNUs need to be resolved, and if good generalisation is to be achieved, adaptations may have to be made. The theory of mental models rests on the hypothesis that a human reasoner, when presented with a statement from which a deduction is to be drawn, creates a model (in the logical sense) from which conclusions may be drawn. The represention of the model is kept minimal by means of an \and so on" notation which allows sentences containing universal quantiiers to be modelled. As new premises are added, individual elements in the model are deleted and previously unexpanded "and so ons" may be expanded and instantiated.