Using Novel Semantic and Informational Manipulations of Rules to Extend an Interpretive Approach to Conditional Reasoning
Alkhalifa, Eshaa Mohamed
This thesis extends a view of human reasoning which emphasises a theory of interpretation in conditional reasoning. It extends work on Wason’s (1968) ‘selection task’, using novel rules and contexts to explore the factors that control subject interpretations, which in turn is reflected in their performance. After reviewing the work on conditional reasoning and particularly the interpretative framework of Stenning & van Lambalgen (2004), the thesis explores subjects’ reasoning with rules that describe processes extended in time in two experiments. The most striking finding is that many subjects exhibit an unusual constant anaphor reading, even though the anaphors involved are tenses rather than pronouns. Results are explained in terms of the temporal constraints involved in the situation described. The thesis then uses novel ‘information packaging’ manipulations which use colour to emphasise different distinctions in Wason’s original task. This manipulation provides evidence of where subjects’ attention already rests. This is combined with a task that gathers data of subjects’ interpretation of negation. Results are consistent with the idea that although subjects in the standard task are focussed on the distinction between cases that fit the rule and ones that do not, there is evidence that emphasising the mapping of the antecedent/consequent onto back/front of the cards is sensitive to these manipulations. The negation interpretation task reveals striking divergences between subjects’ interpretations and the classical model assumed in the literature, and these differences are interpretable in terms of default logic. A few conditions were originally designed as controls only to end up generating striking results of their own. Colour is used in the truth conditional semantics of the rules (black/white replaces number/letter or vowel/consonant) instead of being used as mere information packaging. Sizable increases in ‘classical competence’ responses are observed and this is interpreted in terms of the non-hierarchical structure of the properties used. Studies using LSA and a novel tensor network operating on a database of rules gathered from selection task literature show conclusively that higher frequencies of function words appear in descriptive rules than they do in deontic rules. This thesis concludes that it is possible to direct subjects towards various interpretations on the task through the use of semantic manipulations that include but are not restricted to the ones observed in this work. Issues that include resolving the anaphora in the problem, the hierarchy of the structure of properties and the negation of clauses clearly influence the interpretation subjects arrive at which in turn affects their reasoning and responses. Results indicate that subject assumptions concerning negatives are different from those made by experimenters which inform all major theories including Mental Models, Information Gain and those based on mental logic. In particular, subjects’ most frequent selections indicate that they are selecting an implicit negative which is the opposite of what is expected by the principle of truth of Mental Models.