A computer simulation of a language conditioning of attitude paradigm

Abstract

Two mathematical models (Anderson, 1965; Chalmers, 1966) were designed to account for attitude change and impression formation data. In their experimental paradigm, words are presented to Ss to influence their attitude/impression of some hypothetical person. The models predict that different attitudes/impressions will be formed as a function of the order of presenting these words. Language conditioning of meaning (or attitude) has been studied in accordance with experimental procedures developed by Staats and Staats (1957) in which the meaning (or attitude) of a "meaningless" word is altered as a function of being paired with meaningful words. In the current study it is shown that the attitude change and conditioning of meaning paradigms are analogous, and so these mathematical models are used as the foundation of a computer simulation of conditioning of meaning. A series of computer "experiments" was run in which different conditioned meaning results were generated as a function of the order of presenting the meaningful words. Data was then collected on three groups of human Ss in an attempt to replicate the results produced by the computer simulation program. All groups of human Ss were presented the CSs in the same order and only the sequences of UCS presentations were varied. There were apparent differences in the responses from human Ss as a function of the conditioning procedures. In the ANOVA, the Sequence x CS interaction was significant (p=0.0448). In addition, there were significant differences with sex (p=0.0239) and the CS (p=0.0047) main effects. Inconsistencies in the human S data precluded a clear interpretation of the source of the significance on the CS main effect. The UCSs presented to the Ss previously had been rated on a scale defined as +l=Pleasant, +4=Neutral, and +7=Unpleasant. In the executing of computer simulated "experiments", however, some of the predicted responses were unexpectedly outside the range of +1 to +7. The mathematical models on which the simulation programs were based have no provision for restricting predicted values for the conditioned meaning responses, and therefore only the relative order of group scores was considered as the predictions made by the models with the computer simulation program. The a priori ordinal predictions of group orders were tested with one-tail t-tests. None of the 8 predictions based on the Chalmers model were significant (all p>0.05) and only 2 of 8 predictions based on the Anderson model were significant (p<0.05). Although the computer simulation programs demonstrated that the models had little predictive utility, other useful purposes were served. By using a computer simulation of conditioning of meaning, a number of "experiments" were run very quickly and efficiently. Based on these "experiments" it was predicted that changing the word order would have a significant effect. Data on human subjects bore little resemblance to the actual predicted order of group scores. However, the effect of different results as a function of word order seemed to be supported. In addition, males and females sometimes had very different ratings for the words and the sex differences was found to be significant. There are three implications for future studies on language conditioning of meaning. First, it seems advisable to have separate normative data on the ratings of the words to be used for males and females. Second, even if only one type of word (e.g., pleasant) is being presented, the order of presentation might have a significant effect. Third, it appears that the models proposed by Anderson and by Chalmers are both poor predictors of the outcome of novel data in a language conditioning of meaning paradigm.