Associative models of discrimination and generalisation: an evaluation in Pavlovian conditioning and causal judgement paradigms
Zusammenfassung der Projektergebnisse
Human learning can be applied in areas of basic reflex functioning as well as to more complex behaviours such as decision making and planning. Understanding learning is crucially important in society today, from basic survival to success in the knowledge economy. In the present project we studied Pavlovian eyeblink conditioning (EBC) in humans in order to test the predictions of current theories of associative learning against each other and to examine the extent and the conditions under which human associative learning is flexible along a dimension of elemental to configural processing. Present theories of learning might be divided in so-called elemental theories such as the replaced elements model (REM) of Wagner (2003) or the model of Harris (2006) on the one hand and configural theory (CT) suggested by Pearce (1994) and extended (ECT) by Kinder and Lachnit (2003). In order to “design” special situations for which the theories under consideration make clear different predictions, we manipulated two factors potentially constraining the kind of processing, the structure of the task (Exp. 1, 2, 4) and properties of stimuli (Exp. 4). Exp. 1 and 2 established a firm base for the conclusion that humans may generalise very little between stimuli. The only model being able to deal with all observations was a configural model, ECT, the model suggested by Pearce (CT) extended by a discrimination parameter with a rather high value (high discrimination = little generalisation). The findings of Exp. 1 and 2 clearly converged with earlier observations in causal or predictive learning, very different paradigms at a first glance. Experiment 4 was the by far most challenging part. It adapted the seminal work of Shepard et al. (1961) in categorization learning to basic learning measures, such as EBC. Furthermore, two new theories were tested which bear on the question, often raised in the context of this task, whether human learning uses hypothesis testing or can be explained by associative learning mechanisms. Last but not least, the manipulation of properties of stimuli could elicit different modes of processing. Thus, especially Exp. 4 promised to shed new light on important questions. Although we invested much more work then planned, however, the outcome was a disappointment. The task did not suit the capabilities of the participants. They were unable to show substantial discrimination between stimuli. As a consequence, we not at all could conduct the fine graded analyses necessary for answering our hypotheses.
Projektbezogene Publikationen (Auswahl)
- (2009). Similarity does not affect Feature- Negative discriminations in humans. XIII and XIV. Associative Learning Symposium, Gregynog, Wales, GB
Thorwart, A., Glautier, S., & Lachnit, H.
- (2009). The impact of separable and integral stimulus dimensions on predictive learning: Implications for elemental and configural theories of learning. XIII and XIV. Associative Learning Symposium, Gregynog, Wales, GB
Lachnit, H. & Uengoer, M.
- (2010). Convergent results in eyeblink conditioning and contingency learning in humans: Addition of a common cue does not affect feature-negative discriminations. Biological Psychology, 85, 207-212
Thorwart, A., Glautier, S., & Lachnit, H.
- (2010.). Reduced summation with common features in causal judgements. Experimental Psychology, 57, 252-259
Glautier, S., Redhead, E. S., Thorwart, A., & Lachnit, H.
