Human feature-positive discrimination learning: (not the) conditions for conditioning conditionals

The main goal of this doctoral project is to explore the effect of a variety of factors on human feature-positive discrimination learning and to verify the generality of principles that describe the emergence of Pavlovian modulation. The term feature-positive discrimination refers to a procedure or learning task during which a participant, human or non-human animal, has to learn to respond to a so-called target stimulus in the presence of a feature (FT+), but not in the absence of this feature (T-; e.g., Sainsbury, 1971). Formally, this is described as FT+|T-. In the conditioning literature, it is strongly associated with the phenomenon of occasion setting or Pavlovian modulation (for reviews, see Holland, 1992; Swartzentruber, 1995).Pavlovian modulation refers to a hierarchical knowledge representation which may emerge during feature-positive discrimination learning. In contrast to standard elemental learning theories (e.g., Rescorla & Wagner, 1972), which assume that such discrimination learning results in the formation of a direct feature-reinforcer association, a modulatory perspective proposes that the feature may acquire an indirect association with the reinforcer. Stated differently, that the feature stimulus comes to modulate responding to the target stimulus (e.g., Moore, Newman and Glasgow, 1969).The different accounts on the emergence of modulation during feature-positive discrimination learning, proposed in the animal literature, may either focus on properties of the target stimulus (e.g., Rescorla, 1988) or the feature stimulus (e.g., Holland, 1983; Schmajuk, Lamoureux, & Holland, 1998). However, they converge on a procedural level in the notion that modulation emerges if the feature is, a priori to learning the discrimination, a less likely predictor than is the target stimulus .In three research lines, we investigated whether such a procedural account may hold during human feature-positive discrimination learning. In a first series of experiments, the relative salience levels of the feature and target stimuli were manipulated. During the second series, participants received different types of pre-training. This pre-training was instantiated in terms of differences in feature and target contingency or temporal contiguity information before discrimination learning. In the third line of research, prior differences in common real-world knowledge on the causal likeliness of the feature and target, given certain reinforcers, were used.Results indicated that conditions which were hypothesized to evoke modulation, e.g., feature-positive discriminations with low-salience features, displayed, although not always convincingly, results in accordance with the properties of Pavlovian modulation (Holland, 1992). However, the conditions which were expected to be solved by means of simple excitation of the feature, as predicted by elemental learning models, also displayed one of these properties (for a similar discussion, see Dibbets, Maes, & Vossen, 2002), thus blurring a strict distinction between modulation and elemental learning or simple excitation.We conclude that a strict modulation versus simple excitation perspective does not hold, and that both within-trial emergent properties, which prominent configural models propose (e.g., Pearce, 1987, 1994, 2002), as well as between-trial emergent properties, such as abstract rules (e.g., Shanks & Darby, 1998) or a hierarchical stimulus structure (e.g., Bonardi & Jennings, 2009), are necessary to describe the observed pattern of results.nrpages: 252status: publishe