Addition versus deletion as a signal

Abstract

Subjects typically show superior discriminative performance when a distinguishing feature appears on reinforced rather than nonreinforced trials. The phenomenon is usually attributed to the relative predictiveness of the reinforcer by different stimulus elements. However, stimulus' addition may be more effective than stimulus deletion as a signal. By removing the standard intertrial intervals, we made addition and deletion equally predictive of the reinforcer in four operant experiments involving between-and within-subject comparisons. Pigeons consistently performed better on operant discriminations when the addition rather than deletion of an auditory or visual stimulus served as the cue for food. This general finding persisted despite manipulation of the relative duration and localizability of the signal. Thus mere presence as opposed to absence plays a role in the feature-positive superiority, an outcome that may reflect a fundamental, biologically based difference between addition and deletion as effective signals of reinforcement.In a typical experiment on the feature-positive effect, the subjects receive two kinds of brief trials: one (S+) followed by a reinforcer and the other (S-) followed by nothing. The S+ and S-share a common element-say the illumination of a green light on a pigeon's response key-but a second element, which is the distinguishing feature-say a small white square on the key-is present during one type of trial and absent during the other. Neither the common nor the distinguishing feature appears between trials (the intertrial interval, ITl).Subjects perform considerably better when the feature is present on positive trials (FP) than on negative trials (FN). This result occurs over a wide range of settings involving different species, reinforcers (USs), and stimulus modalities. Often there is little or no evidence of any FN learning, especially when the feature is not very salient (see Hearst, 1978Hearst, , 1984Jenkins & Sainsbury, 1969, 1970Newman, Wolff, & Hearst, 1980). Beyond the intrinsic interest of the FP superiority, analysis of the two discrimination arrangements is potentially valuable for understanding several general issues in discrimination learning (see also . The set of experiments to be reported here mainly addresses two such issues: (1) the predictiveness of different stimulus elements with respect to some outcome, and (2) the effectiveness of presence and absence as positive signals.Prior explanations of the unexpected asymmetry between FP and FN discrimination performance have stressed the role of differential predictiveness (see, for example, Hearst & Jenkins, 1974;Staddon, 1983 standard FP subjects, the distinguishing feature is the best positive predictor of food in the situation, whereas for FN subjects the common element is. If the subjects mainly notice and respond to elements that they learn are the most reliable predictors of the US, they will perform better on FP discriminations. In that arrangement the best predictor of the US is present only on S+ trials, wh...

Simplicity and generalization: Short-cutting abstraction in children’s object categorizations

“…The logic from recognition to generalization might go like this: detecting that an item is different may hinder generalization whereas failure to detect a difference may result in generalization. Since many studies of feature additions/deletions have found that additions are more easily detected than deletions (Agostinelli et al, 1986;Miranda et al, 1992;Hearst & Wolff, 1989 for a review), our results may be a reflection of the idea that more salient additions hinder children's generalizations.…”

Spurious Correlations in Mathematical Thinking

“…The idea that students will pick up on spurious correlations between an absent feature and its correlated algorithm is inspired from the literature on animal learning and behavior (Hearst & Wolff, 1989), which shows that animals can learn to use the absence of a stimulus as a signal (e.g., a predictive cue for food) even though the presence of a signal serves as a stronger cue. Finding that not only symbols, but the absence of particular symbols, can cue the use of a solution strategy would provide additional evidence that even experienced students rely on spurious correlations in their performance.…”

Feature-positive and feature-negative learning in honey bees

“…Experimental comparison of addition and deletion tasks for animals other than humans has been notably rare (reviewed by Hearst and Wolff, 1989), and this problem continues to be understudied (Miranda et al, 1992;Abramson and Buckbee, 1995). However, it is clear that at least some species in a variety of vertebrate groups can solve both feature-positive and featurenegative problems.…”