Some effects on generalization gradients of tandem schedules

The relationship between training conditions and stimulus generalization gradients was examined using tandem schedules of reinforcement. Schedules were selected so that frequency of reinforcement and rate of responding were varied somewhat independently of each other. A peak-shift in the generalization gradient was obtained when extinction had been associated with one of the stimuli. No comparable peak shift was obtained when there were equal response rates in the training stimuli even with dissimilar frequencies of reinforcement. The data imply that response rates at the end of training, rather than reinforcement frequency per se, determine the characteristics of the generalization gradient.     

Force gradients in stimulus generalization

Pigeons were trained to peck with differing force requirements on a key specially designed to measure and control force of pecking without changes in the mechanical threshold. Following training on a key illuminated by a single wavelength, a generalization test was given. Force and rate gradients were obtained. Force gradients were shown to have a form similar to rate gradients. The slope of the force gradients was a direct function of the force of responding at the conclusion of training. Rate gradients were independent of the force of responding.     

Some effects of response-independent reinforcement on auditory generalization gradients.

Two groups of six rats received discrimination training with two auditory stimuli differing in intensity. During one stimulus, the schedule was variable interval; during the other, it was either variable time or extinction. Both the variable time and extinction schedules resulted in differential rates of responding in the presence of the two stimuli. Extinction resulted in an earlier and more stable difference. Stimulus generalization gradients obtained along the noise-intensity dimension revealed peak shift with both procedures. In addition, a secondary peak to stimuli in between the two training stimuli occurred with the variable-time schedule.     

Local contrast and maintained generalization.

Pigeons received variable-interval reinforcement for key pecking during presentations of horizontal and vertical line-orientation stimuli, while pecks during five intermediate orientations were extinguished. Lowest peck rates were observed during presentations of negative stimuli adjacent to the positive orientations while peck rate during 45 degrees (the intermediate negative orientation) was relatively high, i.e., there were negative contrast shoulders. When peck rates were manipulated in the positive orientations, peck rate in neithboring orientations changed in the opposite direction. Contrast shoulders faded after prolonged training. A second type of contrast, local contrast, was correlated with similarity of preceding stimulus and different average peck rates during different stages of the discrimination process. The data suggest that sequential local contrast accompanying the formation of a discrimination contributes to the form of generalization gradients. Blough's model of stimulus control predicts the changes in gradient form described here, but may not accurately depict the underlying process responsible for gradient form.     

Auditory generalization gradients for response latency in the monkey

Two monkeys were trained to press and hold a response key in the presence of a light and to release it at the onset of a pure tone. Initially, all responses with latencies shorter than 1 sec were reinforced without regard to the frequency of the pure tone, and the intensity of the pure tone that resulted in equal latencies at each frequency was determined. The second stage of the experiment consisted of discrimination training, during which releases to one pure-tone frequency (positive stimulus) were reinforced and releases to a second frequency (negative stimulus) were extinguished. Median latencies to the negative stimulus slowly increased as did the variability of the latency distribution for the negative stimulus. There was no evidence of a concurrent decrease in latencies to the positive stimulus indicative of behavioral contrast. The third part of the experiment consisted of determining maintained generalization gradients by increasing the number of nonreinforcement stimuli. The gradients that eventually resulted showed approximately equal latencies to all frequencies of the negative stimulus and shorter latencies to the positive stimulus frequency.     

Stimulus generalization of behavioral history: Interspecies generality and persistence of generalization gradients

Four pigeons were exposed to a tandem variable-interval (VI) fixed-ratio (FR) schedule in the presence of a 50-pixel (about 15 mm) square or an 80-pixel (about 24 mm) square and to a tandem VI differential-reinforcement-of-low-rate (DRL) schedule when a second 80-pixel or 50-pixel square was present. The values of the VI and FR schedules were adjusted to equate reinforcement rates in the two tandem schedules. Following this, a square-size continuum generalization test was administered under a fixed-interval (FI) schedule or extinction. In the first testing session, response frequency was a graded function of the similarity of the test stimuli to the training stimuli for all pigeons. These systematic generalization gradients persisted longer under the FI schedule than under extinction.     

Temporal generalization gradients following an interdimensional discrimination protocol

We investigated the effects of interdimensional discrimination training in the temporal generalization gradient. In a matching-to-sample task, pigeons learned to choose key S after a T–s houselight sample and key NS in the absence of the houselight sample. For one group of pigeons, T?=?20?s; for another, T?=?10?s. Subsequently, houselight duration was varied to obtain temporal generalization gradients. Results showed that (a) proportion S increased as houselight duration ranged from 0?s to T?s and then remained high for houselight durations longer than T; (b) the gradients were well described by negative-exponential functions; (c) these non-flat gradients were present from the beginning of testing, and; (d) the average gradients obtained with T?=?20?s and T?=?10?s overlapped when plotted in relative time. We conclude that temporal control does not require explicit discrimination training along the temporal dimension, and that temporal generalization gradients obtained with an interdimensional protocol show the scalar property of timing. We discuss how these findings challenge current models of timing.     

Concurrent generalization gradients for food-controlled and shock-controlied behavior

In the presence of a bright light, monkeys were trained to press a lever to avoid shock and to pull a chain for food reward. After gradients of generalization to other light intensities had been determined for each response, gradients were subsequently secured after training in a brightness discrimination and under several free-shock conditions. The following results were obtained: (1) Generalization gradients prior to discrimination training were much steeper for the food-controlled response than for the shock-controlled response. This finding was confirmed in another study in which rats served as subjects. (2) After discrimination training, both gradients became much steeper, but the avoidance gradient still showed more generalized responding than that of reward. (3) After a period of continuous testing with all the different test intensities, the two gradients became even steeper. In addition, differences between the two gradients virtually disappeared. (4) The intermittent delivery of free shocks during a previously non-shocked light intensity radically affected the shape of the avoidance gradient, just as the addition of an avoidance contingency did during the same, previously non-shocked, light intensity.     

Contrast and stimulus generalization following prolonged discrimination training

Different groups of pigeons received discrimination training in which the reinforcement-associated and extinction-associated stimuli were respectively either (a) a line tilt vs a blank key, (b) a blank key vs a line tilt, or (c) two different line tilts. The high response rates that developed to the positive stimulus in all groups during discrimination learning were maintained over 64 sessions of training. After these sessions, all subjects were tested for stimulus generalization along the line-tilt dimension. Gradients of relative (per cent) generalization around the stimulus associated with reinforcement (so-called excitatory gradients) and around the stimulus associated with extinction (so-called inhibitory gradients) were as steep as they typically are after much briefer training periods. These results do not support several of Terrace's predictions on the basis of the hypothesis that emotional responses develop to the stimulus associated with extinction during discrimination learning with errors, but eventually dissipate after extended training.     

The interaction of temporal generalization gradients predicts the context effect

In a temporal double bisection task, animals learn two discriminations. In the presence of Red and Green keys, responses to Red are reinforced after 1-s samples and responses to Green are reinforced after 4-s samples; in the presence of Blue and Yellow keys, responses to Blue are reinforced after 4-s samples and responses to Yellow are reinforced after 16-s samples. Subsequently, given a choice between Green and Blue, the probability of choosing Green increases with the sample duration-the context effect. In the present study we asked whether this effect could be predicted from the stimulus generalization gradients induced by the two basic discriminations. Six pigeons learned to peck Green following 4-s samples (S(+)) but not following 1-s samples (S(-)) and to peck Red following 4-s samples (S(+)) but not following 16-s samples (S(-)). Temporal generalization gradients for Green and Red were then obtained. Finally, the pigeons were given a choice between Green and Red following sample durations ranging from 1 to 16 s. Results showed that a) the two generalization gradients had the minimum at the S(-) duration, an intermediate value between the S(-) and the S(+) durations, and the maximum at the S(+) as well as more extreme durations; b) on choice trials, preference for Green over Red increased with sample duration, the context effect; and c) the two generalization gradients predicted the average context effect well. The Learning-to-Time model accounts for the major trends in the data.