出典(authority):フリー百科事典『ウィキペディア（Wikipedia）』「2013/03/06 01:59:53」(JST)
| This article may require cleanup to meet Wikipedia's quality standards. No cleanup reason has been specified. Please help improve this article if you can. (July 2008) |
Stimulus control is said to occur when an organism behaves in one way in the presence of a given stimulus and another way in its absence. For example, the presence of a stop sign increases the probability that "braking" behavior will occur. Typically such behavior is brought about by reinforcing the behavior in the presence of one stimulus and omitting reinforcement in the presence of another stimulus. Many theorists believe that all behavior is under some form of stimulus control.[1] Verbal behavior is a complicated assortment of behaviors with a variety of controlling stimuli.[2]
Contents
|
Proper stimulus control is said to have four characteristics:
In Classical conditioning, when a conditional stimulus (CS) is repeatedly paired with an unconditional stimulus (UCS), the CS will come to elicit a conditional response (CR). In this way, the CR is a type of controlling stimulus. A standard procedure for establishing this relationship involves an arrangement such that the probability that the UCS is delivered, given the CS is presented, is equal to 1, and the probability that the UCS is delivered, given the CS is not presented, is equal to 0; P(UCS|CS)= 1 and P(UCS|~CS)=0. In other words, the UCS always follows the CS, and the UCS is never presented without the CS. The CS can be said to indicate the arrival of a UCS, and the likelihood that the CR occurs will increase when the CS is presented. When this occurs, the CR can be said to be under stimulus control.
An example experimental paradigm that illustrates this is similar to the arrangement used by Ivan Pavlov and involves measuring the amount of saliva a dog produces following a tone of a certain frequency. A tone of a certain frequency is played immediately before the dog receives some food, and after several pairings the tone will be sufficient to produce salivation. The presence of a tone increases the probability that the dog will salivate.
An important finding is that this CR will occur to variations of the CS that were not trained during the acquisition of the initial CS-CR contingency. For example, if the food was originally paired with a 500 Hz tone, the dog may also salivate (to a lesser degree) to a 450 Hz tone. The phenomenon in which a CR is elicited following a stimulus that was not trained in the initial arrangement is called respondent generalization.
This phenomenon can be experimentally verified in the following way: A tone of a certain frequency (e.g., 500 Hz) is paired repeatedly with food, until the salivary response (CR) is elicited following presentation of this frequency. Suppose that in response to the 500 Hz tone being presented (without the UCS) the dog will produce 1 ml of salivation. The experimenter could then vary the some characteristic (e.g., frequency) of the tone and measure the amount of salivation produced. For example, the experimenter could play a 600, 550, 500, 450, and 400 Hz tone, and then measure the amount of salivation produced as a result. The experimenter could then plot the amount of salivation produced as a function of the frequency of tone presented to generate a generalization gradient.
The particular “shape” of a generalization gradient can be altered by making adjustments to the schedule of reinforcement. A relatively “flat” gradient indicates respondent generalization, while a relatively “peaked” gradient indicates respondent discrimination. Respondent discrimination and respondent generalization can be conceptualized as two sides of the same coin: the more generalization that occurs, the less discrimination that occurs, and vice versa. While generalization refers to an organism displaying CR to untrained stimuli, discrimination refers to an organism not producing CR in response to untrained stimuli.
A sharper generalization gradient (respondent discrimination) can be produced through a series of explicit non-pairings. This procedure for producing respondent discrimination is very common. As before, a stimulus is presented and the UCS is delivered, this is called the CS+. Other stimuli are also presented, but do not receive reinforcement, they are called CS-.[4] In this way, the CS-, are explicitly not paired with the UCS. From the example given above, over a series of trials, the 500 Hz frequency would be reinforced with food. Additionally, different frequency tones would be presented without food. Initially, the dog may show generalization to the different tones, but over a series of trials that dog will stop salivating to the different tones and will only salivate to the 500 Hz tone. If a new gradient were generated, it would be much sharper than the original.
It is important to note that this is not the only way to obtain stimulus control in Classical conditioning. Instead only presenting food in the presence of the 500 Hz tone, one could present a large quantity of food in the presence of the 500 Hz tone, and a smaller quantity of food in the presence of other tones. Alternatively, one could present a different type of stimulus, such as a shock or a drug, in the presence of other tones.
Describing generalization as occurring to stimuli that are similar to the original CS is a common mistake that should be avoided. Generalization itself is a measure of stimulus similarity. What is perceived as similar stimuli to humans may be perceived as dissimilar to another species (or may not be perceived at all). For example, homing pigeons are capable of magnetoreception, and they may have the ability to perceive the earth’s magnetic fields. Humans do not have this ability. Pigeons demonstrate discrimination between variations in magnetic fields, while humans do not.
In classical conditioning, the CS is an initially neutral stimulus that comes to play a direct role in indicating the occurrence of a UCS after a series of pairings. In operant conditioning, neutral stimuli do not play such a direct role. Instead, initially neutral stimuli come to signal the availability of reinforcement for producing certain behaviors.[1] When a stimulus signals the availability of reinforcement it is called the SD, or discriminative stimulus. A discriminative stimulus is a type of controlling stimulus; its presence increases the probability that a behavior will occurs, because it reliably signals the availability of reinforcement. Another class of controlling stimuli in operant conditioning is called the extinction stimulus, or S∆. The extinction stimulus signals that certain behaviors will not receive reinforcement.[4]
Stimulus control can be illustrated in operant procedures. A simple arrangement involves training a pigeon to peck a key with different frequencies depending on the presence of certain colors of a light. First, a baseline rate of behavior would be established by reinforcing the pigeon every time it pecks a key. As the pigeon is pressing the key, a light could be presented and only pecks made in the presence of the light would be reinforced. Although the light was initially neutral, it becomes the SD. Over a series of trials the light will be established as indicating the availability of food, and the light will increase the probability that key pecks occur. Similar procedures have been used to train pigeons to discriminate between paintings by Monet and Picasso.[5] Other arrangements might involve different colored lights, with each light controlling different rates of behavior by having them programmed to different schedules of reinforcement. If a green light is associated with a VR 10 schedule and a red light is associated with a FI 20” schedule, the green light will control higher rates of behavior than the red light by providing higher rates of reinforcement.
Procedures for generating generalization gradients in operant conditioning are similar to those for used in Classical conditioning described above. Hanson (1959) conducted a study that illustrates generalization in operant conditioning.[6] A control group of pigeons were placed in an arrangement involving a VI schedule where they received reinforcement for key presses when a 550 nm wavelength light was illuminated. After training the pigeon to peck the key in this way such that a steady rate of behavior was achieved, the frequency of the light was varied away from the 550 nm wavelength. These frequencies were not reinforced, but responses of the pigeons were recorded as the wavelength was varied. The greater the wavelength differed from the trained stimulus, the fewer responses were produced.[6]
Additionally, there were four other experimental groups that received were placed in similar arrangements. The experimental difference being that, in addition to SD at 550 nm, they were presented with explicit non-pairings or S∆. Each of the four experimental groups was presented with S∆ at 555, 560, 570, or 590 nm wavelengths, respectively. The effect of this manipulation was to create a sharper generalization gradient. S∆ that were closer to the SD produced sharper generalization gradients.[6] In this way, introducing S∆ in operant conditioning is similar to the effect of introducing CS- in Classical conditioning. However, there is another important phenomenon that occurs in operant conditioning as a result of this manipulation: introducing S∆ in this way caused the generalization gradients to shift away from the SD. This is called a peak shift, and in addition to the shift, the number of responses produced by the experimental groups at the peak actually increased above the control group.[6]
It seems bizarre that the probability that the pigeon responds is higher at an untrained wavelength then originally trained SD. Based on an earlier theory involving inhibitory and excitatory gradients,[7] Hanson proposed that this occurred as a sort of summation between excitatory and inhibitory stimulus gradients.[6] However, the example of peak shifts described above may be an example of relational control where the pigeons were trained to peck the “greener” stimulus more frequently.[4]
In the quantitative analysis of behavior, stimulus control is examined from a number of perspectives, including Matching to Sample, and signal detection (Nevin, 1965; 1969). Matching-to-Sample is a form of conditional discrimination. In this form of conditional discrimination procedure, only one of two or more stimuli presented on other comparison keys from the sample, shares some property (e.g., shape). Responses to the similar stimulus are reinforced. In Oddity matching, a form of matching-to-sample, responses to the comparison stimulus that does not match the sample are reinforced.
全文を閲覧するには購読必要です。 To read the full text you will need to subscribe.
| 関連記事 | 「discriminative」「stimulus」 |