Студопедия

Главная страница Случайная страница

КАТЕГОРИИ:

АвтомобилиАстрономияБиологияГеографияДом и садДругие языкиДругоеИнформатикаИсторияКультураЛитератураЛогикаМатематикаМедицинаМеталлургияМеханикаОбразованиеОхрана трудаПедагогикаПолитикаПравоПсихологияРелигияРиторикаСоциологияСпортСтроительствоТехнологияТуризмФизикаФилософияФинансыХимияЧерчениеЭкологияЭкономикаЭлектроника






Experimental paradigm of discrimination learning






 

Experimental studies of conceptual abilities in animals demonstrate the substantial range of these abilities as well as their limitations. Such abilities range from categorisation on the basis of shared physical attributes, associative relations and functions, to abstract concepts as reflected in analogical reasoning about relations between relations (Thompson and Oden, 2000). A great deal of experimental procedures for studying species-specific and age-specific abilities to judge about things and relations are based on discrimination learning. It is a general and already trivial sentence that discrimination is a basic element of decision making in the animal kingdom. Organisms must respond appropriately in an enormous number of stimulus situations and learn which stimuli are significant for their goals and which are not. Relatively simple methods based mainly on discrimination learning, serve for solving complex problems concerning the way in which animals represent knowledge. Moreover, the discrimination of presumably more abstract relations commonly involves relatively simple procedural strategies mediated by associative processes likely shared by many species.

When discrimination learning is examined in laboratory, the subject is reinforced to respond only to selected sensory characteristics of stimuli. Discriminations that can be established in this way may be quite subtle.

Ways in which animals solve discrimination problems have been the focus of interest since early experiments conducted by Kinnaman (1902), Pavlov (1917, described in: Pavlov, 1927) and K`hler (1918). The paradigm of discrimination learning appears to be fruitful for studying animal intelligence because simple discriminations lay in the basis of concept formations of highest order, such as “sameness” or the “relation between relations” and thus can be considered “atoms of cognition”.

Skinnerian technique of discrimination learning (discriminative operant) involves the use of environmental cues to signal the availability or unavailability of positive and negative reinforcement as well as punishment (see Chapter 6). For example, for a horse a trial begins when an experimenter brings the horse to the entrance to the testing stall and releases hold on its halter. The horse enters the testing stall unescorted, faces the stimulus wall, and selects one of stimuli by pushing a panel by its nose. When the correct stimulus panel is pushed it opens inward, and the subject is allowed access to a food bowl. The incorrect stimulus panel is locked from behind and would not move if pushed, although the bowl behind it also contains the reward.

One more technique that was used for many years to study discrimination learning is a jumping stand. This apparatus is known as Lashley Jumping Stand referring to: Lashley, 1938. The discriminative stimuli are situated behind ledges that the subject must jump onto from the central platform. If the subject should jump onto the ledge in front of the correct stimulus, then it can push the stimulus card over and gain access to a reward. The position of the discriminative stimuli is normally varied randomly from trial to trial, which ensures that subjects learn that food is behind a certain stimulus, rather than in a particular location.

In experiments with non human and human primates Wisconsin General Testing Apparatus (WGTA) described in Harlow (1949, 1959) is used for a battery of tests including simple and reversal discrimination, matching to sample and the so-called learning set, that is, animals ability to “learn to learn”. We may, for example, try to teach an animal to systematically switch to the other stimulus after every successful response, a discrimination procedure that requires multiple problems and a transfer set-up. Such work on learning sets (also called learning-to-learn) has been done by Harlow (1949), and we will examine it more closely in Part VI. Another example of a complex task which is implemented on WGTA is the Wisconsin Card Sorting Task, a test of the ability of human subjects to flexibly alter their responses to the same stimuli. The sorting rule varies surreptitiously every few minutes and thus any given card can be associated with several possible actions; the correct response is dictated by whichever rule is currently in effect. Impairment on this task is a classic sign of PF damage in humans (Milner 1963), and marmoset monkeys with PF lesions are impaired on analogous tasks (Dias et al. 1997)

The ways in which subjects are presented with stimuli and reinforcement differ in accordance with different schemes of experiments. If there are two or more stimuli present only one of which needs to be responded to, then we have what is termed a simultaneous discrimination. The task in a simultaneous discrimination is to determine which stimulus to respond to amongst the various stimuli present (Fig. V-1). Normally, the animal will initially respond to both stimuli, and thus discover that one of these consistently fails to yield the desired outcome. Over the course of training, then, we will see responding first rise to both stimuli, and then drop off to the S- (a negative stimulus). This is a use of normal discrimination training. If many stimuli are present from which the subject has to make a choice, the task is termed concurrent discrimination. In a variant of simultaneous discrimination procedure, a technique termed errorless discrimination training, however, we typically start with the S- being so reduced in intensity compared to the S+(a positive stimulus) that it is virtually non-salient (here + means that selection of this item is rewarded and “-“ means this item is not rewarded). Over the course of a large number of trials, we gradually increase the intensity of S- until it equals that of the S+. If this is done slowly enough, then we may discover that our subject has never actually made a false response to the S- (hence the name errorless).

An alternative approach is to present just one stimulus at a time. In this case, we have a successive discrimination. Successive discrimination techniques open up several different possibilities. If the successive discrimination is like the simultaneous discrimination in that only one stimulus is associated with a reinforcement, then we have what is termed a “ go/ no go ” training procedure, with rate of response serving as the measure of learning. Subjects can be trained to respond differently to different stimuli in a choice situation. Typically, the trial is initiated by the appearance of the discriminative stimulus. For example, in experiments with pigeons, a peck at one of two stimuli illuminates the choice keys signalling the opportunity to make a choice. As it was described in Chapter 10, in this procedure the intermediate stimulus serves as a “go” signal. In this case, you go (respond) when the correct stimulus (the S+) is present, but do not go (withhold a response) when any other stimulus is present. In successive discriminations, however, we also open up the possibility of requiring several different responses, depending on which stimulus is present. A triangle, for example, could be used to signal pressing a left-hand button, whereas a circle could be used to signal pressing a right-hand button. In the simplest version of this situation, a reinforcement can be obtained on each trial (assuming you are using a continuous reinforcement schedule) so long as the animal knows which response to make to which stimulus. This type of set-up is referred to as a choice situation. Unlike the simultaneous condition or the go-no go situation, each stimulus in the choice situation could, in principle, be associated with the same degree of excitation or inhibition. Or put another way, in a choice situation, the stimulus acts as an occasion setter indicating which response is appropriate.

Conditional discrimination involves successive discrimination. In conditional discrimination, at least two stimuli (or stimulus dimensions) are typically present. The reaction to one will depend on the presence of the other. For example, in an experiment by Nissen (1951), chimps were trained to discriminate between a large square and a small square. Only one of these yielded reinforcement, but which one that was depended on a second stimulus characteristic. Because this was a successive discrimination procedure, each trial involved just one of the four complex stimuli below:

Stimulus Outcome

large black square reinforcement

small black square no reinforcement

large white square no reinforcement

small white square reinforcement

In this example, if the squares were black then large was the S+ and small the S-; but if the squares were white, this assignment reversed. So, colour in this case was the occasion setter informing the animal about meaning of size.

Assigning a transfer tasks to the animals, we ask whether or how prior discrimination training on one problem affects later discrimination training with another problem. Sometimes problems are very similar. A common task involves a reversal shift, in which a subject learns to do the exact opposite of what it did earlier. For instance, if it was trained to respond to a square but not a triangle, in a reversal shift, it would have to respond to the triangle, but not the square. Sometimes the problems are very different, and sometimes the training involves combining problems. In a technique called acquired distinctiveness of cues, for example, we try to speed up the process of discrimination by compounding the two stimuli we want the animal to discriminate with very different stimuli that we know from earlier training are easily distinguished.

A way of looking at the ease and success of discrimination training involves matching to sample procedure (MTS). This is a method to study how animals learn not only about stimuli themselves but about their relationships as well. For this technique a subject is presented with a sample stimulus, and, a short while after it has been turned off, two comparison stimuli are shown. When one of the comparison stimuli is the same as or identical to the sample stimulus, the task often is called an identity matching procedure. To gain a reward, the comparison stimulus that matches the sample must be selected. If a subject can also perform adequately when novel stimuli replace the training stimuli, the performance is often called generalised identity matching or true matching and taken as evidence that training generated an identity concept.


Поделиться с друзьями:

mylektsii.su - Мои Лекции - 2015-2024 год. (0.007 сек.)Все материалы представленные на сайте исключительно с целью ознакомления читателями и не преследуют коммерческих целей или нарушение авторских прав Пожаловаться на материал