Key Points
- Reinforcement follows the response only when the target cue is present, which is how a cue gains control over the response over many trials.
- Simple discriminations turn on one cue; conditional ones, such as receptive identification or matching, require attending to two cues at once.
- Recording which wrong choice a learner makes shows whether a target is truly discriminated or answered from position, order, or an unintended prompt.
Discrimination Training Explained
Much of what a learner needs to do depends on responding to the right thing: touching the cup when asked for the cup and not the spoon, saying 'dog' for a dog and 'cat' for a cat, waiting at a red light. Discrimination training teaches those distinctions. A response is reinforced when a particular stimulus is present, called the discriminative stimulus or SD, and is not reinforced when other stimuli are present. Over repeated trials the learner comes to respond to the SD and not to the others, and the response is then under stimulus control.
Discriminations range from simple to conditional. A simple discrimination has one relevant cue, as when a learner says 'ball' on seeing a ball. A conditional discrimination depends on two. In a receptive identification task with several pictures on the table, the correct picture changes with what the instructor says, so the learner has to attend to both the instruction and the array. Matching-to-sample, sorting, and following multi-part directions are conditional discriminations, and they generally take longer to teach.
Teaching usually starts by making the distinction easy and then making it harder. An instructor might begin with one target and a clearly different distractor, then add distractors that look more alike as accuracy holds. Errorless teaching, which supplies prompts and fades them as the learner succeeds, keeps mistakes from becoming part of what is learned. Discrimination training is often delivered through discrete trial training, where a common progression moves from massed trials of one target, to that target among distractors, to a random rotation of targets. The random rotation is the step that shows whether the learner is discriminating at all.
The errors carry information. A learner who always picks the picture on the left is responding to position. A learner who is correct only when targets come in the same order is responding to sequence. A learner who waits for the instructor's glance or a slight gesture is responding to a prompt the instructor may not realize they are giving. Each of these can look like progress on a percentage-correct graph, which is why many teams record which incorrect choice was made, rotate positions on purpose, and count only independent responses toward mastery.
Discrimination has to be balanced against generalization. Discrimination narrows a response to the right cue; generalization widens it to every version of that cue the learner will meet, so 'dog' should cover a beagle, a cartoon dog, and the neighbor's dog, and still not the cat. Teaching with several examples of each target, instead of a single picture, keeps a discrimination from becoming so narrow that it only holds for the materials used in sessions.
The procedure runs under much of an ABA curriculum: receptive language, labeling, matching, early reading and math, and social skills such as noticing whether someone is available to talk. It also appears in behavior reduction, where a learner may be taught that a request will be honored under some conditions and not others, with a clear signal for each. In every case the program names the target cues, the distractors, and the mastery criteria, and the session data records responses trial by trial.