Psi man mascot

Recognition Testing

Serial learning and free recall are examples of recall testing. The participant in a memory experiment is exposed to information, waits for a period of time (the retention interval), then tries to remember it.

A different approach is to present a stimulus, wait for a period of time, then present it again to see if the research participant recognizes it. Typically, in recognition testing, the subject sees a series of test items and must indicate whether each item is "old" (from the earlier list) or "new" (not shown earlier).

In recognition testing, What is an "old" item? A "new" item?

Notice this does not correspond to what we usually call recognition. Usually if we do not recognize an item, that means we do not remember ever seeing it before in our lives.

In recognition memory research, a subject is typically asked to recognize an item seen earlier in the experiment. It may be a familiar word the subject has known since childhood.

To experience a simple recognition test in the role of a subject (or "participant" as researchers are more likely to say today) see if you can detect the old items in the following list. Three of the syllables appeared on the previous page.

BOF   WUJ   YOZ   VAF    MIB   DUR   RIQ   SOZ   ZYQ

Could you recognize the items you saw before? Probably so, because (as a rule) recognition memory is more sensitive than recall. Chances are you recognize the three "old" syllables, even if you might not recall them in a free recall test.

(Perhaps you would recall them in a free recall test, if you studied the "mibery" example. Natural language mediation is very helpful.)

Recognition succeeds when recall may fail. Most people can recognize the names of their grade-school teachers, even if they cannot recall them in a free recall situation.

Of all the memory testing procedures, relearning is the most sensitive. It can show the effects of memory when recognition fails. Recognition is next most sensitive; it can show the effects of memory when recall fails. Free recall is the least sensitive or most demanding measure of memory.

How do relearning, recognition, and recall compare in sensitivity, as measures of memory?

Most psychologists agree recall is harder than recognition because retrieval is more of a problem in recall testing. Recognition testing makes retrieval easy by presenting you with a stimulus.

As long as you encode the stimulus the same way during testing as you did during the acquisition (learning) phase, recognition is virtually automatic. This is particularly true when to-be-remembered-items are distinctive (easy to tell apart from other items).

Why is recognition more sensitive than recall?

Recognition failure does occur some­times. People do not always encode stimuli the same way on two different occasions.

Recognition failures take place if your current encoding of a stimulus does not match your previous encoding of the same stimulus. Then you may think you did not see it previously, when actually you did (a false negative).

For example, the nonsense syllable ZYQ might look unfamiliar to you. So you say you did not see it earlier, even though it was on the list you were asked to memorize (a false negative error).

If you think you saw something before, but you did not, that would be the other type of error (a false positive). For example, you might think the syllable BIF looked familiar from the earlier list of nonsense syllables, although the actual stimulus was BOF.

When does recognition failure take place?

A change of context can cause partial recognition failure, where you have a sense of familiarity but cannot quite "place" where you saw something. This sometimes happens when you see a person in an unusual setting.

For example, you might recognize some­body at a sporting event, but be unable to recall where you have seen that person previously. It might turn out the person helped you several times at a retail store. The context is different enough to prevent you from recalling that fact.

Similarly, you might recognize a person's face but be unable to recall the name. That would be a recognition success and a recall failure.

Even when the criterion of success is clear and simple, like deciding whether you have seen a face before, recog­nition memory can be probabilistic in nature. You might be "70% sure" you recognize a face.

This would be like saying, "When I feel this level of confidence in my recog­nition, I might be wrong 3 out of 10 times."

How is recognition "probabilistic" in nature?

Subjective judgments do not always correspond with reality, of course. You could be 90% sure a stimulus was familiar when viewing it for the first time. That would be an example of a false positive recognition error.

A person who claims to be guessing, in a recognition task, usually performs at better than random levels of accuracy. This is a typical finding.

Perhaps some sort of unconscious, implicit learning is at work, when a person "knows more than they know." The person has begun to grasp a pattern unconsciously, but the results of this computational activity are not yet available to consciousness.

Because people usually guess at better than random levels in a recognition test, students taking multiple-choice tests should guess rather than leave items blank. The only time this is might be a bad strategy is when students are penalized more heavily for wrong answers than for leaving items blank.

Why does a student who guesses an answer have a good chance of getting it right?

The probabilistic nature of recognition puts it in the same category as other processes in which a person picks a signal out of background noise. The Theory of Signal Detection was designed for situations like that.

The Theory of Signal Detection was discussed in chapter 4. It is a mathematical description of decision-making in uncertain environments.

That framework works well with recognition testing. The d-prime statistic can provide a bias-free measure of the true sensitivity of an organism's recognition abilities.

Forced Choice Recognition Testing

The simplest way to test recognition is to present a stimulus and ask, "Do you recognize this?" Unfortun­ately, this yes/no method does not take into account the probabilistic nature of recognition.

On a true/false quiz, some people are very conservative and will say False if they are not absolutely sure of their recognition judgment. They want to avoid false positives, and that is their bias.

What is the drawback of the yes/no method?

Others have a bias toward saying True. Perhaps they want to be agreeable. This can vary between ethnic groups.

One pollster found that people of Japanese ancestry had a bias toward saying Yes in opinion polls, perhaps because it was con­sidered more polite. This threw off the accuracy of results.

In general, people differ in their bias toward saying Yes or No on a recognition test. One person might require 70% confidence before saying Yes (or True), another might only say Yes when 90% sure.

Experimenters cannot compare the recognition accuracy of different people unless they know the confidence level each requires for a Yes judgment, and that is something that could change even within an experiment.

One solution is to use the Theory of Signal Detection. A person could be presented with a range of biasing conditions so the D prime statistic could be calculated.

In the forced-choice method, what is the target? The distracter?

However, a simpler solution is to use a forced-choice method of recognition testing. In a forced-choice procedure, subjects are faced with two or more stimuli, only one of which is "old."

The old item is called the target item; the other items are called distracters. To give yourself a forced choice recognition test, decide which of the following items was used earlier in this chapter.

VEW   DOX   XAJ

What are advantages of the forced-choice method?

In a forced choice test, you must guess even if you are not sure. This solves the problem of response bias, because every­body must say Yes to one and only one item.

This also makes forced-choice testing a very sensitive measure of recognition memory, because people are forced to make a guess even if they are not sure about the answer. Usually they will guess at a better-than-chance level.

The "Almost Limitless" Capacity of Recognition Memory

How sensitive is human recognition memory? Standing (1973) tried to find its upper limits. He gave people a single exposure to pictures cut from magazines then tested them in a forced-choice situation using similar pictures as distracters.

Standing found that subjects had no difficulty learning 10,000 pictures in one sitting. They could recognize them the next day and tell them apart from the distracters. Animals like chimps and pigeons proved to have equally impressive recognition memory for images.

He concluded, "The capacity of recognition memory for pictures is almost limitless." In similar research, Shepard (1967) found that word recognition was over 90% accurate in forced choice situations.

What did Standing conclude, and how did he arrive at this conclusion?

The excellence of recognition memory creates a potential problem for teach­ers. A poorly constructed multiple-choice test resembles a forced-choice recognition test.

Suppose a teacher puts familiar vocabulary words from lectures into correct answers on a multiple choice test, but not in any of the distractors. According to Shepard's research above, a student who simply recognizes the words (without understanding the concepts) could pick out the right answer nine out of ten times.

Students are efficiency experts. If they learn they can ace tests simply by familiarizing themselves with words, some will do that instead of studying.

Therefore a teacher who con­structs tests carelessly, making it easy to pass a test using recognition memory, subverts the goal of encouraging in-depth understanding. No wonder "multiple guess" tests have such a bad reputation among teachers with high standards.

How can a multiple-choice test be constructed to measure more than recognition?

However, it is not too difficult to create a multiple-choice test that measures comprehension rather than familiarity with words. The test writer must construct distracters that trigger a recognition response. More than one answer must "sound right" to a student who has merely skimmed and does not understand the concepts.

That is easy to accomplish by using familiar words from the text in the wrong answers. Of course, the instructor must take care that only one answer is actually correct, and the correct answer should be easy to select for a student who understands the concept targeted by the question.

Also desirable: the correct response might not repeat the exact language of the lecture or book. It might use paraphrasing (expressing the same idea in different words).

The goal is to encourage a correct understanding of the subject matter. Recognition memory alone should not be enough to guide a student to a correct answer.

Recognition Errors

In an article titled "Attributes of Memory" (1969) the influential verbal learning researcher Benton Underwood summarized evidence about false recognition errors. He pointed out that they occur when words have similar attributes.

Even opposites can be falsely mistaken for each other. If you see the word UP in a list of words, you might falsely recog­nize the word DOWN on a recognition test the next day. Both are directions, and both involve the vertical dimension, so they have many features in common, despite being opposites.

Underwood tested other relationships that might provoke false recognition errors. He obtained the results shown in the following table.

RelationshipList word(s)Test wordError rate
Oppositesbottomtop.32
Associatesbread, buttercrumbs  .28
Categories/Examplesmaple, elm, oaktree.25
Sense Impressionsdoughnut, globeround.08 (not signif.)

It shows, for example, that people who saw the word "bottom" on a list would falsely recognize the word "top" about 32% of the time. Conceptual features (related to meaning) proved more likely to produce recognition errors than per­ceptual features (related to appear­ance).

What sorts of relationships did Underwood discover to provoke false recognition?

Underwood summarized his conclusions as follows:

A memory is...a collection of at­tributes that serve to discrim­inate one memory from another and to act as retrieval mech­anisms for a target memory....Encoding represents the process by which the attributes of a memory are established.

Why was Underwood's conversion to the encoding perspective considered important?

When Underwood wrote that in 1969, other researchers took note. Under­wood was a grand old man of memory research, highly respected. He had always accepted the assumptions of S-R theory that went back to the days of Ebbinghaus.

Underwood was joining the encoding revolution. The old S-R (stimulus-re­sponse) theories were being replaced. Now researchers had to consider how subjects interpreted, represented, and manipulated information.

How did changes in memory research foreshadow changes in other areas of experimental psychology?

Memory researchers were among the first psychologists to use information processing concepts. Before long, the encoding revolution in memory research was followed by the cognitive revolution in experi­mental psychology as a whole.

---------------------
References:

Shepard, R. N. (1967). Recognition memory for words, sentences, and pictures. Journal of Verbal Learning and Verbal Behavior, 6, 156-163.

Standing, L (1973). Learning 10,000 pictures. Quarterly Journal of Experimental Psychology, 25, 207-222.

Underwood, B. J. (1969). Attributes of memory. Psychological Review, 76, 559-573.


Write to Dr. Dewey at psywww@gmail.com.


Don't see what you need? Psych Web has over 1,000 pages, so it may be elsewhere on the site. Do a site-specific Google search using the box below.