Rule Leaning and Attribute Learning

Psychologists have used several experimental procedures to study how people make classifications. For instance, a visual image of a four-footed animal can be classified into either “dog” category or “wolf” category and people instantly react differently according to their classification. The most basic distinction between two categories occurs when they can be distinguished by the values of a single dimension. A rule for sorting objects, for example, could state that large objects belong to one category and small objects belong to another category. Psychologists have often studied how people learn concepts defined by logical rules requiring two dimensions, such as color and shape. Note that each dimension has three attributes – black, white, and gray for color; and square, triangle, and circle for shape. Logical rules are rules based on logical relations, such as or, and, if-then, to relate stimulus attributes.

_

The following are four fundamental logic rules:

Conjunctive rule uses the logical relation AND to relate stimulus attributes, such as small and square.

Disjunctive rule uses the logical relation OR to relate stimulus attributes, such as small or square.

Conditional rule uses the logical relation IF …, THEN ... to relate stimulus attributes, such as if small, then square.

Biconditional rule uses the logical relation IF and only IF..., THEN … (i.e., if, then in both orders) to relate stimulus attributes, such as (1) if small then square and (2) if square then small.

Suppose that the two relevant dimensions are color and shape and the relevant color attribute is gray and the relevant shape attribute is square. The figure of category assignments shows how four logical rules divide stimuli into two categories: positive examples with a “+” sign and negative examples with a “-”sign.

Rule learning and attribute learning tasks are both empirical ways to study how people learn correct conceptual rules. Rule learning refers to the process of discovering a logical rule from known relevant attributes. Attribute learning on the other hand refers to the process of discovering relevant attributes based on a known logical rule.

The present experiment, is similar to the one developed by Bourne (1970) to test the comparative difficulty of the four logic rules by asking the participants to solve a series of nine successive rule-learning problems.

Method

Participants

The participants of this lab experiment would be your selves.

Materials

The experimental stimuli are visual displays of objects varying on four dimensions – color, shape, number and size. Each dimension has three attributes: blue, green, red for the color dimension, cube, pyramid, and sphere for the shape dimension, one object, two, identical objects, and three identical objects for the number dimension, and small, medium, and large for the size dimension. This gives a total of 81 (3 color X 3 shape X 3 number and X 3 size) object-displays.

Procedure

The experimenter always specifies the two relevant attributes before each rule learning session (e.g., color and shape). Each rule learning session requires the participant learn to classify the stimuli as either positive or negative examples of the to-be-learnt rule (concept). The participant receives one stimulus at a time, make a classification, and will be told whether the response is correct. The presentation of the objects continues until the participant achieves a solution by making 10 correct responses in a row.

Now imagine yourself as a researcher conducting a study on consumers’ preference for different modals of newly marketed sports utility vehicles (SUVs). You are interested in four dimensions that are of key relevance to consumers’ choice preference. Suppose that these SUVs are all in the price range that your target consumers can afford. The four dimensions are color, shape, size and gas-mileage. Each dimension has three attributes. For the color dimension, the three attributes are blue, green, and red; for the shape dimension, the three attributes are “cube” body shape, “pyramid” body shape and “sphere” body shape. For the size dimension, the three attributes are small interior, medium interior and large interior. For the gas-mileage dimension, the three attributes are low, medium and high gas-mileage represented by the number of identical objects (i.e., one, two or three) on display. A single object represents low gas-mileage, two identical objects represent medium gas-mileage, and three identical objects represent high gas-mileage. Accordingly, a red color, cube-shaped, small sized SUV with a medium gas-mileage would be presented by a display of two red colored, small cubes.

Suppose our computer database has the consumers’ preference ranking for all the new SUV models. Your current research aim is to find out for each consumer who participated in the survey study which of the four logic rules described above explains the observed preference pattern the best. For instance, a consumer who uses the Conjunctive rule to determine his preference for SUVs may prefer all the SUVs that are in red color AND cube-shape.

Results

For each group, record and plot the average number of trials used to learn each of the four logic rules.

Discussion

Several hypotheses regarding the comparative difficulty of learning the four logic rules are conceivable:

1. Rote memorization hypothesis. If participants use simple stimulus-response (S-R) association logic, no difference in difficulty among the rules will obtain. This is because in all cases the participant must deal with the same number of stimuli (i.e., 81) and make the same number of S-R associations.

2. Positive instances hypothesis. Alternatively, participants might attend to and derive more information from positive than from negative instances. If so, the fewer distinctly different positive instances in the stimulus displays, the better participants performance should be.

3. Negative instances hypothesis. In contrast to the second hypothesis, participants might attend to and derive more information from negative than from positive instances. If so, the fewer distinctly different negative instances in the stimulus displays, the better participants performance should be

4. Category size hypothesis. Participants might rely primarily on instances of the smaller and less variable of the two categories (i.e., positive instance category and negative instance category). For example, of the two categories of the conjunctive rule, the positive category has a smaller size with only one instance and is smallest category size compared to the smaller categories of the other three rules. The negative category of the conditional rule has the next small and less variable instances that differ only along the shape dimension (see the Figure of category assignments). According to this hypothesis, the fewer instances in the smaller category of a rule, the easier for the participants to learn that rule.

5. Frequency (relevant attributes) hypothesis. Still a fifth possibility that a rule is easier to learn if the frequency of the relevant attributes appearing in the positive category is high. As shown in the figure of category assignments, for the conditional rule, the frequency for the two relevant attributes occurring in the positive category is 29% if the time (the average of 1/7 and 3/7).

Discussion Questions

1. What does this experiment suggest about rule learning?

2. What are the four fundamental logic rules?

3. Define rule learning and attribute learning. How do they differ?

4. Which of the hypotheses proposed in Discussion explains your results the best? Why? Is there a better explanation than these proposed hypotheses to account for your results?

3. Psychologists have posited two distinct mechanisms for forgetting: decay and interference. Describe each of these and briefly review the experimental evidence supporting each, and a possible way to theoretically integrate the both viewpoints.

References

Bourne, L. E. Jr. (1970). Knowing and using concepts. Psychological Review, 77, 546-556.

Bourne, L. E. Jr., Ekstrand, B. R., Lovallo, W. R., Kellogg, R. T., Hiew, C. C., & Yaroush, R. A.(1976). Frequency analysis of attribute identification. Journal of Experimental Psychology: General, 105, 294-312.