Stephen K. Reed

Pattern recognition and categorization

Four experiments are reported which attempt to determine how people make classifications when categories are defined by sets of exemplars and not by logical rules. College students classified schematic faces into one of two categories each composed of five faces. One probability model and three distance models were tested. The predominant strategy, as revealed by successful models, was to abstract a prototype representing each category and to compare the distance of novel patterns to each prototype, emphasizing those features which best discriminated the two categories.

The role of analogy in transfer between similar problem states

Abstract The study investigated the effect of transfer between two problems having similar (homomorphic) problem states. The results of three experiments revealed that although transfer occurred between repetition of the same problems, transfer occurred between the Jealous Husbands problem and the Missionary—Cannibal problem only when (a) Ss were told the relationship between the two problems and (b) the Jealous Husbands problem was given first. The results are related to the formal structure of the problem space and to alternative explanations of the use of analogy in problem solving. These include memory for individual moves, memory for general strategies, and practice in applying operators.

Structural descriptions and the limitations of visual images

Two experiments are reported that require Ss to judge whether or not the second of two sequentially presented patterns is a part of the first pattern. The results suggest that Ss code the pattern as a structural description and find it difficult to recognize a part of the pattern which does not match the description. It is proposed that a structural description is a combination of visual and verbal codes and that visual images lack detail when not supported by verbal concepts.

Modeling Strategy Shifts in a Problem-Solving Task.

Abstract A computer simulation model was fitted to human laboratory data for the Missionaries and Cannibals task to explain (1) the effects upon problem performance of giving a hint, and (2) the effects of solving the problem a second time after one successful solution had been achieved. Most of the variance in the relative frequencies of different moves can be explained by positing that the effect of the hint, or of previous experience in solving the problem, is to cause subjects to switch more promptly from a strategy of balancing the numbers of missionaries and cannibals on both sides of the river, to a means-ends strategy.

Detection of parts in patterns and images

Subjects performed in an embedded-figures detection task which required them to judge whether one pattern was a part of another. In the perception condition, the part was presented before the complete pattern, but in the imagery condition, the part was presented after the complete pattern. Subjects made fast, but inaccurate, responses in the perception task when RT s were recorded, but the error rate declined substantially when they were given 10 sec to make a decision. In the latter condition, subjects failed to detect a part on 14% of the trials in the perception condition and on 72% of the trials in the imagery condition when a correction was made for prior perception of the part. A subsequent experiment showed that the high error rate in the imagery task was not the result of the subject’s inability to remember the complete pattern. The complexity of mental operations and the limited accuracy of visual images are considered as possible alternative explanations of the results.

Category vs. item learning: Implications for categorization models

A paradigm that required that subjects learn two responses to each of 10 schematic faces was used to study the relative rate of discrimination and generalization learning. One response uniquely identified each face, whereas the second response classified each face as a member of one of two categories. Rapid category learning and slow item learning suggested that category responses were learned on the basis of abstracted information, but item responses depended on the more difficult task of discriminating among patterns. The results are related to categorization models and to task variables that should influence the relative rates of discrimination and generalization learning.

Perceptual vs conceptual categorization

The study considered whether Ss use the same strategies in categorizing biographical descriptions as in perceptual classifications. A biographical description consisted of a person’s age, income, number of children, and years of education. The Ss were asked to classify these descriptions as residents of two different suburbs in order to compare the results with a previous study using schematic faces. The 123 Ss were assigned to one of three alternative organizations of the descriptions: a table, a “name” organized paragraph in which each person was successively described, or an “attribute” organized paragraph in which each attribute was successively described. Essentially the same results were obtained for schematic faces and biographical descriptions, except for a few differences which were attributed to the use of more realistic categories (suburbs) in the present study. When more realistic categories are used, a S’s prior experience can influence which attributes he emphasizes and possibly his formation of an abstract prototype.

Memory for Problem Solutions

Publisher Summary This chapter examines traditional research on paired-associate learning and serial learning sufficient for understanding problem solving or new concepts to theorize about what a person remembers after solving a problem. The three experiments presented in the chapter makes use of a variety of paradigms in an attempt to determine what is remembered after solving a problem. The problem in all three experiments is a version of the missionary–cannibal (MC) problem in which there are five missionaries (5M), five cannibals (5C), and a boat which can hold three people. The task requires the transportation of all people across a river under the constraint that the missionaries are never outnumbered by cannibals on any side of the river or in the boat. The issue of intentional versus incidental learning provided an independent variable for all three experiments. The emphasis on strategies and subgoals raised the question of whether subgoals are differentially effective in producing good strategies and whether subgoals result in more forward planning or merely change the evaluation function used to select the next move. The purpose of the recognition memory task is to provide data on recognition memory and study the effect of an intervening task on memory for their first solution.

Structural models and embedded-figure difficulty for normal and retarded children

60 normal and 20 mentally retarded children were tested on an embedded-figures task to determine what stimulus variables correlate with the difficulty of finding a figure hidden in a larger pattern. The number of shared contours, the complexity of the ground, and analysis complexity (defined as the amount by which the total complexity of the parts exceeds the complexity of the whole pattern) resulted in significant, positive correlations with the difficulty of finding an embedded figure. The complexity of the figure, the complexity of the complete pattern, and the number of overlapping lines resulted in nonsignificant, positive correlations with the difficulty in finding the figure. The task was discussed within the theoretical framework of the structural model advocated by Narasimhan and Sutherland. The results suggested how a pattern might be analyzed into a hiearchy of articulations with each articulation consisting of a hierarchy of component parts.

Temporal organization of pattern structure

Two pattern reproduction experiments examined the relations among the figural goodness of a pattern, the organization of two parts within the pattern, and the interpart interval (ISI), which ranged from 40 to 200 msec. If the parts contained connected line segments, performance was slightly better (3%-5% gain in accuracy) at a 40-msec ISI than at a 200-msec ISI. If the parts contained unconnected line segments, reproduction accuracy of the first part declined sharply between 40 and 200 msec. These results were interpreted by assuming that the parts were perceived as a single whole pattern at a 40-msec ISI but as two separate patterns at a 200-msec ISI. One surprising finding, the lack of an interaction between figural goodness and ISI, was explained in terms of a response bias in favor of figural good patterns. A secondary manipulation revealed that a part was more accurately reproduced in a good figure context than in a poor figure context but was most accurately reproduced when it appeared alone.