William Rybolt

Recall and learning from broadcast news: Is print better?

Recall and learning from television, audio and print news was measured to determine how much is recalled by Kenyan and American college students. For both groups, recall from television and print was similar and significantly higher than from an audio source.

Literacy, Illiteracy, and Learning From Television News

This paper reports on an investigation of the abilities of literates and adult nonreaders to recall and use information from a national network television news program. On a test of unaided recall of news stories the literates recalled 55% more stories than the nonreaders. On a multiple-choice test of information gain from the news program, the literates performed 63% better than the nonreaders. The samples were found to be virtually identical in their use of and opinions about television news. A control group of college students was given the information test without first seeing the news program and did no better than chance. The most important factor affecting recall was the length of time the story was on the air. Human interest stories were recalled much better by both groups than any other type of story. The level of performance among the adult nonreaders correlated highly with their reading levels.

Classifying the uncertainty arithmetic of individuals using competitive learning neural networks

Abstract The application of artificial neural network technology to a host of problems in pattern recognition has long been advocated. Several analyses comparing the performance of neural networks to the standard methods for achieving machine classification and machine learning, such as statistical pattern recognition and ID3, have been reported. Typically, supervised learning has been used and the specific learning algorithm has been back propagation. For many classification type problems, a priori categories are not available, that is, one does not know explicitly the number of categories existent nor the boundaries delineating these categories. Therefore, known targets with which to train the network are not available. A supervised learning approach is not appropriate under these circumstances; an unsupervised learning algorithm is required. In this article we report on the use of an unsupervised competitive learning algorithm as a classifier. The network was used to classify individuals into categories based on differences in the manner in which individuals manipulate the uncertainty associated with the chaining of rules. The experiment, from which the data to be classified were obtained, is described, results of the neural network approach are compared to classification using a distance measure and to classification using a standard clustering algorithm.

A comparison of the manipulation of certainty factors by individuals and expert system shells

The treatment of uncertainty in expert system shells is addressed, starting with a review of the modeling of uncertainty by expert system shells. An experiment to replicate earlier work investigating the manner in which individuals manipulate certainty factors in comparison to commercial shells is discussed. Comparisons are made among seven commercial shells, both personal-computer (PC)-based and mainframe-based, and individuals. A significant difference between individuals and shells themselves is indicated. Some implications for both expert system and decision-support-system methodologies are discussed.<<ETX>>

The application of artificial neural networks to quality control charts

Reports on the development of artificial neural networks that function as alternatives to conventional quality control charts. Multilayered feedforward networks using a backpropagation learning algorithm were trained and tested. The results illustrate the feasibility of using artificial neural networks to detect out-of-tolerance conditions in a manufacturing process.<<ETX>>

The processing of numerical uncertainty associated with the components of if then rules: experiments with human subjects

The authors present the results of an experiment that was conducted to explore how individuals interpret and manipulate uncertainty in basic logical operations. Results indicate that the commonly used models are not as universally appropriate as has been assumed. These results have implications for the process of knowledge acquisition, the design of the system/user interface, and general issues of system development. Collecting information to construct an expert system, duplicating reasoning by using an expert system, and displaying results to decision makers require an understanding of how people actually reason with and use uncertain information.<<ETX>>

Artificial Neural Networks as Alternatives to Statistical Quality Control Charts in Manufacturing Processes

The concept of quality has reentered the vocabulary of American business. The perception, whether founded in reality or not, that American products are inferior to their foreign counterparts, has contributed to the competitive disadvantage now faced by many American firms.

The human machine interface in decision support systems: comparing the functional mappings of human subjects to those generated by feedforward neural networks

Feedforward neural networks have been used to perform classifications and to learn functional mappings. This paper compares human performance to feedforward neural networks using backpropagation in generating functional relationships from limited data. Many business judgments are made in situations where decision makers are required to infer relationships from partial, incomplete and conflicting information. Experiments conducted to compare the interpolations of humans to the behavior of the neural network models are presented. The experiments are described and the results and their implications in designing human-computer interfaces for decision support systems are discussed.