PhD Thesis: A Power and Reliability Model for Error Prone Technologies: Improving Speech-Based Support for Spatial Navigation

UMBC
Maryland, US
May 2005

Navigation is one of the major activities that people engage in when interacting with computers. When using traditional navigation solutions, such as the mouse, inherent limitations of human capabilities are the primary source of errors. However, the development of assistive technologies and ubiquitous computing have highlighted the importance of system- and context-based errors both of which differ in important ways from user-based errors. Traditional theories such as the speed and accuracy tradeoff and Fitts' Law may effectively explain user-based errors, but are insufficient when errors are caused by limitations of the underlying technologies or the context in which the interactions are occurring.

To address this issue, a power and reliability model (PRM) was proposed, which expands on traditional theories to more effectively address all three types of errors. This model consists of two components: power and reliability. Power addresses the efficiency of a technique or application while reliability addresses the errors that occur during a task. The tradeoff between power and reliability, which is inherent when employing error-prone technologies such as speech recognition, has significant implications for system and interface design. To operationalize the concept of reliability, a metric is proposed that incorporates both the likelihood of command failures and the severity of the outcomes caused by those failures. This metric builds on the concept of entropy in information theory to provide a systematic solution that allows the reliability of various techniques and applications to be quantified, evaluated, and compared.

Several studies were conducted to begin the validation of the PRM. One study involving ninety participants focused on the validation of the reliability metric. The results indicate that, compared to failure rates, reliability scores are more effective for describing both objective task completion times and subjective user satisfaction ratings. A case study involving 29 participants illustrated the tradeoff between power and reliability and demonstrated how the proposed model could be used to guide the design of more effective solutions. Interaction solutions developed based on design guidelines derived from the PRM resulted in lower command failure rates, less severe failure consequences, the development of more effective strategies, and more efficient interactions.