Humans, Machines, and Conversations

Abstract

This essay investigates the design of automatic speech recognition (ASR) technologies as a site at which the human qualities of ‘hearing’ and ‘understanding’ are mimicked in machines. On the basis of an observational study, it explains the major work components and debates in ASR, such as research paradigms and strategies, as well as the intricacies of instruments and experimentation. Key arguments concern the tensions, conflicts, and ambiguities that emerge from the intersection of disparate research paradigms in ASR. The study emphasizes the actors’ success in establishing a local context of stable, practical rationality in which they negotiate the complementary and conflicting research objectives of building speech recognition machines, on one hand, and understanding human hearing, on the other. The conclusion is cast in terms of a ‘functional contingency’, a concept that characterizes this setting in which researchers successfully make machines ‘recognize speech’ and in so doing make the machines part of specific types of ‘conversation’.