Puzzle-solving science: The quixotic quest for units in speech perception

Although speech signals are continuous and variable, listeners experience segmentation and linguistic structure in perception. For years, researchers have tried to identify the basic building-block of speech perception. In that time, experimental methods have evolved, constraints on stimulus materials have evolved, sources of variance have been identified, and computational models have been advanced. As a result, the slate of candidate units has increased, each with its own empirical support. In this article, we endorse Grossberg's adaptive resonance theory (ART), proposing that speech units are emergent properties of perceptual dynamics. By this view, units only "exist" when disparate features achieve resonance, a level of perceptual coherence that allows conscious encoding. We outline basic principles of ART, then summarize five experiments. Three experiments assessed the power of social influence to affect phonemesyllable competitions. Two other experiments assessed repetition effects in monitoring data. Together the data suggest that "primary" speech units are strongly and symmetrically affected by bottom-up and top-down knowledge sources.