Reflections on Phonetic Convergence: Speech Perception does not Mirror Speech Production

Abstract

Since the beginnings of speech research, the issue of variability has been a central topic in the field. Until recently, solutions to this problem have focused mainly on resolving linguistic forms from variable acoustic realizations, with little systematic consideration of talker variability or social settings. However, acoustic-phonetic variability within and between talkers does not arise by chance -such variation is attributable to a variety of causes including talker physiology, dialect, affect, and social ⁄ situational factors. Studies of conversational interaction demonstrate the influence of situational factors on many aspects of spoken communication. In particular, interacting talkers converge and diverge in acoustic-phonetic form over the course of a single conversational setting. Although phonetic convergence likely reflects a close connection between speech perception and production, it is not an automatic consequence of priming mechanisms or a mirror neuron system. In studies of the neural activity associated with speech perception, it is increasingly clear that utterances are processed in separate dorsal and ventral cortical streams, one that is part of the selfregulation of spoken utterances, and another that is part of conceptual understanding. Links between perception and production appear to be part of the self-regulatory system that participates in speech production control monitoring, which makes this system a likely candidate for talker control of acoustic-phonetic variability. Whether talkers converge or diverge depends on multiple aspects of the social setting, which have a much greater influence on the form of phonetic expression than automatic links or mirror mechanisms.Spoken communication works so well that its efficacy is taken for granted. Two complete strangers who share a language can converse with relative ease from the moment of introduction. However, this feat is accomplished under conditions of enormous variability in the sounds that are produced by talkers. Decades of research have not yet yielded a complete explanation of how individuals accommodate this variability when perceiving speech. As a consequence, recent accomplishments in speech recognition technology fall short of producing systems that are robust across multiple talkers and unlimited lexical ⁄ topical sets. Many of the old debates persist in the research literature, in particular, the dispute over the nature of the relation between speech perception and production.Typically, this dispute has been cast in stark terms: Listeners rely primarily either on auditory representations or on motor representations when perceiving speech. Currently, there is ample evidence in favor of both propositions, and current debaters may wield Occam's razor and cite a multitude of studies to support either side (Kluender and Kiefte 2006;Liberman 1996; Pardo and Remez 2006;Trout 2001). With the increased interest in mirror neuron systems, the dispute has moved beyond psycholinguistics into neuroscience (Rizzolatti and Craighero 2004). Wi...

Prediction and imitation in speech

“…Abstracting from the reliability and directionality of single findings, data are clearly consistent with two assertions: (i) that speech perception influences speech production; (ii) that the link between speech perception and speech production is mediated by a number of variables (Pardo, 2006, 2012; Pardo et al, 2010). The first assertion (i) is incorporated in three of the four theories we have presented: the ET, the MT, and the ST.…”

“…It is possible that imitation may sometimes serve as an intentional signal with an intrinsic communicative value (e.g., “I want you to like me”; cf. Pardo, 2012), but we propose that it generally occurs as a by-product of the internal mechanics of speech comprehension.…”

“…It has been suggested that the overall impression of similarity reported by naïve listeners in AXB tasks could be realized phonetically in different ways, possibly with relatively minor adjustments along several dimensions (Pardo, 2012). This could explain why some studies have failed to confirm results from perceptual judgment tasks when they looked at a few acoustic dimensions that were hand-picked by researchers (Pardo et al, 2010, 2012).…”

Native voice, self-concept and the moral case for personalized voice technology

“…[50] Voice synthesis technologies are improving but we remain some way from cost-effective, widely accessible personalized voices for use in SGDs. The paucity of rigorous studies on the effects of loss of native voice on self-concept leaves a large number of individuals suffering, literally, in silence.…”

The extent of engagement, the means of invention: measuring debate about mirror neurons in the humanities and social sciences

“…For example, looking across the texts starting from 2007, six of the thirteen articles in that time period cite a handful of early MN sources exclusively, effectively establishing the MN finding and then working from their own field-specific sources [See: Aboitiz, Aboitiz and García, 2010;Connolly, 2011;Griggers, 2009;Jajdelska et al, 2010;Pardo, 2012;Powell, 2007]. 8 The remaining articles cite sources up through 2011 that continue to support the Action Understanding Theory [including MN sources such as: Arbib, 2005;Dapretto et al, 2006;Corballis, 2005;Fogassi, 2011].…”