Toward an (even) more comprehensive model of speech production planning

ABSTRACT

Since the publication of Speaking in 1989, with its extraordinary goal of modelling the entire process of human speech generation from message conceptualisation to articulation, encompassing results from a wide range of empirical studies, much new information has emerged about three aspects of speech production that were not clearly in focus at that time. This evidence has revealed 1) the systematic patterns of context-governed surface phonetic variation, and the active control of these patterns exercised by speakers and listeners, 2) the depth and pervasiveness of prosodic influences on those patterns, and 3) the close alignment of co-speech gestures with the prosodic structure of an utterance. This paper reviews some of that evidence, and suggests how its implications may constrain models of speech production planning, as those models become more comprehensive in their treatment of higher-level structures, and of aspects of the communicative act beyond the articulation of lexical and syntactic elements.

KEYWORDS:

• Speech production planning
• feature-cue-based processing
• prosodic control of phonetics
• co-speech gesture
• prosody-gesture timing

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 Examples cited here are drawn from the MIT Speech Error Corpus, a collection of 11,000+ errors collected by the author and her colleagues, by listening to everyday speech over the past decades; individual errors have been largely labelled for error type, error unit, direction of error and error ambiguity (which is pervasive, Shattuck-Hufnagel, 1987).

2 Levelt et al. (1999) describe a mechanism by which such cross-PWd errors can occur, postulating for example that in the encoding of red socks, the phonemic segments for both syllables are available during the process of Phonetic Encoding, by which syllable-position-specific elements like syllable onsets are associated with their appropriate slots. the result that the onset /s/ can be mis-selected for the first onset position and the onset /r/ for the second onset position, and if their binding-by-checking mechanism (Roelofs, 1997) fails to detect the mis-associations, the error sed rocks will be produced. This account opens the door to the possibility that longer stretches of the phrase or utterance are planned phonologically.