This dissertation examines the evidence of linguistic units in speech production on the basis of speech errors. Traditionally the qualitative and quantitative patterning of speech errors has been taken to mean that the cognitive processes involved in word form encoding manipulate a-temporal symbolic units, which get implemented 'normally' at later stages in the production process. The categorical properties of speech errors (identified on the basis of auditory evaluation) have thus been taken to corroborate the psychological reality of a-temporal segmental representations. The data presented in this dissertation run contrary to these assumptions: On the basis of tongue movement data the present work shows that many errors are not categorical segmental substitutions; rather, speakers spontaneously add extra articulatory gestures that are communicatively inert and grammatically illegal. These temporally mislocated gestures occur along a continuum of values. A tongue twister-like and a non-repetitive error elicitation method yielded convergent results. The perceptual consequences of these illicit articulatory constellations vary greatly: A perception study included in this dissertation shows that errorful tokens can indeed be perceived as error-free and well-formed.
The grammatically incorrect coproduction of two gestures (one appropriate, one erroneous) is conceptualized as arising from a weakening of lexicalized, complex coordination modes which usually are stable attractors in speech production. In errors, an attractor of fundamental dynamic stability comes to temporarily dominate over the phonologically stable patterns. Coordinating multiple events in 1:1 frequency coordination has been shown to be the most stable pattern in movement synchronization tasks: Under certain conditions, such as increased rate, transitions from more complex patterns to in-phase 1:1 synchrony can be triggered. While mainly finger-tapping and limb oscillation experiments have been used to investigate this phenomenon, the speech error data can be interpreted as the emergence of a 1:1 frequency, in-phase pattern in speech production and can thus be understood as emerging from fundamentally dynamic characteristics of the speech production process.