We provide a summary of our recent research on the control of rapid action sequences in speech production, emphasizing findings about the advance planning and hierarchical organization of such sequences. The effects of number of elements in the utterance (its "length") and other factors on maximum production rates of short utterances lead us to infer that a "motor program" for the whole utterance, prepared in advance, controls the execution of each of its "units". Findings from studies of typewriting as well as speech production have led us to a model in which the performance of each unit is controlled by two processes arranged in sequence: one (subprogram selection) whose duration increases linearly with sequence length, and the other (command) whose duration depends on type of unit. Quantitative aspects of the production of utterances composed of different types of element suggest that the action unit in speech is the stress group or metrical foot. The virtual identity of the timing of word and nonword utterances implies that the utterance program is sufficiently detailed so it can be executed without reference to learned routines for words stored elsewhere in memory.

We review our search for properties of performance that are suggested by the model: First, the time from a reaction signal to the first unit (the latency) increases linearly with utterance length. Second, the maximum length utterance controlled by one program depends on unit size. Third, the effect of utterance length on production timing is localized (intermittent), rather than affecting all parts of the articulatory stream. And fourth, the effect of utterance length on production timing appears in just one epoch per unit.