Studies of speech production and its control have traditionally focused on acoustic parameters and/or articulatory configurations of the end result of the production process as the target of speech production. The representations of production target, however, are far richer than the stereotypical acoustic/articulatory features that have been identified from those traditional studies. Speech production is a process through which mental representations of a language are transformed into vocal tract movements and sounds as physical entities. Thus, the speech production target includes these complete representations, and in order to fully understand how the target is controlled, we need to perturb the production process. This paradigm allows us to see how the production system as a whole corrects behavior in response to the perturbation. By systematically introducing perturbation, it is possible to examine what is needed for the system to detect an error, and how such an error is reduced. This, in turn, enables us to have a better understanding of what the speech production target is and how it is defined. A series of experiments were carried out to examine this issue in this thesis, using a real-time auditory perturbation paradigm. This paradigm perturbs the auditory feedback speakers receive while they are producing a speech segment. In response to the perturbation, subjects spontaneously change their articulation to compensate. The results showed that 1) the speech production target is not a list of independently controlled acoustic features but is a multi-dimensionally defined category that is language dependent 2) spectral and temporal aspects of speech motor control show the same results 3) similar compensation behavior is observed even with using an unfamiliar tool to produce a vowel-like sounds, and 4) an intention to produce a speech category may be manifested in ways that are different than behavior in other motor control studies such as reaching.