We report the development of an interactive approach to single-word language production studies in fMRI. The approach, adaptive pacing, involves real-time adjustment of stimulus presentation times based on individual subject performance timing and content. At the same time, it maintains a stochastic distribution of interstimulus intervals to avoid confounding task covariates with speech-related signal variance. Adaptive pacing of overt speech production is an example of a new class of paradigms that require an observational approach to data acquisition and benefit from a "time-aware" acquisition and processing environment. The advantages of adaptive pacing in fMRI of impaired subjects are expected to be the acquisition of more informative data per unit time, less contamination of data by correlates of non-language processes such as emotion, and facilitation of experiments that combine normal and impaired subjects.