Many social processes are adaptive in the sense that the process changes as a result of previous outcomes. Data on such processes may come in the form of categorical time series. First, the authors propose a class of Markov Source models that embody such adaptation. Second, the authors discuss new methods to evaluate the fit of such models. Third, the authors apply these models and methods to data on a social process that is a preeminent example of an adaptive process: (encoded) conversation as arises in structured interviews. © 2007 Sage Publications.