Research over the last decades has shown that language development in its multiple forms is characterized by a succession of stable and unstable states. However, the variation observed is neither expected nor can it be accounted for on the basis of traditional learning concepts conceived of within the Universal Grammar (UG) paradigm. In this paper, I argue that modularly organized grammars bear much more of a dynamic potential than admitted thus far, and I propose a dynamic approach to the development of grammars, based on a conception of change as developed in the realm of Dynamic Systems Theory (DST). In my discussion of the available evidence of system-internal inconsistencies in different types of language acquisition and diachronic language change, I suggest that the nonlinear behavior observed results from a complex information flow modeled by internal and external feedback processes and that changes in grammars are tied to the amplification of new information leading to system-internal conflicts. Finally, I reconsider the role of UG in the apparent dichotomy of chance and necessity in the evolution of grammars. I argue that their stability is tied to universal principles and constraints on the format of natural languages (hence the self-similar or fractal nature of language), whereas the potential for change is given in the functional categories and their associated properties (the loci of grammars' bifurcation sensitivity).