A topic that has been of perennial interest in political science is that of government actors' political power. However, measurement of political power has thus far been a highly costly process involving hand-coding by experts; consequently, only the executive power of a limited number of polities has been examined in depth. This paper proposes a novel methodology to extract measures of political power directly from the text of US state constitutions using machine learning, text as data, and network methods. With these measures, I find support for conventional wisdom surrounding the changes in state executives' relative power over time and validate my the measure both through comparison of predicted power of government actors with the power estimated using hand-coded data and qualitative examination of constitutional texts. This project contributes to the development of a systematic, objective, and cost-efficient method of measuring de jure political power using constitutional texts.