Abstract: How do elites understand foreign and monetary policy? In the past, scholars have explored this question by modeling it as a psychological mapping process: elites focus on particular features of policies and map them onto particular psychologically relevant categories, such as “adversary” or “danger,” each of which usually has one or more action implications. The problem with understanding policy interpretation in this way is that it requires specifying multi-level rules by which particular words or phrases, in order to be mapped onto categories, are first mapped onto context, with the contexts themselves being mapped onto higher-level contexts, and so on ad infinitum. As an alternative, we reconceive of policy interpretation as a matter of nonhierarchical textual entailment, in which different combinations of syntactic and semantic information will be associated with stylized accounts of what is, or might, or should not be happening and why. Specifically, we use machine learning techniques to generate newspaper accounts (actual or possible actions, attributed motives, and so forth) of two lengthy streams of policy announcements: one, covering 24 years, from the White House and State Department about the Soviet Union/Russia; and the other, covering 51 years, from the Federal Reserve, about monetary policy.