The role of journalism as a filter between events and the way in which they are responded to by elites has been studied by political scientists from Lasswell’s work in the 1920s to contemporary work on framing. However, the majority of these studies have, for the most part, elided the specifics of what journalists are reacting to: the trends, the events, or, as is our focus in this paper, the announcements by state agencies. For the fact is that when states act, they announce those actions in the form of speeches, press releases, news conferences, interviews, and reports, to name some of the more common forms those announcements take. When journalists cover, and, through that coverage, frame, announcements, they do a significant amount of interpretive work. This paper is a report on how we model those interpretations computationally, specifically, how journalists attribute motives to state actors, and how they make claims that a given action is or is not continuous with previous actions. We report on a project that a) hand-codes newspaper articles about central bank actions and foreign policy, then b) uses machine learning techniques on pairs of hand-coded articles and announcements, thereby permitting us c) to generate articles from new or counterfactual announcements.