During election campaigns party elites often communicate which coalition they might or might not be willing to enter after the election. These public statements about prospects of future governments are often picked up by media outlets. In this paper, we present a supervised learning model to classify coalition signals in newspaper articles. The method identifies segments of text in which an affiliate of one party rules out or promotes a coalition with another party. We evaluate our approach based on hand-coded newspaper articles in Germany and Austria and validate the method based on pre-electoral coalitions in state elections. Both applications reveal the advantages of our method over existing dictionary approaches. Research can use our approach when interested in identifying targeted party elite communication about a specific subject in newspaper articles.