Producing language is a kind of action – if the circumstances are right, certain effects may ensue, and specific physical, cognitive or social goals can sometimes be achieved by following appropriate courses of verbal action. The problem of synthesizing an organized collection of actions that leads to goal achievement can often be solved with automated planning methods. It is thus natural that such methods have found application to the automatic production of understandable text in natural language, i.e., to natural language generation. In this article, we survey a number of earlier and ongoing computational approaches to natural language that generate utterances by modeling speech acts or words as particular types of actions in planning a problem. After discussing strengths and weaknesses of the different models, we outline some possible directions for future work that could further advance this field.