Certain verbs (e.g. bend, break) can undergo the causative alternation, appearing both in a transitive form (John bent the lever) and an intransitive form (The lever bent). To explain this alternation, linguists have proposed that the transitive form of the alternation conveys an action with an explicitly causative structure— for example that John bent the lever indirectly conveys the proposition John caused [the lever bent]. While this analysis is syntactically appealing, it makes strong assumptions about explicit representations of causality in the motor system that must be independently justified. In fact there is a large body of research within motor neuroscience showing that motor programmes can be defined by the perceptual effects they cause. In this paper we argue that the causative action representations posited by linguists are direct references to the effect-based action representations identified by motor neuroscientists. We develop a neural network model of how an agent can learn actions that are defined by the effects they bring about, and deploy this model in an account of the syntax of causative sentences.