We hypothesize that canonical sentence schemas (e.g. Agent—verb-Patient) can sometimes assign argument structure to verbs. In particular, they provide a default argument structure early in learning when a verb's lexical entry may record the nature of the action but lack a specific argument structure. To test the theory and its application to causative verb errors (e.g. stay it there), novel action verbs were modelled, some as causative, some as intransitive, and some unmarked for transitivity. Spontaneous usage was recorded, along with responses to agent-questions (‘What is the [Agent] doing?’) and patient-questions (‘What is the [Patient] doing?’). Comparable data were obtained for familiar English verbs, some of fixed and some of optional transitivity. Subjects were willing to use all novel verbs both transitively and intransitively, although adults respected assigned transitivity more than children. All subjects largely respected the transitivity of familiar verbs. The discourse pressure of the agent- and patient-questions greatly affected observed transitivity. No evidence was found for the intransitive-to-causative derivational process postulated by Bowerman. We propose that the kind of causativity error observed by Bowerman is due to assignment of argument structure from canonical sentence schemas, especially under pressure of a need to make a sentence with a particular argument (Agent or Patient) as subject. The theory has the advantage of explaining errors without postulating the acquisition of erroneous lexical entries that have to be unlearned, and it can be extended to other kinds of errors in the choice and placement of arguments.