English resultative constructions are widely believed to obey the generalization that the predication subject for the resultative phrase must be the verb's (underlying) object (the Direct Object Restriction; DOR). This talk will show that the DOR is factually incorrect and will explain the distribution of resultatives in terms of the semantic selectional restrictions imposed in control relations. We distinguish between control resultatives ('John hammered the metal flat') and non-control resultatives ('The dog barked itself hoarse'). In a control resultative, the predication subject of the resultative phrase is a semantic argument of the matrix verb; in a non-control resultative it is not. Agentive verbs of motion like run, walk, and jump are problematical for the DOR, as noted by Levin and Rappaport Hovav (1995). It will be shown that these motion verbs require that any added resultative must express a locative relation, when there is a control relation. But in a non-control resultative, the selectional restrictions are relaxed. In short, resultatives can be predicated of unaccusative or unergative arguments, as long as the semantic conditions on control (if there is control) are observed. Intuitively the control resultative must express a 'normal' type of result state for the verb. A strongly atelic predicate like 'bark' is not associated with hoarseness or any other property as its normal result, so '*The dog barked hoarse' is bad; while 'The dog barked itself hoarse' does not involve control, so it is acceptable.
