There is a longstanding debate over how much idiosyncratic, item-specific knowledge is contained in our mental grammars, in addition to productive knowledge of item-general rules and constraints. A key source of evidence is ordering preferences for syntactic alternations like the dative (“throw him an apple” vs. “throw an apple to him”) vary depending on which words they contain. But the quantitative extent of this variability is poorly understood, especially in relation to superficially similar, non-dative constructions which are not alternating (“throw the man to the floor” vs. “*throw the floor the man”). To address this, we built a large corpus of naturally-occurring sentences including either dative or superficially similar non-dative structures, and analyzed the unique contributions of productive and verb-specific knowledge in predicting argument ordering preferences