Firstly, we investigated the frequency of occurrence of single vs. double object sentences in German corpora (Negra and Tiger). An overall strong prevalence of single object structures was found, irrespective of case. When sentences with animate referents only were looked at, this prevalence persisted for sentences beginning with a nominative and an accusative, whereas for those beginning with a nominative and a dative, more double than single object sentences occurred. Loglinear analyses confirmed the predictive power of case and animacy.
Secondly, a completion questionnaire was administered that included sentence fragments of the type "X will Y ...". X was always marked for nominative, while case of Y and animacy of X and Y were varied. When the first constituent referred to an animate entity, there was a case effect mirroring that found in the corpus study: Most completions after an accusative consisted of a verb only, whereas most completions after a dative consisted of a third constituent and a verb. In contrast, subjects tended to insert other completions (resulting neither in a single nor in a double object construction) when the first constituent referred to an inanimate entity.
We propose that syntactic projections in verb-final sentences are based on prototypical thematic roles (agent, theme, and recipient, cf. Primus, 1999) and the actions they are typically involved in. Processing data from German indicate that the processing system assigns thematic roles on-line (e.g., Knoeferle et al., 2005) and that it engages in the construction of thematic hierarchies before the verb (e.g., Bornkessel, Schlesewsky, & Friederici, 2003). However, a corpus study including inanimate referents resulted in many empty cells in the double object conditions. This finding supports the thematic projeciton hypothesis, but it also indicates that the impact of case and animacy on sentence projection is hard to disentangle in an online task because of the lack of verbs in some conditions.