This talk investigates phenomena involving long object promotion [LOP]—cases in which an embedded argument is promoted to matrix subject. We provide a comparison of the distribution of different types of LOP, specifically raising (of an embedded passive object), long passive, and crossed control, and suggest a unified account based on several syntactic and semantic similarities. The account employs two mechanisms: i) syntactic Voice restructuring which creates a local feature dependency between two Voice heads in certain control contexts; and building on that dependency, ii) (unless the matrix verb is non-thematic) semantic argument sharing via semantic binding which enforces coreference between the matrix and embedded Agents (whether overt or understood). As a broader implication, the Voice restructuring model also extends to forward and backward exhaustive control, and thus offers an alternative to both, syntactic control models involving movement or PRO, and pure semantic control models.