Abstract: This paper analyzes the grammatical nature of the preceding verb in compound verbs, so as to clarify the de-causal conditions and the mechanism of causative alternations in Japanese compound verbs. The results show that when compound verbs are de-causative, it is necessary to realize not only the background of the preceding event, but also its semantic restriction on the subsequent event, so as to ensure that the preceding verb participates in both the conceptual and argument structures. Moreover, the compound verb derived from de-causative actions, whose preceding verb represents the “result like state” of the changed object, is designated as the subordinate of BE in the semantic conceptual structure, and further defines the “result state” shown by the following verb. This conclusion helps to explain the applicability of the “Unique Path Constraint” to Japanese compound verbs and the hypothesis of “Result Regression”.