The goal of this study is to account for the observed surprising discrepancy between the relatively early production of Hebrew Object Relative clauses as opposed to their late comprehension (based on the findings in Günzberg, Shvimer, & Friedmann (in press)). The main hypothesis advanced in the paper is that at a certain developmental stage children treat relative CPs as simple modifiers, namely as constituents with a slot, resulting from externalization of one of the verb's θ-roles. Based on this, and assuming the processing model of Pritchett (1992), the chance level comprehension of Object relatives, as opposed to good comprehension of Subject relatives, is argued to derive from the externalization mechanism assumed in the parsing of relative clauses and its interaction with the processing guideline, leading to two equal parsing analyses randomly chosen by the automatic processor at a certain processing stage. The variety of the well-comprehended Relative Clauses (e.g. Free Object Relatives, relatives including a null arbitrary subject) is shown to lack such stage, providing sufficient and unambiguous linguistic information for the automatic processor.