In this paper we argue for the following properties of clause-bound scrambling, as they are manifested in German. First, scrambling presupposes head-final projections. Only selected constituents, notably arguments, scramble, the reason being that phrases selected by a head have a unique base order. Second, scrambling involves antecedent-gap dependencies with A-chain properties. Third, scrambling is overt and non-string-vacuous. Fourth, scrambling is syntactically optional, clause-bound, category neutral, and may apply to more than one phrase per clause. Fifth, scrambled elements remain transparent for extraction; they are licit binders and take scope.Furthermore, we evaluate our conclusion that scrambling is contingent on the "OV" property by examining Yiddish, an uncontroversial scrambling Germanic language with controversial VO versus OV status. We argue that Yiddish is a variant of an OV language-thus allowing scrambling-and that it is the only Germanic language with alternative V-positions in a VP-shell structure, like Hindi, and, arguably, like Slavic languages. * 1 Chomsky (1995:324) suggests that operations such as extraposition and scrambling "may not really belong to the system we are discussing here as we keep closely to ... movement driven by feaure checking within the N-£ computation," where N is a numeration and £ is LF.206 Haider and Rosengren scrambling operates. In section 4 we derive the fact that scrambling applies to selected items only, notably arguments. Section 5 recapitulates the coverage of the facts under our analysis. In section 6 we argue that Yiddish is basically OV and not VO, whence scrambling is expected. Section 7 summarizes the paper.
Theoretical Background.The main concern of this paper is an empirically adequate modeling of scrambling, paying attention to a nontechnical key issue of the minimalist program in its later versions (Chomsky 1998(Chomsky , 1999, namely optionality. Chomsky's (1998Chomsky's ( , 1999 ideas differ from those of Chomsky 1995 with respect to "bare output conditions": Covert movement is avoided and replaced by a matching operation that erases uninterpretable features of the target by matching them with features of the probe. As expected, these features are case and agreement features. Movement (or chain formation) is thus overt and feature-triggered. We argue that accounts in terms of feature-driven movement are inadequate for scrambling.In our view, overt operations are allowed when grammar does not forbid them and are licensed either by structural requirements, such as one that requires the movement of the finite verb in root V2-clauses, or by the possibility to exploit them in a systematic way at the syntaxexternal interfaces. This we assume to hold for the EPP as well as for scrambling. 3 Our starting point is a traditional CP-IP-VP structure for the Germanic VO languages. We do not see the need or justification for any inherently Agr-type projections. We assume only projections that may match the inflectional features of the verb, among others, TP. ...
