<<   >>   TOC

4.2 Chain Formation

As mentioned above, Haider (1993) proposes that scrambling structures are created by base generating a trace to link the scrambled argument to its (empty) default position. This proposal has the advantage that the argument surfaces in the scrambling position not through application of Move-a, but by cost-free merging. Whereas Haider emphasizes that his model is designed to be purely representational, it is straightforwardly translatable into the terminology of the strictly derivational minimalist program, which I will attempt here:

It is commonly assumed that the features associated with lexical items are organized in clusters, i.e. a set of formal features (F-features) and a set of morphophonological features (P-features). One may argue that in English, the P-features of an object are saturated within the VP, whereas the F-features enter the syntax via an AgrO-position. Since English P-features supply only minimal information about associated F-features, a convergence condition may then require that at the point of Spell-Out, both P- and F-features are to be found in the same position, which would trigger movement of the overt object up to [Spec, AgrO]. Haider's proposal can now be reformulated as follows: German has overt case morphology, i.e. German P-features do inform about associated F-features. Thus, German allows saturation of F-features in the default argument position, whereas saturation of the P-features may or may not be postponed.[1] Because of the 'traceability' of the P/F-feature association, convergence does not depend on feature unification before Spell-Out as it does in English.

This, at first glance, extremely attractive proposal fails, because it shares one characteristic with the traditional A-/A-bar movement accounts: it crucially relies on the existence of a trace (i.e. the saturation position of the F-features). A close scrutiny of the relevant data reveals, however, that the hypothesis of a scrambling trace is untenable.

Haider (1993: 199) mentions a circumstance that makes scrambling traces suspicious from the start: multiple scrambling does not show relativized minimality effects in terms of Rizzi (1991). Consider (24):


Die Prinzessin should be a potential antecedent for ei and thus intervene between ei and der Kuß, but no such intervention is discernible. The theory would have to be embellished with ad hoc assumptions granting scrambling traces a special status, certainly an unwelcome complication.

In the context of V2, various elements can undergo fronting in German. To recapitulate, (16b) - (16f) show that subject, object and verb can be fronted alone. Verb fronting is possible regardless of the argument order in the residual VP. Object and verb can undergo V2 together, but not subject and verb.


            a.         ..., daß die Prinzessin   den Prinzen   geheiratet hat.

                             that the princessNOM the princeACC married    has

            b.         Die Prinzessin hat den Prinzen geheiratet.

            c.         Den Prinzen hat die Prinzessin geheiratet.

            d.         Geheiratet hat die Prinzessin den Prinzen.

            e.         Geheiratet hat den Prinzen die Prinzessin.

            f.          Den Prinzen geheiratet hat die Prinzessin.                      

            g.         * [ VP Die Prinzessin geheiratet [V ei ] ]j hati den Prinzen ej.

Haider claims that the ungrammaticality of (16g) stems from the fact that the moved VP contains the trace of the ACC-argument, i.e. the derivation of (16g) would involve an intermediate scrambling step, such that the resulting structure should be that shown in (17):

(17)   *[VP Die Prinzessin [VP ej [VP [V geheiratet [V ek ]]]]]i  hatk [VP den Prinzenj ei ]. 

The fronted scrambling trace ej would not be c-commanded by its antecedent and result in an ECP-violation.

One problem with this analysis is the fact that the fronted VP already contains a trace, i.e. that of the auxiliary hat: since no independent I-position is available, auxiliaries in base structures such as (16a) are assumed to merge with V, yielding the representation (18):


In the formation of a matrix structure, the auxiliary would have to move up to F, leaving a trace. Whereas (16d) and (16e) would allow for the verb to move without the auxiliary trace, (16f) requires it to be tagged along, since den Prinzen and the verb would otherwise not form a constituent. The auxiliary trace obviously does not violate the ECP, which requires explanation. It is often assumed that typical A-bar movements (which include V2-fronting) either allow LF-reconstruction of the moved element in its original position or do not 'really' move the constituent, but just copy it onto the landing site. In either case, the auxiliary trace would be saved at LF. If this were the case, however, the same should be true for the scrambling trace in (17). Again, a possible conclusion may be that scrambling traces are of a different quality than traces created by Move-a. So far, however, no underlying principle has been proposed that would decide which syntactic processes can 'see' the scrambling trace (e.g. V2-fronting), and which cannot (e.g. relativized minimality).

The picture is further complicated when 3-place predicates are considered:


            a.         ..., daß der Prinz        der Prinzessin  den Kuß     gegeben hat.

                             that the princeNOM the princessDAT the  kissACC given     has.

            b.         Den Kuß gegeben hat der Prinz der Prinzessin.

            c.         Der Prinzessin gegeben hat der Prinz den Kuß.

            d.         Der Prinzessin den Kuß gegeben hat der Prinz.

            e.         Den Kuß der Prinzessin gegeben hat der Prinz.

            f.          Der Prinzessin den Kuß gegeben hat der Prinz.

            g.         *Der Prinz gegeben hat der Prinzessin den Kuß.   

            h.         *Der Prinz gegeben hat den Kuß der Prinzessin.

            i.          *Der Prinz der Prinzessin gegeben hat den Kuß.

            j.          *Der Prinzessin der Prinz gegeben hat den Kuß.

            k.         *Der Prinz den Kuß gegeben hat der Prinzessin.

            l.          *Den Kuß der Prinz gegeben hat der Prinzessin.

If it is correct that scrambling elements leave traces, and if these traces cause ECP-violations, because e.g. they have to be c-commanded by their antecedent at S-Structure (Spell-Out), (19g)-(19l) are predictably ungrammatical. But (19c) should have a structure as shown in (20) and by the same argument should not be grammatical either, which is not correct.  

(20)      [VP Der Prinzessin [VP ej [VP [V gegeben [V ek ]]]]]i             hatk      [VP der Prinzj    den Kußj  ei ]. 

I am thus led to believe that V2-fronting not only fails to provide evidence for scrambling traces, but that it casts doubt on any analysis involving chain formation.

Haider argues that the following scope ambiguities provide evidence for the existence of scrambling traces:


            a.         ..., daß er fast            jedem Lehrling       mindestens einen Trick

                             that heNOM almost every apprenticeDAT at least        one   trickACC

                        beigebracht  hat.

                        shown         has

            b.         ..., daß erNOM [mindestens einen TrickACC]i fast jedem LehrlingDAT ei                  beigebracht hat.

                                                                        (examples from Haider 1993: 205)

In (21a), the dative object has unequivocally wide scope over the accusative object, whereas the latter can have either wide or narrow scope in (21b). Haider (1993: 205) argues that c-command conditions on the surface order[2]of the quantified arguments determine their scope relation in (4a), i.e. the dative object fast jedem Lehrling has unequivocally wide scope. (4b), on the other hand, is ambiguous, which Haider explains with the hypothesis that scope relations can be determined on the basis of the structural positions of both overt arguments and corresponding empty positions. Thus, the ACC object may thus have wide scope, since it c-commands the DAT object, which in turn may also gain wide scope, because it c-commands the empty ACC position.

Instead of assuming that scope relationships are exclusively read off from S-structural c-command relationships, one may take into consideration that default A-structures are availableat LF (as an underlying structural tier). Scope at LF could thus be determined along two dimensions: either on the basis of syntactic (c-command) relations, or on the basis of A-structure precedence (or prominence). Ambiguities may thus arise from a clash of overt syntactic structure with default argument orderings, which would make the assumption of scrambling traces superfluous.

<<   >>   TOC

© Philipp Strazny 1997

[1]               "Postponing" does not mean that a principle like Procrastinate applies: since Merge is cost-free for the P-features in either position, it is not more economical to postpone saturation. The contrary may be the case: since the act of tracing an argument back to its default position may not be cost-free, saturation of P-features may be more economical in the default position. However, this cost does not apply within the context of sentence generation, which may explain that economy only causes a weak 'preference'. 

[2]               Cp. Reinhart (1983), Frey (1994).