Case assignment versus licensing
At the outset we suggested that Case assignment and Case licensing might well be distinct phenomena, which would lead to a very different view of the Case Filter than has standardly been supposed. We can now provide further evidence for this view based on a subject-object asymmetry with respect to lexical Case in Icelandic. As noted in section 5, the selection of lexical Case in object position can license the occurrence of a lexical NP in object position of a passive predicate, in contrast to the configurationally accusative Casemarked object, which is not licensed in the same position. With a lexical Case selected subject, however, selection is not sufficient to license the occurrence of a lexical NP in subject position.
The asymmetry in Case licensing for lexically Case-marked subjects versus objects shows up when we compare the failure of passive morphology to block (lexical) Case licensing with the following paradigm.
(1)
The bracketed construction in (1) is an infinitival sentential subject. Crucially, the verb in this construction is one that selects a lexical Case subject (in the dative). Given the PLS, we assume that PRO is Case-marked dative so that the lexical property of Case selection is satisfied in (1a). The fact that a lexical subject in the selected Case is not possible in this construction shows that lexical Case selection for subjects is not sufficient to license the presence of a lexical NP in that position. Thus, (1b) is ruled out as a violation of Case licensing. Under this analysis, the lexical Case subjects in the following examples must be licensed configurationally, and not via lexical Case selection.
(2)
In (2a–b) the actual Case of the subject of batna is determined by the verb, but the licensing of the lexical NP is done independently in terms of its structural position. In (2b), for example, the licensing of the lexical NP Jóni is done by the matrix verb tel, which governs the NP. Thus, Case assignment and Case licensing appear to be distinct processes.
This same asymmetry holds with respect to lexically Case-marked derived subjects where lexical Case selection is to an object position. The paradigm corresponding to (1) is given in (3).
(3)
Given the PLS, we assume once again that PRO is Case-marked dative to satisfy the lexical Case selection property of hjálpa (namely, that it selects a dative object). Though this Case selection is sufficient to license the presence of a lexical NP in the object position of the passive predicate, it has no effect when the lexical object is moved to subject position, as illustrated in (3b). Thus, in (4) we have yet another example where Case assignment is determined via lexical Case selection, whereas the licensing of the lexical NP is done configurationally.
(4)
In this way, (3)–(4) provide striking confirmation of the subject-object asymmetry for lexical Case selection and that Case licensing is primarily a configurational phenomenon, with the exception of lexical objects in passive constructions, which can be licensed by lexical Case selection.
The separation of Case assignment and the licensing of lexical NPs leads us to a reconsideration of what have been assumed to be Case Filter effects. Our investigation of lexical Case phenomena suggests that the determination of the Case of an NP is not the relevant factor; rather, it is whether a lexical NP occurs in a configuration that licenses the presence of a lexical NP. If this view is correct, then the Case Filter as formulated in section 1 should be replaced by a licensing principle along the lines of (5).
(5)
Under this analysis, (1b) and (3b) constitute violations of Case Licensing.
Though it seems that the Case-licensed positions are exactly those where a configurationally Case-marked NP can occur, there is in Icelandic and German the notable exception of the object position for passive predicates, where a lexically Case marked NP can occur but the canonical configurational Case (accusative) cannot. One way to deal with this exception is to assume that a lexical NP can occur only in positions that receive a “structural index.” Thus, objects of verbs and prepositions receive a structural index under government, and presumably subjects of finite clauses receive such an index via identification with respect to agreement. If, for reasons yet to be determined explicitly, a passive predicate cannot assign such a structural index, then its lexical object must get one from some other position. If, however, lexical Case selection for objects can assign such an index as a marked option (for instance, Icelandic versus Russian), then the Icelandic facts would follow. Note that we would have to assume that assignment of the structural index is to a position and remains on the position even when a lexical NP in that position is moved to another position (as in (3b)). Assuming that structural indices are assigned within a government domain, the asymmetry of licensing with respect to lexically Case-marked subjects and objects would follow since subject position (the position designated as SPEC of IP) is outside the government domain of V.
What we have demonstrated in this paper is that lexical Case phenomena manifest some rather different properties from configurational Case phenomena. With configurational Case phenomena, Case assignment and Case licensing are not distinguished. The analysis of lexical Case, in contrast, requires a distinction between the assignment and licensing of Case—the former being a lexical property of certain heads and the latter a configurational property of constructions. As we have shown with the analysis of Icelandic, it is licensing rather than Case assignment that distinguishes well formed from ill-formed constructions. Since this analysis will apply equally well to configurational Case phenomena, it is possible to revise Case theory by replacing the Case Filter with a principle of Case Licensing, as we have proposed.
