48 Steven Franks
compositional—it is never a function of the meaning of the lexical item which heads
the phrase to which it adjoins. Thus, for (51) one might imagine that some func-
tional head (whatever immediately dominates VP, e.g. v or Asp or AgrO) values case
on complements to verbs, which is accusative everything else being equal. However,
when V combines its properties with those of that functional head, V adds its own
specific information. For example, V can impart the information that it assigns no
object theta-role in (52a) or that it assigns “quirky” instrumental in (52b). I call the
addition of features from the V head “fertilization.” The crucial idea here is that
the next head down needs to fertilize a traditional functional category probe (and
I remain agnostic as to whether this is head movement, morphological merger, or
even if there is a difference) in order to empower that probe to value case on an argu-
ment. Extending these ideas to the SD, and assuming that dative is valued by C, since
the semipredicative is not an argument it should be able to be assigned dative in any
CP, whether or not a dative subject argument can occur. And just as with accusative
direct objects, the possibility of valuing the subject as dative depends on properties
of the head below C (e.g. Mod or Inf/TINF). Using C in this way allows us to connect
overt dative subjects in Russian with the SD. The same probe, C, values dative on
overt subjects of infinitives in (44) as on the semipredicatives in (7). The only differ-
ence is that the latter, as non-arguments, do not care if C has been fertilized or not.
That is why the SD typically appears freely in infinitival clauses even when these do
not tolerate overt dative subjects.6.1.4 Predicate adjective agreement is local
The most conservative hypothesis is that controllers of agreement must be in a tight
structural relationship with their adjectival targets. Case valuation by probes, on the
other hand, is not so constrained. For ordinary predicate adjectives, this means that
whenever the controller does not seem to be local the relationship must somehow
be reduced to a local one. In Section 3.2 we saw one straightforward way of accom-
plishing this, namely, by movement of the controller. That is, any apparently non-local
agreement relation implicates movement and, consequently, standard restrictions on
movement should be at play. For the SD, on the other hand, case is directly assigned to
the semipredicative, so the relevant C probe can be some distance away. Since ordinary
adjectives can only get case through agreement, a MTC-account, as in (25a) based on
Grebenyova (2005), seems sensible.
This leaves us however with two problems: (i) How can the instrumental on ordi-
nary adjectives be analysed as agreement? – and, the flip side of this, namely, (ii) How
can the nominative (or accusative) on semipredicatives be analysed as direct assign-
ment? There are various ways one might solve these problems. For the first, one could
either adopt Grebenyova’s proposal of assuming an instrumental pro with which the
adjective agrees or one could let instrumental be a true default case here, as suggested