This suggests that the character of a semantic argument is not always sufficient to constrain the category of the
corresponding syntactic argument. To be sure, semantically related verbs usually are rather close in their syntactic
behavior (Pinker 1989; Jackendoff 1990a; Levin 1993), but the detailed semantic differences do not seem to correlate
reliably with detailed syntactic differences. Thus we tentatively conclude that verbs can stipulate the category of their
syntactic arguments, though not every verb does. (The standard term for such stipulation, going back toAspects,is
“strict subcategorization.”)^71
Another aspect of syntactic argument structure that cannot be entirely predicted from semantics concerns the
“governed prepositions”that are sometimes required to glue heads to their syntactic arguments. (46) offers examples,
some repeated from the previous section.
(46) a.Susan provided Amywitha cupcake,vs.
Susan served Amy (*with) a cupcake.
b. We countonyou.vs. We trust (in)you.
c. The road approaches the castle,vs. the approachto/*of thecastle.
d. We talkedwith/toJane.vs. our talkwithl*toJane
e. pridein/*of Billvs. proudof/*inBill
f. angryat/withHaroldvs.madat/*withHaroldThemembersofthese pairsdiffer minimallyintheirsemantics butdodifferintheirgovernedprepositions. Sometimes
the choice of preposition is semantically motivated (for instancethe approach to/*with the castle). But when words are as
closely related as the adjectivesangryandmad, there seems little choice for the governed preposition other than pure
stipulation. Thetentativeconclusionis that verbsdo at least sometimes specifysome ofthesyntax (and phonology!)of
their syntactic arguments. (We return to strict subcategorization and governed prepositions in section 6.6.)
THE PARALLEL ARCHITECTURE 141
(^71) Grimshaw (1979; 1990) suggests that a verb can stipulate the syntactic category only of arguments within VP. Thus if an argument appears elsewhere, its category is
sometimes constrained only by semantics. The interesting cases are pseudo-clefts: compare (45b) with (i).(i)What John didn't express was that he was disappointed.In (i) a
clausalargumentis possible,eventhoughthe“normal”positionintheVP is restrictedtoNP, as seenin(45b).Thus thesesyntacticallyunrestrictedpositionsletthesemantic
realizations of the argument shine through more clearly.