370 LAURA Α. MICHAELIS
former case represents the default coding of non-macrorole direct core
arguments, the latter two cases are associated primarily with a particular
marked linking. Hence, the present analysis has provided for this deviant
class of predicates an account of case-pattern selection, case-pattern paral
lelisms, and the relative frequencies of particular case-patterns. This
account might then be said to impose some order upon a seemingly unruly
realm of Latin grammar.
Notes
I would like to thank Robert Van Valin, Harm Pinkster, Joan Maling, and Eve
Sweetser for their helpful comments on earlier drafts of this paper. I would also
like to thank Charles Fillmore, and Paul Kay for valuable discussion. Abbrevia
tions used in narrow glosses of example sentences are: A "accusative", AB "abla
tive", D "dative", F "feminine", G "genitive", IMP "imperative", IMPER "imper
sonal", INF "infinitive", M "masculine", Ν "nominative", and SG "singular".
Abbreviations of text and author names are those used in Lewis and Short's Λ
Latin Dictionary (Oxford: Clarendon Press).
There are several examples of two-place verbs which can take either a dative or
accusative non-subject argument, with concomitant meaning modification. (The
dative here is often said to code "indirect effect".) The following pair illustrates
this case-marking variability:
(i) Senatum consulit.
senate(A) (he)consults
"He consults the senate."
(ii) Reipublicae consulit.
republic(D) (he)consults
"He considers the interests of the state".
Pinkster (to appear) has argued that among verbs which, like consulo, sanction
variable object-marking there is no opposition between dative and accusative
case, but that these two cases code different arguments — an object and a '"be-
nefactive' satellite" — one or the other of which may be omitted in context. Evi
dence for this claim derives from the use of these verbs as three-place predicates
sanctioning both dative and accusative non-subject arguments (cf. Pinkster's
example (22) (op. cit:8). Such evidence would suggest that the so-called variable
object verbs are not in fact two-place predicates sanctioning dative "objects" but
rather three-place predicates allowing object omission.