Advances in Role and Reference Grammar

84 ROBERT D. VAN VALIN, JR.

analyses that the extent of the cross-linguistic validity of steps like 2c in

(70), for example, can be determined.

5.3 Lexical vs. syntactic phenomena

Over the past decade most grammatical theorists have come to distinguish

between lexical and syntactic phenomena, starting from the criteria put

forth in Wasow (1977).^49 While the criteria for deciding the clear cases are

generally agreed upon (e.g. lexical idiosyncracies should be treated lexi­

cally, exceptionless phenomena syntactically), the criteria for distinguishing

the two classes of phenomena are ultimately theory-internal. In RRG the

line between the two is clear-cut and falls out from the linking algorithm

in Figure 16: lexical phenomena affect the LS of the predicate, its argument

structure, and actor and undergoer assignment, whereas syntactic

phenomena deal with the morphosyntactic realization of the macroroles

and other core arguments. This may be represented as in Figure 21.

The importance of the lexical-syntactic contrast can be seen in the fol­

lowing phenomena from Georgian and Japanese. In Georgian, according to

Harris (1981), only "initial subjects" can control reflexivization; in RRG

terms, control of reflexivization is a property of syntactic pivots only, and as

argued in Van Valin (1990a), the hierarchy for pivot in Georgian is actor >

highest ranking non-actor direct core argument. Causative, inversion and

passive constructions present an interesting problem with respect to the

determination of the controller of a reflexive possessor. Georgian causa-

tives are morphologically derived accomplishment verbs which have two

potential actors, the effector of the causative predicate and the effector of

the caused predicate. The LS of such a form will be discussed in detail in

section 7, but it can be given the preliminary representation in (71).