480 MICHAEL SILVERSTEINthan those of the minimal set coding Agent-Patient-Subject. For some lan
guages, there is incredible formal consistency in the mode of expression,
leading to such large suffixal paradigms, or elaborate form-order classes of
cross-referencing pronominal series, or elaborate sets of uniform clitics or
pre- bzw. post-positions on noun phrases. For other languages, there is an
abrupt or gradual differentiation of sets of formal devices from the most
central, the minimally "grammatical" cases, down through the rest of the
system, even becoming indeterminate as to real "case" like status without
some language-external criteria. For languages which make such a sharp
differentiation, the question is what are the interactions between specific
peripheral and specific central case-markings (this is the whole defining
problem of relational grammar, as noted), in the various transformations of
the propositional schemata.
Considered in this way, we have broadened the dependency of case-
marking from its relationship to inherent lexical content, to consider in
addition the nature of the predicate-argument relationships case-marking
purports to code. We have been claiming that you cannot know how to dif
ferentiate case-marking phenomena and properly to label them until you
locate the formal devices in the set of language-independent possibilities
dictated by an hypothesis about predicate-argument types as well as inher
ent lexical content types. Surface data, even in transformational batteries,
are uninterpretable without such an ordering hypothesis, so that we know
that, say, the active:passive relationship is coded with mirror-image formal
case-markings and NP-Predicate relationships in a "nominative:accusative"
system by comparison with the active:antipassive relationship of an "erga-
tive:absolutive" system. This and this alone allows us to use formal indi
cators to make analytic and explanatory arguments about surface forms, at
the level of noun phrases functioning in clause-level structures.
Now let us turn to multi-clause structures. It seems to me that we must
treat this factor with a view to explaining reference-maintenance in both
compound and complex sentences and typical discourse contexts of sen
tences (even those with only one clause) in sequence. In fact, in many lan
guages, the formal surface structures of many types of logically-related
clauses in sequence is indistinguishable from what we would get from
underlying embedding plus extraposition; we will see this in Chinookan in
one sense, in Djirbal in another. There seems to be a hierarchy of tightness
of linkage of clauses, from minimal linkage of clause sequence in indepen
dent sentences that just happen to be in the order S 1 S 2 -... with no implica-