universality of grammatical functions

[Joan Bresnan]

Responding to Paul Kay's comments:

> It is useful to distinguish between a formal architecture and a
> particular theory, or batch of theories, expressed within it.  The formal
> architectures of LFG and CG are very similar -- perhaps identical, viewed
> at a sufficiently abstract mathematical level.  This is not necessarily
> the same thing -- and to all appearances not the same thing in fact -- as
> the substantive commitments of the grammatical theories that have been
> constructed within those architectures. 

This certainly true, and it is a point which has been often
misunderstood over the past couple of decades.  (Witness those
well-meaning but underinformed linguists who believe that
non-Chomskyan theories are "explanatorily weak" just because they may
have powerful formal architectures.)  I've emphasized this distinction in
my own writings about the LFG formal system.

Unlike many current theories of syntax, LFG does not attempt to build
the substantive theory of universal grammar into its formal
architecture--the former is treated as a metatheory over the space of
structures provided by the latter.  This division in fact is what
makes it so easy to incorporate LFG into an Optimality Theoretic
setting.  A universal LFG (the formal system) can generate the
typological space of structures, which can then be optimized in
language-particular ways against the universal constraints (the
metatheory, which can largely be stated in the formal constraint
language).

>      [...]                   The theory of CG does
> not contain a commitment to the universal prominence of grammatical
> functions, unlike LFG (as I understand it). Such a theoretical commitment
> could be expressed within the CG architecture, just as one could formulate
> a theory within the LFG architecture which lacked commitment to the
> universality of grammatical functions.  (Of course one might want to find
> a different name for this theory.)
> 

My impression is that substantive linguistic work in LFG has much
stronger universalist aims than substantive linguistic work in CG.  CG
celebrates constructions in all their rich and irreducible
idiosyncracy (the more, the merrier), while LFG attempts to provide an
abstract universal characterization of the typological space of
possible constructions.  Grammatical functions represent a very high
level of abstraction; they are essentially equivalence classes of
possible mappings between semantic roles and expressions.  This level
of abstraction is useful in (among other things) attempting to provide
a typological theory of alignments: active/stative, or ergative, or
accusative, etc., as well as in extracting generalizations from
languages with flexible relations.  But like all abstractions, one can
always eliminate it by global substitution of the extension for the
concept.

I like the CG focus on irreducible language particularities; but I
also like the LFG/OT focus on the universal structure in the
typological space.

Joan