universality of grammatical functions

[Joan Bresnan]

In reponse to Rachel's msg:

>I agree with this definition of universal [as available to all lgs,
not necessarily present in all--jb].  However, it seems to me
that LFG accords a more primary position to grammatical functions than
it does to other universal features of this type.  For example, in the
theory of c-structure lfg is much freer: while VP is a universally
available category, if a language has no evidence for its existence
it simply is not present in that language.  This is different from the
approach to grammatical functions, which must be present in some form
in all languages.  So, my original question to Joan was prompted by my
own thoughts as to whether this prominence of grammatical functions in
lfg was a strength or, as some people here seem to believe, a weakness.
------

This sounds like a reasonable approach: why not make everything
optionally available to all lgs?  So some might lack a VP, others
might lack a Subj, still others both.  But when you think a bit more
about it, there is an important asymmetry between functions and
categories in lfg.  The inventory of c-structure categories can vary
across lgs precisely because the GFs are there to do the work of
mapping from expressions to roles.  

A framework/theory/analysis that eliminates grammatical functions from
some active/stative language must still capture the relational
generalizations that exist in the lg.  If complex c-structure
hierarchies are also eschewed, then recourse will be made to
abstractions over semantic roles.  Thus Durie in his (beautiful)
grammar of Acehnese captures the agreement of the pronominal clitics
with the NP arguments they may crossreference in terms of the role
abstractions Agent and Undergoer.  The Agent pronominal clitic is
obligatory; it agrees with an Agent NP argument in person.  So if we
were to formalize this description in our generic feature-logic
framework, we would have something like:

	Acehnese agreement:  [ Agent [ Pers  2]
                               Undergoer [ Pers 3] ]

Compare English, where the Subject agreement marker, also obligatory,
specifies person, number, and gender information of the Subject:

	English agreement:  [ Subj [ Pers  2 
                                     ...     ]}

What we find is that roles like Agent, Undergoer (or whatever the
choice may be) must bear exactly the same kinds of grammatical
attributes as Subject and Object in other lgs.  These are grammatical
functions by another name.  (As Bruce Mayo argues in his msg, *some*
abstraction over individual roles is necessary.)  Hence the proposal
to eliminate the universality of Subject, Object, etc as grammatical functions
succeeds only by introducing a larger set of universally available
grammatical functions: Agent, Undergoer, Actor, Subject, Object, etc.
This is all really a notational variant of LFG; its essential
difference from LFG lies in a different choice of labels for
attributes, as I indicated in my first message to Rachel.

In this sense, Rachel's point that the differences between CG and LFG
are "terminological" seems absolutely correct.

Joan






          





*----------------------------------------        ______     __o       __o
Joan Bresnan	bresnan at stanford.edu           ______     _`\<,_    _`\<,_
*----------------------------------------       ______   (*)/ (*)  (*)/ (*)