Universality of GR's

[Joan Bresnan]

I think Chris Manning's point is very well put, and Avery's proposal
is also cogent.  (But why call this recent work "lfg-like"?  I am
Joan-like, I suppose, but why put it that way?)  I think the idea that
functions are decomposed along the two dimensions (i) core
vs. periphery (or term/non-term) and (ii) subject vs. nonsubject (or
pivot/non-pivot) is useful.  As Avery notes, it would let us
hypothesize that some languages are degenerate along the
subject/nonsubject axis; the typological consequences would be an
absence of all syntactic subject properties--e.g. passivization,
raising, etc.--and use of semantic indices to differentiate among core
arguments.  [Such a proposal came up in the discussion of Peter
Austin's presentation about an Austronesian lg at lfg96 in Grenoble, I
recall.]  A further possibility is the opposing language type in which
functions are degenerate along the core/periphery axis, leaving only a
subject core argument, all else being oblique or adjunctive.  It seems
that such languages might also really exist...

However, I am still somewhat skeptical of the argument that the
absence of the subject-nonsubject distinction can explain the
typological clustering associated with active languages, such as the
absence of passive, raising, etc.  The reason is that I recall very
well Ken Hale's explanation in his early work on W* languages of why
Warlpiri lacks a passive: in a radical nonconfigurational lg having no
phrase structure, the passive transformation couldn't gain a
toe-hold--NP-movement requires configurational phrase structure to
operate on.  At the time I first heard it, this argument really
stumped me.  I didn't believe that there was really NP-movement, but
what could explain the absence of passive, raising, etc. in a language
like Warlpiri?

As Chris astutely points out, a parametric explanation for the lack of
a passive in active languages depends on the genuine clustering of the
proposed typological properties.  Even if we accept the explanation
for why Acehnese lacks passive, raising, applicatives, etc., we still
have to explain why Warlpiri lacks passive, raising, applicatives,
etc.  And Warlpiri clearly has both the subject/non-subject and the
core/non-core distinctions among its arguments; it is not an active lg.
My impression is that many factors are partially correlated with
absence of passivization, ranging from morphology to discourse, and
there is also historical development to consider.  

For this reason, my own best guess about active languages remains that
they have a relatively rigid mapping between roles and functions:
Actor = Subject, Undergoer = (Primary) Object, in this lg type.  Apart
from the typological issues just mentioned, the only objection to this
proposal that I see is that it entails that there are intransitive
verbs whose sole overt arguments are not subjects, and this
contradicts Dixon's definition of subject (which rests on
transitivity).  We are left, then, on this approach with the problem
of redefining Subject, Object, etc. in a way that differs from the
most commonplace, familiar, and simple notions.

Can this be done?  We can use the core/periphery distinction, and then
among the core arguments, distinguish one that is more prominent than
the others along some grammaticalized dimension--perhaps in topicality
(the English subject) or in semantic role (either controlling the
eventuality-the Acehnese subject, or determining its temporal
structure--a la Dowty's incremental theme idea applied to a
syntactically ergative language with pt/th subject) or possibly in person
(in *some* inverse lgs, perhaps Ojibwe?, where 1/2 person must be the
subject). 

The quality of prose in this last sentence tells me that I'd better
stop.

Joan-like