universality of grammatical functions

[Rachel Nordlinger]

Joan and I have been having a discussion by email over the past few days
that we feel would be an interesting one for this list, so we are moving it
here in the hope of stimulating some discussion.

The issue has arisen out of discussions I have had with people here (at the
Max Planck Institute) who work on languages for which there is apparently
no evidence for grammatical functions distinct from thematic roles.  One of
these is the Australian language Jaminjung, which has no controlled
clauses, relative clauses, valence changing operations, etc. that can be
used to identify grammatical functions such as "subject" and "object".
Rather, the few syntactic processes that there are can all be described
solely with reference to thematic roles.  The researcher working on this
language has therefore found Construction Grammar to be more useful than
LFG, since LFG requires her to assign grammatical functions (in the
f-structure), while Construction Grammar does not.  So, my general question
is, is this a drawback for LFG?

As a way of initiating discussion, here is the exchange between me and Joan:



Rachel Nordlinger, Tuesday June 9:

>I was wondering if you know of any work that has been done on
>'active/stative' languages in LFG --- i.e. languages like the Mayan
>languages that arguably have no evidence for grammatical functions such as
>subject and object distinct from thematic roles like agent and patient.
>Such languages would seem to challenge the LFG assumption that all
>languages have grammatical functions like SUBJ and OBJ in the f-structure.
>I am asking because there are a number of people who are working on
>languages of this type here, and who are attracted to Adele Goldberg's
>Construction Grammar approach since it does not force you to assume that
>grammatical functions exist universally as LFG seems to do.  I am sure
>that this is something that has been thought about within LFG before, but
>I myself can't think of a good solution....I was wondering if someone else
>has come up with one?


Joan Bresnan, Tuesday June 9:

> The best analysis of an active language I have seen in LFG is Wayan Arka's
U. Sydney Ph.D. dissertation on Balinese:

	I Wayan Arka. 1998. From Morphosyntax to Pragmatics in
	Balinese: A Lexical-Functional Approach.


The issue of active languages and grammatical functions has been
around for a long time--similar points about the nonuniversality of
Subject, Object, etc. were made by Foley and Van Valin in their work
on Role and Reference Grammar, and before that, by Schachter in his
work on Philippine lgs.  I believe that Durie in his Grammar of
Acehnese also echoes this point.

However, I don't think this argument against GFs is valid.  It has the
same logic as the following (fallacious) arg: Dependency grammar is
better than LFG because it doesn't force you to assume a c-structure
for every language, and there are languages like Jiwarli that have no
c-structure.

"Subject" in lfg is not assumed to have either a uniform expression
across lgs or a uniform set of semantic roles.  Rather, there is a
flexible many-to-many mapping between a-structure, f-structure, and
c-structure, constrained by some universal principles (lmt together
with the theory of structure-function mappings).  The same goes for
the theory of phrase structure variation (as in Ch. 6 of my book ms,
and your book).

Finally, I don't want to oversimply, but if you squint, Construction
Grammar is just LFG with the f-structures annotated on the c-structure
nodes, and a few labelling differences with functions (coarser
functions are used for c-structure positions--cf. my Ch. 6, and
Optimal Syntax).  It uses some formal ideas that were anticipated in
Manning and Andrew's  "Aussie LFG".


Rachel Nordlinger, Wednesday June 10:

>Thanks for your reply.  So, I guess what it comes down to is that for some
>languages for which there is no evidence for grammatical functions
>distinct from thematic roles of agent and patient, the separation of
>thematic roles in a-structure and grammatical functions in f-structure in
>LFG is just redundant.  So, it is not problematic to account for these
>languages in LFG, it is just that for these languages LFG may require you
>to make a distinction between agent in the a-structure and subject in the
>f-structure that is terminological only.


Joan Bresnan, Wednesday 10 June:

>What Arka/Manning/others have shown (building on Kroeger) is that
there is a split between the a-structure "subject" and the f-structure
"subject" in terms of binding, relativization, control, etc.  There is
claimed to be a universal correlation between which syntactic
phenomena can reference a-structure subjects and which reference
f-structure subjects.  It is not merely a terminological distinction,
it is substantive and makes typological predictions.


Rachel Nordlinger, Wednesday 10 June:

>Right, I understand that.  But the point is what of languages that are
>claimed not to have this split between a-structure "subject" and
>f-structure "subject"?  For example, Jaminjung (an Australian language
>being worked on by Eva Schultze-Berndt here at MPI) has, as far as Eva can
>tell, no syntactic processes that pick out f-structure "subject" as
>opposed to a-structure "subject".  There are no controlled clauses, no
>relative clauses, no valence changing derivations, etc.  So, for a
>language like this, the distinction between a-structure and f-structure
>"subject" does seem to be more one of terminology, rather than substance.


Joan Bresnan, Thursday 11 June:

>Well, consider the analogy with phrase structure configurationality.
If all we had were highly configurational languages, you could argue
that GFs are redundant, since they are in (near) one-to-one
correspondece with p.s. positions/configurations.  And indeed, many
generative linguists have argued this position.  But from a
typological perspective, this is not an attractive position.  We
assume that real variation exists in the structure-to-function
mapping, so that some languages have more flexible or rigid mappings
than others.  We *wouldn't* say that the distinction between GF and PS
configuration is merely "terminological"--we wouldn't say that even
for one language with a rigid structure-function mapping.  So why say
the analogous thing for Jaminjung, just because it has a rigid a-s to
f-s mapping?


Rachel Nordlinger, Thursday 11 June:

>Yes, I see your point. So maybe calling the distinction "terminological" is
not exactly what I mean.  What I mean is that a researcher analysing
Jaminjung, in which all syntactic properties can apparently be stated in
terms of thematic roles like agent and patient has to make some decision,
when working within LFG, as to which of these to call SUBJ in the
f-structure; and since there are apparently no controlled clauses or
relative clauses or anything that would help you to make the decision on
typological grounds, the decision could just as easily go either way (based
on language-internal evidence, that is).  So, if you're a descriptive
linguist on this language and
you're looking for a framework to work within, a framework like
Construction Grammar, which doesn't require you to make this decision
(although still allows the distinction to be made for languages in which it
is substantive), might appear more attractive than a framework like LFG,
which forces you to attribute grammatical function labels to these thematic
roles when there doesn't seem to be an language-internal reason to do so.