universality of grammatical functions

[Tara W Mohanan]

Thank you, Rachel, for moving your interesting discussion to the public doma=
in.

We would like to make a couple of clarificatory comments:

Note 1: By "universal", we mean a construct or law that is available in
linguistic theory, but need not necessarily appear in every language. Take
the feature [=B1 distributed]. We can assume that it is a universal feature
without claiming that every human language must exhibit the contrast
between [+distr] and [-distr] in terms of phonological phenomena whose
analysis crucially appeals to this distinction. Similarly, the assumption
that the constraint against *[+nasal, -voice] is universal does not entail
the claim that every language exhibits this constraint. If we accept this
conception of "universal", the hypothesis that SUBJ is a universal does not
entail the claim that every language exhibits linguistic phenomena whose
analysis crucially calls for the construct of SUBJ.

Note 2: A model for a class of objects may call for the specification of
values along a set of well defined parameters, but quite often it may be
difficult to choose a specific value for a given object in terms of the
parameters of the model. There is nothing undesirable about this. Take a
model of phonology that uses underlying representations and phonetic
representations. Alternations between schwa and full vowels (as in
solid/solidity, drama/dramatic), coupled with the distributional
restrictions on schwa, justify the assumption that schwa is not an
underlying segment in English. This means that all schaws in phonetic
representations must correspond to a full vowel or null in underlying
representation. But when it comes to instances like the first vowel in
'about', we have no evidence from alternation to make a choice from a set
of vowels permitted by the grammar, say between /=E6/ (in hat) and /o/ (in
hot).  If we agree that this is not necessarily a *defect* of the model or
the grammar, we must accept that the relevant evidence for the options
required by a model may not always be available, and that such absence of
evidence does not constitute a defect of the model, even though it is
frustrating for researchers.

When faced with the unavailability of evidence to choose between "guilty"
and "not guilty", conventions of the legal system demand that we choose the
unmarked option "not guilty", a choice that is justified on moral grounds
if not on epistemological grounds. When faced with the unavailability of
reasons to choose between "true" and "false" in mathematics, mathematicians
use the term "conjecture" to refer to propositions that have not been
proved to be either true or false. If linguists were to follow the jury's
practice, we would say something like, "In the absence of evidence, I will
adopt the unmarked option of assuming that argument-x is linked to GF-i."
If we follow mathematicians, we would say something like "In the absence of
evidence, I will not make a choice between GF-i and GF-j for argument-x."

In the light of the above comments, languages in which we have not found
phenomena that call for reference to grammatical functions do not
constitute an argument against the universality of grammatical functions.
We don't see what light the unavailability of relevant evidence in Mayan
languages or Australian languages by itself can shed on the choice between
LFG and Construction Grammar.

Mohanan and Tara

----
Tara W Mohanan, Dept of English & Literature, NUS