Question about functional approaches

[Liz Bates]

I agree with the other respondents that the issues of innateness and
functionality are orthogonal (at least in principle), although I would
phrase it more in terms of innateness vs. domain-specificity (i.e. the
extent to which language in general and grammar in particular are based on
idiosyncratic principles and mechanisms, free-standing, independent of the
principles and mechanisms at work in non-linguistic domains).  As we all
know, the supposed autonomy/peculiarity/domain-specificity of grammar plays
an important role in arguments about its innateness: if grammar is entirely
idiosyncratic in its structure and operations, then (so the argument goes)
it cannot be acquired by "bottom up" mechanisms that transcend language
(e.g. distributional analysis, statistical induction, pattern extraction)
nor can it be inferred, constructed, constrained or assimilated by "top
down" structures or mechanisms that transcend language (e.g. the kind of
body-based cognition that George Lakoff wrote about as a basis for
cognitive grammar, or the kinds of information processing constraints
(perception, memory, motor constraints, attention) that some functionalists
have proposed). In the absence of a case for bottom-up or top-down learning
mechanisms, grammar must be innate.  The structure of this argument makes
it clear why any claim that language is "caused" by cognition (or grammar
is "caused" by meaning) is viewed with such suspicion in the nativist camp.

In this regard, Brian MacWhinney and I found it useful back in our 1989
book ("The cross-linguistic study of sentence processing") to distinguish
between four levels of functional causation of grammar, ordered from the
weakest to the strongest, in the sense that the higher levels presupposed
the lower ones.  By "functional causation" we are referring both to message
constraints (the cognitive content that underlies those semantic structures
that are mapped onto grammar) and to information processing constraints
(perceptual, memory, motor output, etc.).  The four levels refer to the
ways in which these constraints could operate to affect the nature and
shape of grammar.  In a nutshell:

Level 1 functionalism refers to claims that grammars are the way they are
because of functional constraints on evolution, i.e. the process by which
grammars appeared in the first place.  Note that, at this level, no claims
are made about cause-and-effect relations in today's speaker-listeners, nor
about the effect of functional constraints on language acquisition, nor
about the need to include functional categories and operations within our
characterization of the grammar (i.e. in competence models).  In other
words, Level 1 functionalism does not preclude strong, domain-specific
nativist beliefs.  I interpret the well-known Pinker & Bloom paper on
Natural Language and Natural Selection to reflect a kind of Level 1
functionalism.

Level 2 functionalism presupposes Level 1, with an additional claim:
specifically,  that the same functional constraints that brought grammar
into the world continue to operate synchronically, maintaining grammars in
their current form and (perhaps) bringing about continuing change (i.e.
historical language change following inter-language contact, gradual
erosion of morphological markers, and so on).  Note that one can embrace
Level 2 functionalism without assuming that the cause-and-effect relations
that operate in adult discourse are available to small children as an aid
in language learning (indeed, many of these facts may be completely
inaccessible to children, of no use at all in learning).  Hence one can
still be a nativist, assuming that grammars are unlearnable without
substantial a priori knowledge.  Nor is it necessary for a Level 2
functionalist to build non-linguistic constraints directly into his/her
characterization of the grammar (i.e. one can maintain a modular
architecture, with an autonomous grammar).

Level 3 functionalism presupposes Levels 1 and 2, with a further claim:
that children can exploit the form-function correlations used by adults,
rediscovering in each successive generation the reasons why grammars look
the way they do.  In other words, the functional constraints described at
Levels 1 and 2 also operate causally within the learning/discovery process.
Although this approach eats away at strong nativism (because less
grammar-specific innate knowledge is required), it is still possible to
maintain some kind of modular architecture in which grammar eventually
"decouples" from the rest of cognition.

Level 4 functionalism presupposes Levels 1 - 3, with the additional claim
that grammar cannot be adequately described without reference to the
functional constraints (structures and operations) that play a causal role
in its evolutionary history, current use by adults, and acquisition by
children. In other words, modular accounts must be abandoned in favor of
accounts of linguistic knowledge that are "functionalist all the way down."
This is, I take it, where most functionalist and cognitivist grammarians
take their stand.

In our 1989 proposal, Brian and I suggested that each of these four levels
requires different kinds of empirical evidence.  Lower forms of evidence
(in the logical sense outlined above) are required for the higher levels,
but are not sufficient to establish claims at the higher levels.  We hoped
back in 1989 that this taxonomy might prove useful.  It seemed to work for
us.  Maybe it will be useful to some of you in the current discussion.
-liz bates