Bruce,
There is a sense in which
cognitive approaches involve string-level phenomena. To the extent to which a
cognitive linguistic theory imports concepts from psychology, it has to
acknowledge “chunking” in some way. “Chunking” here can
be seen as a not-specific-to-language equivalent of “constituency.”
Research in the 50s, for example, showed that when subjects who didn’t
play chess were presented with a chess board in mid-game, and then distracted
for a moment, they could later recall where only a few of the pieces were;
chess Grand Masters, on the other hand, could position every piece given the
same amount of exposure. It wasn’t because the Grand Masters had better
memories, really – it was because they didn’t see a whole bunch of
pieces; they saw four or five patterns they were familiar with. In general, we
appear to recognize that particular sequences or visual arrangements constitute
“chunks,” and handle large amounts of information by reducing it via
hierarchically-organized chunking schemes. The same thing happens with
telephone numbers – we remember “area code + prefix + number”
as basically three chunks, rather than as ten numerals, and there’s a
syntax to those chunks. As someone who watches a lot of very bad
science-fiction movies, I know the major chunking sequences of the genre quite
well, and can use that knowledge to make predictions (“Next, the threat
posed by <insert monster HERE> should cause the estranged couple to agree
to temporarily lay aside their differences, clearing the way for their
re-engagement during the last five minutes”).
You’ll notice, though,
that I started with the term “cognitive approaches” rather than “Cognitive
Grammar.” The capitalized version is a specific theory founded by
Langacker, and I honestly don’t know exactly how it instantiates chunking
(although I should, and I’m sure it does). I do know that
Stratificational Grammar, another cognitive approach, has no trouble with
constituents/chunking at all, and has moved toward a neural-network
interpretation of it.
A relevant point of contrast
here is that Generative approaches tend to approach constituency as a kind of
abstract, static phenomenon (“movement” rules don’t really
entail actual movement, so the effects of movement on constituency don’t
involve actual changes during production, etc.). Constituents exist as a kind
of set of potentials in an abstract space; the grammar determines what the
potentials are. Cognitive approaches tend to approach it as in issue involving real-time
processing and storage, and thus motivate constituency partly via performance.
That breaks down the performance/competence distinction, and thus creates a
fundamental mismatch between the two platforms. I think functionalists in general
don’t mind claiming that performance (including comprehension in a social
context, rather than just production) partly creates competence as an
epiphenomenon, but Generativists take the opposite view. It’s Aristotle
vs. Plato all over again (but then, that’s the history of social science
for the past two millennia).
Bill Spruiell
Dept. of English
Central Michigan University
From: Assembly for the
Teaching of English Grammar [mailto:[log in to unmask]] On Behalf Of Bruce
Despain
Sent: Monday, February 09, 2009 3:09 PM
To: [log in to unmask]
Subject: Re: Quick note on SFL and Cog. Grammar
As long as we’re
commenting on linguistic approaches, allow me add a few words about
“generative grammar.”
I believe that generative
grammar was originally concerned (as many linguists still are) with the
structures (in the brain, like the CG linguists) that enable language.
For example, Chomsky, in his 1965 Aspects of Syntax pointed out five
primitive structures, as it were, that characterize human language syntax:
nesting, right-branching, left-branching, multi-branching, and
self-embedding. These devices of language may be imitated by generative
algorithms, but I have yet to see any other approach characterize them in any
other way. In that sense, perhaps all approaches have a generative
base.
1.
I
called the man who wrote that interesting book up. [nesting]
2.
That
man who the boy who the students recognized pointed out is here.
[self-embedding]
3.
Tom,
Bill, John, and several of their friends are coming. [multiple branching]
4.
John’s
brother’s father’s uncle died. [left-branching]
5.
The
uncle of the father of the brother of John died. [right-branching]
There must be structures or
switches in the brain that help us keep track of these relationships. Is
that simply part of our general cognitive ability? Perhaps. I need
to see some examples of other behavior that uses these structures. That
might settle the argument for me.
Our general cognitive ability
seems much easier to map to the structure of a neural network. The
structure of the syntax of a string, which is characterized by generative
rules, is perhaps part of, but mostly quite distinct and specialized from that
of a neural network. Such a network allows associations and connections
of a great variety and at varying strengths and “distance.”
When we express the ideas and concepts of a neural network, we are obliged to
stringify it, i.e., map parts of it into a one dimensional string of symbols.
Each one of these symbols (“clumps of concepts”) has associations
that can in turn be mapped to new strings.
My impression is that Grammar is
involved principally with strings of symbols and must relate to syntax in the
above sense. On the other hand, the linguistic approaches of SFL and CL
take a few steps back (higher?) and create other kinds of structures that do
not have the syntax of a string. These structures move away from the
constraints that language syntax has on the form of our expression and concentrate
on the effects of other aspects of the neural network.
From: Assembly for the
Teaching of English Grammar [mailto:[log in to unmask]] On Behalf Of Spruiell,
William C
Sent: Monday, February 09, 2009 11:48 AM
To: [log in to unmask]
Subject: Quick note on SFL and Cog. Grammar
Dear All:
I’d like to address one point in the recent debate
about developmental phases of grammar – but I want to be careful to
emphasize that it’s a very focused (in other words, it doesn’t have
a large impact on the debate as a whole, but hey, it came up). And I
think I may be able to address it noninflammatorily (Word just red-lined that,
but I have a deriving license).
Halliday is quite clear about his grammatical model being a
statement about social practice, rather than about cognition. In a sense,
he’s recapitulating an old trend in linguistics: we’re much more
confident with statements about what we observe going on than we are with
statements about what we think might be going on in people’s heads,
unless we have some way to measure the latter directly. He’s also from
the “hocus-pocus” approach to linguistics rather than the
“god’s truth” approach, for the same kinds of reasons. In
other words, if the grammar describes what’s going on well, and acts as
an explanation insofar as it lets you predict the kinds of things
you’ll encounter, why go out on a limb and claim Full Truthiness?
Cognitive grammar, a la Langacker and others, is a
“god’s truth” model, and does make claims about
what’s going on in people’s heads. It would thus seem at first to
stand in opposition to Halliday’s – and it does, if the only
dimension we’re organizing along is the internal/external-phenomena one.
There are, however, other dimensions along which CG and SLF
tend to “cluster” together. Both acknowledge that social context
directly affects what is produced, and more importantly, both consider the
social environment to have a *direct* effect on the basic structure of
language. In most Generative approaches I’m familiar with, any kind of
selection effect due to social context is outside the scope of
“grammar” – the grammar defines the set of what is within the
realm of possibility, and that set has nothing to do with social interaction.
In a particular social situation, a speaker might choose a subset of that set,
but that’s not an issue for the grammar. There’s a sense in which
sociolinguists, to generativists, are looking at something fundamentally
different from what “core linguists” look at.
There’s an additional reason for CG/SFL clustering,
but it’s one that exists “outside” either theory. CG,
by its nature, must also acknowledge that cognitive processing constraints
– short term memory limitations, etc. – have a direct effect on the
structure (and structures) of language. SFL theorists have spent a fair amount
of time describing what Halliday terms the “textual metafunction,”
which among other things is concerned with maintaining new vs. old information
contrasts, and cohesion. The kinds of constructs one needs for the textual
metafunction happen to dovetail fairly well with notions of processing
constraints – in other words, part of CG may be quite useful as a kind of
backdrop to SFL and vice versa. In generative models, the *basic*
structure of language has nothing to do with more general nonlinguistic
processing constraints, so there isn’t that kind of “contact
point” between the theories.
Sincerely,
Bill Spruiell
To
join or leave this LISTSERV list, please visit the list's web interface at: http://listserv.muohio.edu/archives/ateg.html
and select "Join or leave the list"
Visit ATEG's web site at http://ateg.org/
NOTICE: This email message is for the sole use of the intended recipient(s) and
may contain confidential and privileged information. Any unauthorized review,
use, disclosure or distribution is prohibited. If you are not the intended
recipient, please contact the sender by reply email and destroy all copies of
the original message.
To
join or leave this LISTSERV list, please visit the list's web interface at:
http://listserv.muohio.edu/archives/ateg.html and select "Join or leave
the list"
Visit ATEG's web site at http://ateg.org/