|
|
Questions and Summary of Fauconnier and Turner (Conceptual Integration Networks)
discussed by Ilya Farber
Fauconnier & Turner (hereafter F&T) discuss "blending", a
cognitive operation whereby elements of two or more "mental spaces" are
integrated via projection into a new, blended space which has its own, unique
structure. They present examples of blending, analyze the blending process,
provide a taxonomy of blends, and argue for the ubiquity and importance of
blending as a cognitive resource. In this summary I'll focus on the basic
structure of blends and blending, since the taxonomy sections are sufficiently
technical and detail-dependent that a brief summary would not be very useful.
The first illustration goes a long way toward clarifying the concept, so
I'll quote it here and then relate it to the subsequent analysis:
"The riddle of the Buddhist monk: A Buddhist monk begins at dawn one day
walking up a mountain, reaches the top at sunset, meditates at the top for
several days until one dawn when he begins to walk back to the foot of the
mountain, which he reaches at sunset. Making no assumptions about his starting
or stopping or about his pace during the trips, prove that there is a place on
the path which he occupies at the same hour of the day on the two separate
journeys."
[stop and try to figure it out first ...]
The solution: "... imagine the Buddhist monk walking both up and down the
path on the same day. Then there must be a place where he meets himself, and
that place is clearly the one he would occupy at the same time of day on the
two separate journeys."
In this example, there are two INPUT SPACES, one which has a monk
traveling up the mountain and one which has him traveling down. There is a
MAPPING between the two, which links up the mountains, monks and paths (all
identical), days and motions (nonidentical but superimposed). Out of this
mapping is constructed the BLEND, which is another mental space.
Most of the central claims of the paper can be illustrated using this
case:
SELECTIVE PROJECTION AND FUSION: Only certain features from each input
space are projected into the blend; for example, absolute date has no place in
the blend. Some features may be projected from one input but not the other(s).
Of those that are projected from multiple inputs, some may be fused into single
elements in the blend (eg the mountain), while others may remain separate (eg
the two motions). These decisions will be based on (1) the purpose of the blend
and (2) a set of optimality constraints, discussed below.
EMERGENT STRUCTURE: To be useful, the blend must have structure above and
beyond what's present in the inputs. In the monk case, there are now multiple
interacting figures and motions, which allow us to bring new resources to bear
on the problem. We may also wind up adding futher structure to the blend space,
either because it makes internal sense or because the blend fits some other
pattern that we know something about.
OPPORTUNISM AND ENTRENCHMENT: As in the monk case, blends may be
constructed on-the-fly to exploit useful relations between input domains that
are under consideration. In other cases (such as the notion of "digging your
own grave"), the blend may become sufficiently entrenched that it can serve as
a conversational and/or cognitive shortcut.
The construction of blends is guided by six optimality principles. I'll
present all six, since (in my opinion) they represent F&T's strongest
claims about actual human psychology:
INTEGRATION: the blend should be tightly integrated. F&T say this
means it "can be manipulated as a unit"; though it's not clear what this
depends on, in the analysis of the examples it usually means that all the
elements are strongly linked by some sort of functional or metonymic
association. It also seems to help if the whole thing can be imagined as a
single picture or scene (this may even be a necessary requirement, judging by
the examples).
TOPOLOGY: the blend should preserve the character of the relations between
elements in the input spaces. Obviously, there's a tension here, since the
elements may have different relations in the two spaces; how this tension is
resolved determines the blend's place in F&T's taxonomy, which I'm not
dealing with here.
WEB: the blend<->input space connections should be robust across
manipulations of the blend, so that results achieved in one space can exert
influence on the others.
UNPACKING: you should be able to pull apart the blend and get the inputs,
mapping, and other things that went into it. F&T don't spend much time on
this, and it was never clear to me why this was necessary (or at least, why it
had to be possible based on the blend alone, which is what they claim).
GOOD REASON: there should be a justification for everything that shows up
in the blend, with respect to that particular blend's intended function.
Basically, this is the parsimony constraint; it prevents unnecessary details
from getting projected into the blend.
METONYMY PROJECTION: _if_ the reason that something is projected from an
input is that it has a metonymic relation to (ie is associated with) another
element that's being projected, that relation should be even tighter in the
blend.
Comments and Questions:
My primary worries all had to do with the status that F&T are claiming
for the blending mechanism (can you tell I'm a philosopher?). They prove pretty
much beyond a doubt that blending is a coherent and frequently-used move in the
realm of cognitive and communicative operations, "on a par with analogy,
recursion, mental modeling, conceptual categorization, and framing." What's not
clear to me, though, is whether blending is the *result* of some deeper
cognitive process, or whether the presented structure of blending is supposed
to represent the actual structure of cognition. F&T seem to want it to be
the latter, but that claim would open up a number of questions that they don't
address, such as:
-- What are the alternative hypotheses?
-- How do we decide when to blend, what spaces to blend, what elements from
those spaces to project, and in what ways?
-- How do we understand the structure of others' blends?
-- Does blending depend on the manipulation of images or language? Do they
depend on it?
-- How, if at all, does any of this relate to the brain?
Put more generally, my worry is that in hybridizing the argument styles of
linguistics, cognitive psychology and the various modeling sciences, F&T
(and many other eminent cognitive scientists) risk losing track of two things:
(1) the status that they're claiming for the entities in question -- are they
"real" mechanisms, or abstractions, or models, or convenient fictions, or ...?
(2) the relation between their theory and the evidence -- are they saying that
their theory is the *best* explanation, or that it's *an* explanation, or are
they taking the engineering approach and saying "this is what we've got, let's
see how we can rig it up to cover as much territory as possible"? All three are
useful, but I think one of the dangers of cogsci is that we can blend them in
ways that make it unclear what exactly we're trying to do.
A few more specific things I found myself wondering about: -- Are there
systematic differences between blends that we use for internal cognitive
purposes and ones that we come up with specifically for communication?
-- How much linguistic and/or cultural variation is there in the use of
blending, and in the importance assigned to the various optimality constraints?
-- How important is imagery in the construction of blends? Some blends seem
easier to communicate verbally, and some seem easier to draw (perhaps with
labels) -- is there anything systematic about this difference?
--Ilya
ifarber@ucsd.edu
|
