CREATIVITY IN VERBALIZATION AS 
EVIDENCE FOR ANALOGIC KNOWLEDGE 
Wallace L. Chafe 
Department of Linguistics 
University of California 
Berkeley CA 90406 
An issue which I take to be of great 
interest at the present time is whether 
human knowledge can be adequately 
represented in terms of a network of 
propositions. The alternative is that 
knowledge has a significant analog 
component, which I understand to mean that 
continuous differences in the real world are 
stored in a continuous mental 
representation. Three views are possible. 
A strict propositionalist would say that a 
propositional network is adequate to account 
for everything in memory. A strict 
analogist would claim that everything is 
stored analogically. But I suppose that 
many people are willing to admit some 
combination of these two views, and allow 
that the mind has the capacity for both 
kinds of knowledge. 
Evidence for analogic storage is in 
part introspective. Many people seem to be 
able to observe themselves using mental 
imagery for thinking about the world, and 
even for solving logical problems. Although 
it is easy to demonstrate that mental images 
are not "pictures" in any literal sense, 
nevertheless they do seem to retain some of 
the analogic quality of reality. Those 
whose thinking has a strong imaglnal 
component are predisposed to believe that 
both language and propositions are 
derivative, in the sense that they are ways 
of talking about or describing something 
more basic which shows up in consciousness 
in the form of imagery. But there is a 
counterargument which says that the basic 
processes of the mind take place outside of 
consciousness, and that we deceive ourselves 
if we think we can observe cognitive 
processes directly. Furthermore, it has 
been noted that imagery does not allow us to 
individuate objects and events that were not 
individuated when first perceived. The 
familiar demonstration of the power of 
imagery in which you are asked how many 
windows are in your living room or the like, 
and then you seem to go through a process of 
imaging the room and counting the windows, 
can in fact be used to show how much 
preprocessing has taken place. For one is 
evidently able to count in one's mind only 
those objects which have already been 
attended to and individuated, just as one 
might count the nodes in a propositional 
network. Images don't seem to allow for 
further individuation, but are dependent on 
whatever individuating took place at the 
time of perception. 
But what needs to be done is to sort 
out those interpretive processes which must 
take place during perception, if they are to 
take place at all, and those which can still 
be performed on material stored in memory. 
If there are processes of the latter kind, 
and if they depend on analogic properties of 
144 
the stored material, then the analogic view 
will have strong support. Roger Shepard and 
his coworkers, for example, seem to have 
demonstrated that such processes do exist in 
people's ability to continuously rotate 
mental objects. I want to suggest that the 
way people talk about things provides 
another demonstration of this kind. 
In my earlier paper I suggested that 
verbalization -- the set of processes by 
which a speaker turns stored experience into 
words -- requires that choices be made at 
various points. These choices, I suggested, 
are of three kinds. I called them 
schematizing, framing, and categorizing. 
What is of interest here is that all three 
of these processes are interpretive in just 
the sense we are looking for. All of them 
require the speaker to interpret stored 
material, related to particular events and 
objects, as instances of stored prototypes. 
All of these processes may take place at the 
time of perception, but they do not have to. 
In fact, it is clear both from hesitations 
in speech and from the frequency with which 
people use different verbalizations for the 
same material at different times that many 
of these choices are made only while the 
material is being verbalized. 
Categorization is the clearest example, 
if only because the data are more extensive. 
It has been known for a long time that items 
stored in memory differ in their degree of 
"codability": the degree to which they can 
readily be categorized. Highly codable 
items tend to be named with short words, to 
be named with single words rather than 
phrases, to be named without hesitation, to 
be given the same names by different people, 
and to be named the same way by the same 
person on different occasions. Items of low 
codability show the opposite symptoms. It 
is these latter items that are of interest 
here. 
Imagine a person who has observed a 
scene and individuated events and objects 
within it. Certain of these indlviduated 
events and objects are likely to be highly 
codable, and undoubtedly the observer will 
have categorized them immediately as part of 
his perception of the scene. There might, 
for example, be a typical table which he 
immediately categorizes as a table. On the 
other hand, there might also be objects and 
events which he has individuated but not 
categorized, or categorized only in a highly 
general way (as a piece of furniture, for 
example). He will have stored his knowledge 
of this object with enough analogic 
information included to enable him to search 
for an appropriate categorization at a later 
time when he refers to that knowledge in 
speaking. At that time he will arrive at a 
point in the verbalization where he realizes 
the necessity of expressing his knowledge of 
the object in words. Typically at that 
point he will hesitate so that he can 
compare his mental representation of the 
object with his mental representation of 
available category prototypes, will decide 
on the best match, will perhaps add 
modifiers to improve the match, and will 
finally utter something llke "a 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
! 
! 
! 
! 
! 
! 
lozenge-shaped green marble coffee table". 
This choice, made in the course of speaking, 
may or may not be remembered on a subsequent 
occasion when he has to verbalize the same 
knowledge again. At that time he may have 
to think about how to categorize it all over 
again, and may very well choose differently. 
It is important that categorization is 
often a more-or-less affair, not a matter of 
yes-or-no. The choice seems to be made on 
the basis of comparing the "experient.ial 
content" associated with the individuated 
item with the experiential content 
associated with the category. The match 
between the two takes place on a continuous 
scale. If both individual items and 
categories have associated with them a 
content which allows such continuous 
comparisons to be made, the best conclusion 
seems to be that the content itself has 
analogic properties. 
I will just mention here that the two 
other kinds of choices which may be made 
during verbalization -- the choices of what 
I have called frames and schemata -- seem to 
lead to the same conclusion. While, for 
example, a particular chunk of experience 
may be interpreted as an instance of a 
specific schema while it is being 
experienced, the schematization process is 
something that can be applied while one is 
talking. I take this to be evidence that 
people must be able to store enough analogic 
information about an experience that they 
are able to interpret it later as an 
instance of some schema. The same can be 
said for framing. A particular event might 
be framed on one occasion as "Jim finally 
planted the seeds" and on a later occasion 
as "Our garden finally got started", where 
the objects and their roles are quite 
different. The speaker's memory for this 
event was flexible enough to allow for these 
two interpretations. 
In short, although some interpretive 
processes such as individuation seem to be 
confined to the perceptual stage of 
information processing, others such as 
categorization, framing, and schematization 
are applied for the first time, in some 
cases at least, while a person is 
verbalizing knowledge recalled from memory. 
In order for this to take place it would 
seem that some nonpropositional 
representation of knowledge must exist. 
145 
