COGNITIVE GRAMMAR 213
words in some others, or a combination of both that in turn produce a mental
model through conceptual blending. The mental model activates that part of the
network where sentence patterns are stored. The structure of the propositions in-
herent in the mental model specifies a range of possible sentence patterns. One is
selected as a “best fit” and is then filled with words that match the model and the
person’s intention. Cognitive grammar accounts for the high degree of creativity
in language on the basis of the essentially limitless supply of mental propositions
and the flexibility inherent in English word order. Language’s creative character-
istics are not the result of a generative grammar.
This model of production allows cognitive grammar to offer a viable expla-
nation of errors in language without recourse to rules or competence and per-
formance. It is often the case that, when speaking, we intend to say one thing
and end up saying something different. We usually catch these “slips of the
tongue” and self-correct, but the question remains: What caused the error?
Consider the following example: The family and I are going to drive to the
beach, and before we leave I want my son to bring in the dog and put out the cat.
But what I actually utter is “Bring in the cat and put out the dog.” Why did my
intention fail to produce the desired sentence?
Our experiences of the world are defined and processed as patterns. Mam-
mals have four limbs, people laugh when they are happy, birds fly, dogs bark,
the sun rises in the morning and sets in the evening. Many patterns necessarily
overlap because they have similar characteristics. Numerous people, for exam-
ple, have to remind themselves that tomatoes are a fruit, not a vegetable, and
that dolphins are mammals, not fish. Language acquisition at the word level in-
volves recurrent encounters with, say, dogs and cats, resulting in mental mod-
els of “dog-ness” and “cat-ness.” Hearing the word or deciding to utter it
triggers an association between one set of neural patterns and another set that
contains subsets of the various features related to the target. Each triggering in-
creases the strength of the connection between the appropriate patterns, raising
the probability of correctly matching strings of phonemes.
In the case of dogs and cats, we can imagine several subsets, clustered, per-
haps, under the general set of pets or mammals, depending on how one primarily
categorizes these animals. The subsets will contain not only the features of dogs
and cats—hairy, lovable, licks, ownership, and so forth—but they also will con-
tain entries for other animals, such as mice, guinea pigs, turtles, and skunks. In
my scenario, when I formulated the intention to tell my son to bring in the dog
and put out the cat, the entire network associated with pets/mammals was acti-
vated. Because the individual representations of dogs, cats, skunks, and guinea
pigs have numerous overlapping features and because they are all intercon-
nected, they will compete as targets (see Rumelhart & McClelland, 1986). This