Open-Ended Elaborations in Creative Metaphor - Computational Creativity Research: Towards Creative Machines - page 229

In ATT-Meta, a transfer of information from pretence into surrounding space is
just a reasoning step like any other, and an individual correspondence created by a
correspondence rule is just a proposition like any other. (Recall that a correspondence
rule such as (8) creates an individual correspondence when it fires, such as: Mary's
having a motive to believe that Mike is adorable corresponds to at least one subper-
son having a motive to believe this.) Further, correspondence rules just sit with other
rules in the system, firing when able to. The overall reasoning engine that handles the
firing of rules makes no distinction between correspondence rules and other rules. In
addition, while reasoning actions occur in different spaces, there is no regimentation
of when reasoning steps happen in different spaces, except as is dictated by the nat-
ural linkages between the outputs of some rules and inputs of others. The actions in
different spaces are interleaved in whatever way is convenient for the goals for rea-
soning. Thus, as already hinted, the peculiarly metaphorical aspects of ATT-Meta's
reasoning (reasoning within pretences; and the action of correspondences to trans-
fer information) are seamlessly woven into the overall reasoning endeavour. This
gives tremendous flexibility especially when dealing with more elaborate cases of
metaphor, and especially when a metaphorical scenario is developed over a multi-
sentence or larger region of discourse. It is typical in such cases for reasoning to be
needed in the ordinary terms of the target subject matter, not just metaphor-based
reasoning, partly because metaphorical stretches can be separated by literal language.
In the ATT-Meta approach (and implemented system) metaphorical correspon-
dences cause information to be transferred in both directions between a pretence
and the surrounding space. I call the surround-into-pretence direction the “reverse”
direction. The motives for this are detailed in Barnden et al. [ 9 ]. In brief, one reason
is to try to maximize the degree of consistency between what is being inferred within
the pretended scenario and what is being inferred within the surrounding space's
scenario, and another is to allow useful enrichment of the pretended scenario based
on information independently known about the surround. For example, in a Mind
as Having Parts that are Persons example, if we know that the real person believes
X then it may be useful to infer that each subperson believes X too, by virtue of
correspondence rule (9). This could allow important inference about different sub-
persons' thoughts to take place. This case of reverse transfer is a simple instance
of the “metaphorization” of information in the surrounding space. A more elaborate
case, concerning a creative metaphor of doubts and grievances moving in and out of a
conversation as though they were independent-minded cats, is discussed in Barnden
et al. [ 6 , 9 ].
At the time of writing, ATT-Meta is being developed towards metaphor genera-
tion as well as understanding. The reverse transfer capability puts ATT-Meta in an
advantageous position in this respect.