Mystery Number Four - Computational Linguistics and Talking Robots - page 73

and the prn value of the consequent, e.g., K+1 . Within a chain, any reoccur-
rence of a variable must be bound to a value at the Word Bank level equivalent
to that of the initial occurrence of the variable. The equivalent value is either
an address (5.2.3) or an indexical (5.2.5).
Step 1 is an R inference (defined in 5.2.3) with the connective cm (for coun-
termeasure) and triggered by a sensation of hunger. Step 2 is a D inference
with the connective pre (for precondition), while step 3 is a D inference for
downward traversal (defined in 6.4.1) with the connective down . Steps 4, 5,
and 6 are E inferences with the connective exec (for execute).
Step 4 may be tried iteratively for the instantiations of food provided by the
consequent of step 3 (see the restriction on the variable ). If the agent cannot

locate an apple, for example, it tries next to locate a pear, and so on. Individual
food preferences of the agent may be expressed by the order of the elements
in the variable restriction.
Step 7 is based on a D inference for upward traversal, defined in 6.4.4,
with the connective up . This step is called the completor of the chain be-
cause the consequent of the chain-final inference, called the completor conse-
quent , equals the consequent of step 1. The completor indicates the successful
blueprint of a countermeasure to the imbalance characterized by the antecedent
of the chain-initial reactor inference.
While R inferences are activated by triggers provided by the agent's recog-
nition, external (e.g., hot ) or internal (e.g., hungry ), D and E inferences are
usually initiated by other inferences which are already active. D(eductor) infer-
ences establish meaning relations (Sect. 5.3), and are illustrated by synonymy
(5.3.1), antonymy (5.3.2), cause and effect (5.3.3), summarizing (5.3.5), down-
ward traversal (6.5.9), and upward traversal (6.5.12). One D inference may
activate another D inference or an E inference.
E(ffector) inferences provide blueprints for the agent's action components. 7
Because E inferences connect central cognition with peripheral cognition,
their definition has to be hand in glove with the robotic hardware they are
intended to use.
A limiting case of a chain is a single R/E inference, such as the following:

α

5.2.2 O NE - STEP CHAIN BASED ON AN R/E INFERENCE

R/E: α feel full K cm/exec α stop eating K+1
Here the response to a deviation from balance results in a countermeasure
which can be executed directly (i.e., without intervening D inferences).