Mystery Number Five - Computational Linguistics and Talking Robots - page 109

The inverse of downward traversal is the upward traversal of a semantic hi-
erarchy. An upward inference assigns a hypernym like food to concepts like
salad or steak . Consider the following application of the inference for upward
traversal derived in 6.5.7 ( consequent 2 ).

6.5.12 H IERARCHY - INFERENCE FOR UPWARD TRAVERSAL

antecedent consequent

⎡ ⎤ ⎡ ⎤

noun: α noun: food

fnc: ( β K)
prn: K+M

⎣ ⎦ ⎣ ⎦

α {apple, pear, salad, steak} fnc: β

prn: K

rule level & up

matching and binding

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

noun: Julia
fnc: prepare
prn: 23

verb: prepare
arg: Julia salad
prn: 23

noun: salad
fnc: prepare
prn: 23

noun: food
fnc: (prepare 23)
prn: 29

content level

As in the downward inference 6.5.9, the antecedent of the upward inference
consists of a single pattern proplet, now with the restricted variable as the

core value. Due to the use of a pointer address as the fnc value of the out-
put there is sufficient information to complete the output proplet into the new
proposition Julia prepares food (not shown), with the prn value 29 and the
pointer proplets (Julia 23) and (prepare 23) .
The automatic derivation and restriction of schemata like 6.5.4 and 6.5.5 di-
rectly controls the automatic adaptation of the hierarchy inferences by adjust-
ing their restriction sets. In this way, DBS fulfills the three functions which
define an autonomic system: “automatically configure itself in an environ-
ment, optimize its performance using the environment and mechanisms for
performance, and continually adapt to improve performance and heal itself in
a changing environment” (Naphade and Smith 2009).

α

6.6 Natural vs. Artificial Language Learning

The mechanism of automatically deriving and adjusting DBS schemata holds
at a level of abstraction which applies to natural and artificial agents alike.
The simplicity of this mechanism allows us to design artificial agents as nat-
ural agents in that they adjust over time. Thereby, the following differences
between natural and artificial agents do not stand in the way:
In natural agents, adjusting to a changing environment and optimizing come
in two varieties, (i) the biological adaptation of a species in which physi-
cal abilities (hardware) and cognition (software) are co-evolved, and (ii) the
learning of individuals, which is mostly limited to cognition (software) alone.