Mystery Number Four - Computational Linguistics and Talking Robots - page 81

5.4.3 C OMPLETING AN INTERSECTION BY SPREADING ACTIVATION
⎡

⎡
⎣

⎤
⎦

verb: look_for
arg: John, potato
pc: cook 19
nc: eat 21
prn: 20

⎡
⎣

⎤
⎦

⎤ ⎡ ⎤

noun: potato
fnc: look_for
mdr: hot
prn: 20

noun: John
fnc: look_for
prn: 20

adj: hot
mdd: potato
prn: 20

⎣ ⎦ ⎣ ⎦

This primary subactivation is based on the addresses potato 20 in the hot pro-
plet, look_for 20 in the potato proplet, and John 20 in the look_for proplet.
While a primary subactivation utilizes the intrapropositional relations of
functor-argument and coordination (cf. NLC'06, Chaps. 6 and 8), a secondary
subactivation is based on the corresponding extrapropositional relations (cf.
NLC'06, Chaps. 7 and 9). For example, using the pc (previous conjunct) and
nc (next conjunct) values of the look_for proplet in 5.4.3, a secondary subacti-
vation may spread from John looked for a hot potato to the predecessor and
successor propositions with the verb values cook and eat and the prn values
19 and 21 , respectively. 12
The degree of subactivation that is automatically selected at any current
moment, including no subactivation at all, depends on the computational re-
sources available. These depend on the agent's interests and current preoccu-
pation. Compare, for example, running for one's life and a leisurely walk, both
in the same park: the very same triggers will be completely ignored in the first
case, but may result in rich subactivations in the second.

5.5 Analogical Models for Problem Solving

Sect. 5.2 showed a data-driven 13 application of the inference chain 5.2.1
(forward chaining). Let us consider now a goal-driven 14 application of this
same inference chain (backward chaining). 15 While a data-driven application
is for instinctive, habitual, and rote behavior, a goal-driven application is for
nonroutine behavior such as analogical reasoning and problem solving. The
backward-chaining of a goal-driven inference application makes heavy use of
subactivation and intersection.
12 Thus, the movie title All the President's Men (Pakula 1976) will likely activate Couldn't put
Humpty Dumpty Together Again as a continuation, referring to R. M. Nixon. In fiction, our notion
of triggering a spreading subactivation is illustrated by the madeleine experience of Proust (1913),
which brings back an almost forgotten area of what he calls “l' édifice immense du souvenir. ”
13 So-called because the inference chain is triggered by data matching the initial antecedent.
14 So-called because the inference chain is triggered by a goal matching a consequent. A goal-driven
application is a form of case-based reasoning (Schank 1982). See also Muñoz-Avila et al. (2010).
15 We follow the standard terminology in computer science, even though from the viewpoint of DBS
the input to goal-driven inferencing, i.e., the goal, is data too - no different from that triggering
data-driven inferencing.