Conclusion - Computational Linguistics and Talking Robots - page 244

sign-oriented [-sense, -constructive] ontology to an agent-oriented [+sense,
+constructive] ontology (cf. FoCL'99, Sect. 20.4). 24
12.5 Autonomous Control
An agent-oriented approach requires an autonomous control for sensible be-
havior, including natural language production. In DBS, autonomous control is
driven by the principle of balance. A state of balance is an absolute like truth,
and like truth it provides the fix point necessary for a system of semantic in-
terpretation. But while truth is (at least 25 ) bipolar, balance is monopolar. This
may be the reason why balance, unlike truth, is a dynamic principle, suitable
for driving the cognitive agent trying to survive, and to survive comfortably,
in a constantly changing world, short-, mid-, and long-term.
In DBS, the agent's search for balance is implemented as a set of inferences.
An inference is triggered by an activated content matching the antecedent.
After binding the variables of the antecedent to constants in the input, e.g.,
core values, the consequent of the inference derives a blueprint for action as
output. The output is written to the now front of the agent's Word Bank.
This new approach to inferencing is based on the belated insight that the
database schema of a Word Bank, designed before FoCL'99, is in fact a
content-addressable memory. 26 Content-addressable memories happen to be
the most efficient for content written once and never changed. By structuring
the memory like sediment , written once and never changed, all processing for
real-time behavior control is restricted to the now front .
Stored content which never changes makes practically 27 no processing de-
mands on the system. At the same time, the personal history contained in an
agent's static sediment provides an individual notion of what is true and what
is right. The agent's current state is defined by the most recent data, stored last
(rightmost) in the token lines.
The agent's overall moment to moment behavior may be viewed as regulated
by of a basket of weights which represent the current options for maintaining
24 The binary feature notation without attributes used here, e.g., [+constructive] , resembles the
“feature bundles” of Chomsky and Halle (1968).
25 Cf. FoCL, Sect. 20.5.
26 Originally, the Word Bank had been naively derived from classic (i.e., record-based) network
databases as described in Elmasri and Navathe (1989). Driven by functional concerns, our interest
was focused primarily on a running program, working as intended. Over the years, the implementa-
tion of a Word Bank has been explored in a sequence of at least four projects at the CLUE.
27 With the exception of occasional cleanups, cf. Sect. 5.6.