Mystery Number Three - Computational Linguistics and Talking Robots - page 66

these lexical proplets incrementally via copying (7) in a strictly time-linear
derivation order.
Next consider the language production from stored content 3.3.3 within the
component structure of diagram 4.5.3:

4.6.2 M APPING STORED CONTENT INTO SURFACES ( SPEAK MODE )

noun: α verb: β noun: γ

pattern proplets

rule level: fnc:

prn:

β arg:

prn:

α γ fnc:

prn:

β

K K K

7

matching and binding

noun: John
fnc: know
prn: 625

content level:

8

noun: Julia
fnc: know
prn: 625

content proplets

verb: know
arg: Julia John
prn: 625

6

synthesis

surfaces: Julia knows John

The impulse (8) may be provided by another agent's question (Sect. 4.2) or
by a request to recount a certain event. 15 The navigation through the content is
driven by the rule level and transmitted to the content level by means of pattern
matching (7). 16 Once the LA-think grammar has been started by an initial
impulse matching its start state, the derivation is powered by successful rule
applications calling successor rules with their rule package. The blueprints for
well-formed language expressions passed to the I/O component (6) are derived
in frequent (7, 8) interactions between the rule and the memory component.
Finally consider language production based on the on-the-fly inferencing of
the agent's autonomous control. Anticipating the discussion of inferences in
Chaps. 5, 6, 10, and 11, it is sufficient for present purposes that the output
of a DBS inference be a sequence of proplets which (i) have core and prn
values and (ii) code the semantic relations of functor-argument, coordination,
and coreference.
In other words, the content produced by on-the-fly inferencing has the same
format as stored content resulting from language recognition - and can there-

15 For a general discussion of statement, question, and request dialogues, see Chaps. 10 and 11.
16 For simplicity, the navigation is more direct than in 3.3.3.