Mystery Number Three - Computational Linguistics and Talking Robots - page 64

token lines. It is just that the proplets stored in a Word Bank support several
different views , among them those of diagrams 4.3.2 and 4.5.3.
Different views are created by referring to different kinds of proplet values:
proplets with the same prn value are viewed as propositions, proplets with
non-NIL sur values as language content, proplets with NIL sur values as con-
text content, proplets with interlocking pc and nc values as coordinations,
proplet sets with successive prn values as text, etc. (1.3.2, 6.6.1-6.6.5).
Therefore the technically correct way of integrating diagram 4.3.2 into dia-
gram 4.5.3 is by changing from a naive-conceptual view to a database view:
instead of viewing content proplets as sets with a common prn value (propo-
sitions), separated into a language and a context level (4.3.2), and so on, 13 the
same proplets are viewed solely as items to be sorted into token lines according
to their core values and in the order of their arrival. 14
This conclusion is formulated as the Third Mechanism of Communication:
4.5.5 T HE T HIRD M ECHANISM OF C OMMUNICATION (M O C-3)
The operations of cognition in general and of natural language com-
munication in particular require a memory with a storage and retrieval
mechanism supporting (i) extensive data coverage, (ii) functional com-
pleteness, and (iii) efficiency which enables real-time performance.
In DBS, extensive (i) data coverage is achieved by allowing an unlimited num-
ber of proplet values. These are used to define an unlimited number of views
which are the basis of (ii) functional completeness. Performance in (iii) real-
time is based on combining a simple content-addressable database schema
(Word Bank), a simple data structure (proplets), and a time-linear algorithm
running in linear time (C1-LAG).
The need for completeness of data coverage is illustrated by automatic word
form recognition (Sect. 2.5). It may be systematically improved by parsing
corpora to find missing entries to be added to the online lexicon, and by
fine-tuning the rules for inflection/agglutination, derivation, and compound-
ing. Other areas requiring completeness of data coverage are the syntactic-
semantic interpretation and the semantic-syntactic production of different lan-
guage constructions, as well as nonlanguage and language inferencing.
Functional completeness has been illustrated by the operations of (a) stor-
ing content derived in the hear mode (3.3.1, 4.3.1), (b) selectively activating
content in the think mode (3.3.2), (c) realizing natural language surfaces from
13 In DBS, these different views are established by patterns, and not by create view ,asinSQL.
14 See 6.4.3 for an example showing reference as a “horizontal” relation within a token line.