Conclusion - Computational Linguistics and Talking Robots - page 246

semantic relations of coordination and functor-argument, defined at (a) the
content and (b) the schema (rule) levels. Introduced to model the cycle of
natural language communication, the semantic relations provide the structural
basis also for subactivation, intersection, and inferencing.
Because the content in a Word Bank is structured by the semantic relations
of natural language, the system can respond in kind, i.e., it can be as specific or
general as formulated in the query or any other search request. This ability to
respond to language questions with language answers, developed for natural
language dialogue with a talking robot, may also be used for more conven-
tional applications. For example, a database structured as a Word Bank, sitting
on a standard computer in some geographically remote warehouse, may be
queried, and may answer, in natural language.
12.6 Applications
The DBS approach to practical (commercial) applications of natural language
processing is based on solving the most important theoretical question first:
How does the mechanism of natural language communication work?
To protect against accidentally neglecting some crucial interface, compo-
nent, or ability, the overall design of a DBS robot aims at functional complete-
ness. By modeling all essential structural aspects of natural language com-
munication by humans it is hoped that there will be no application-motivated
requests which cannot be satisfied.
If a functional framework works properly at all levels of abstraction, though
with small (and highly relevant) data coverage only, then all that remains to
be done is to increase the data coverage. For natural language communication,
this is a mammoth project, though nothing compared to projects in physics
(CERN) or biology (human genome project), for example.
Extending the data coverage as a form of upscaling has immediate conse-
quences on commercial applications using the system for their natural lan-
guage processing needs. Take for example LA-morph, the automatic word
form recognition software, running with a certain natural language of choice.
The data coverage of such an instance of LA-morph may be extended by
adding to the lexicon and by optimizing the allo- and combi-rules for the nat-
ural language at hand. This broadens the base for syntactic-semantic analysis
and inferencing. It also provides practical applications with better results for
retrieval based on content words.
A second area for completing data coverage is extending the syntactic-
semantic analysis. When applied to a new (i.e., previously unanalyzed) natural