Conclusion - Computational Linguistics and Talking Robots - page 248

This work will not go unrewarded from the application side either. The obvi-
ous application is query answering in natural language. Thereby the LA-speak
part is only the tip of the iceberg. Prior to answering, the query is converted
automatically into several schemata which are used to subactivate correspond-
ing contents. These data are processed into the content for the query answer
by means of intersection and inferencing. Once the resulting answer content
has been derived, it is passed to LA-speak for realization as an unanalyzed
external surface.
While specific applications may benefit selectively from the nurturing of a
particular component, all applications will benefit simultaneously from a me-
thodical upscaling of the DBS robot as a whole. An application which does
not require certain abilities may be run with a DBS version in which they have
been switched off. 33
The systematic, theory-driven upscaling of a talking robot is of general inter-
est for the following reasons. First, it provides the opportunity to ensure com-
patibility between the system's components in a declarative manner. 34 Sec-
ond, the neighboring sciences, for example, psychology, ethology, neurology,
philosophy, etc., may use the computational model to test some of their own
issues, which may in turn contribute to the long-term effort of upscaling the
talking robot. 35 Third, by making regular version updates available to the pub-
lic, progress in pure research may quasi automatically improve the language
processing of participating applications.
The orderly transfer from a continuously improving DBS system to com-
mercial applications of human-machine communication may be illustrated by
the following vision. Every year, when the current monitor corpus (Sect. 12.2)
has been put through the automatic software grinder of word form recognition,
parsing, frequency analysis, and comparison with preceding monitor corpora,
the results are used for a software version with improved data coverage.
By making new versions available to paying subscribers for their natural lan-
guage processing needs, all or most of the research costs may be recovered.
For this to work long-term, a new release must not require any labor from the
subscriber (e.g., additional personnel training), except for the routine installa-
tion. Also, each new version must enhance service directly and noticeably, so
that subscribers are attracted and kept in sufficient numbers.
33 For example, a dialogue system over the phone may omit the ability to read.
34 In addition, typing (in the sense of computer science) could be used in DBS. It is not really needed,
however, because of the simplicity and formal uniformity of proplets and the associated interfaces
and algorithm.
35 For example, a system of control implemented in analogy to the sympaticus/parasympaticus nerve
network.