(Copyright 2009 David Dodds)
Semantic information located in ontologies can be
wrapped in (English), or other) natural language
statements by building an instance graph network
of the instances in the system's ontologies where
items match or depict the semantic processing
undertaken by the system. As we have seen in
past articles elements / items in an SVG graphics
file may have ontological items associated with
them. Also, MathML items may have ontological
items associated with them. One way to indicate
(for a given moment) which ontological items are
relevant (in depicting the semantics of the data set
under examination, such as the SVG bargraph or
an MathML function) is to display the ontological
terms / items as their RDF URIs. This might be
done with the aid of a network graph. While
accurate such a depiction is ugly and unhelpful to
the more casual user, one who is not versed in
W3C RDF experience. This means most people.

A more useful way to depict the semantic content
of material, such as the bargraph or a MathML
equation is to depict the information in a way
which is like the communications used every
day by non-experts (in the semantic web domain)
and this means natural language constructions.
It is implied, therefore, that when possible,
special usage symbols and terms are replaced
by verbiage and that of more likely common usage.

This leads us to the point that we are then faced
with explaining to the computer (via programming)
how to decide what to say. Some will view this as
the inverse problem faced by programmers
constructing natural language understanding systems,
a far greater complexity than building a fancy
parser for grammatical categorizing (parsing).

One of the ways that we can approach the
programming for making natural language-like
output of semantic content is to restrict the
complexity of the modes of expression (fluidity)
to be less sophisticated than a human natural
language user would produce if he were outputting
similar semantic information using natural language,
such as English.
In the next part of this article we see how
language semantic templates are used to perform
this processing. We also see how the TaleSpin
process contributes to this activity.
(Discussed are : How are Conceptual
Dependency Theory , and Frames, of value to
this pursuit? How do we decide what to say next?)