|
Research >> The ILIT R&D Methodology
R&D can be method-oriented or task-oriented.
Many R&D groups in academia tend to concentrate on particular
methods and techniques – algorithms, logics, formal languages,
etc. – and then seek applications for evaluating and testing
these methods. Task-oriented research starts with a real-world
problem and then devise methods to best tackle it. (This distinction
is, to a degree, an abstraction; in reality, many more factors
play a role in the choice of research paradigm.) Both basic and
applied research at ILIT is guided by our belief that the real
breakthrough in language technologies will take place only if
the computer systems will gain the capability to understand
texts, not just to manipulate and compare uninterpreted
or underinterpreted symbol strings on the basis of their occurrence
in texts.
To attain this capability, computer systems
must become self-aware and aware of the world that surrounds
them. They must internalize descriptions of kinds of events and
objects (both physical and mental) that exist in the world; they
must be able to model their own goals and plans; they must also
be aware of other intelligent agents and their goals, plans and
knowledge. Finally, in order to model intelligent communication
with humans, the systems must internalize the knowledge about
one or more human languages, including the connections between
elements of language and elements of their internal world model.
In view of the above, the main line of
ILIT’s basic research is devoted to the many and various
facets of extracting and manipulating meaning in natural languages.
Since our goal is to build comprehensive working systems (rather
than develop limited-purview theories), we cannot work on semantics
in isolation; determination and manipulation of meaning relies
on computational descriptions of several non-semantic strata
of language: ecology, lexis, morphology and syntax – in
addition to semantics proper. Our approach is strictly computational – we
are developing algorithms and building systems for text analysis
that result in meaning specification as well as for text synthesis
on the basis of its meaning. Our approach integrates rule-based
and corpus-based methods, in that the conditions in our processing
rules may include both linguistically and ontologically motivated
constraints and results of frequency and co-occurrence analysis
of text elements in corpora.
Our basic scientific paradigm, incorporating
all of the above facets, is embodied in the theory of ontological
semantics. This theory is realized as a society of “microtheories” of
particular language and world phenomena. This arrangement helps
to organize the vast descriptive and computational enterprise.
It also makes it possible to “import” descriptions
available for particular phenomena and then adapt them for our
needs. At the center of our current basic research are a number
of lexical-semantic microtheories as well as the study of reference.
Adding treatment of meaning in various
languages to the battery of existing natural language processing
technologies is the mainstay of ILIT’s applications research.
Development of tools and knowledge resources
to support our basic and applied research has been an important
component of our work.
|
