Task-Oriented Semantic Interpretation
                  John F. Sowa and Arun K. Majumdar
                             VivoMind LLC

Text understanding requires knowledge of the language and knowledge of
the subject matter. During the past 40 years, many ways of using both
kinds of knowledge have been implemented for natural language
processing. This talk describes a method called task-oriented semantic
interpretation (TOSI), which combines a broad, but shallow lexical
semantics with a narrow, but deep axiomatized semantics specialized
for a particular task. To enable the system to process all information
with common mechanisms, both kinds of knowledge are represented in
conceptual graphs (CGs), and both are used in the same way to
interpret input text. The primary difference is that the CGs
associated with a specialized task may trigger additional processes to
perform further reasoning or to execute application programs. TOSI was
successfully used for legacy re-engineering, and it can also derive
CGs customized for molecular biology from published papers or
abstracts.