Irene Langkilde-Geary
ilangkil@isi.edu
http://www-scf.usc.edu/~langkild/

Talk title:  Statistical Sentence Generation
Abstract: 
Sentence generation is an important subtask of applications like machine
translation, human-computer dialogue, explanation, and summarization.  The
recurring need for sentence generation suggests the usefulness of a
general-purpose, domain-independent natural language generator.  However,
such a system, like any AI system, has enormous needs for knowledge, such
as lexicons, grammars, ontologies, collocation lists, and morphological
tables.  Acquiring and applying accurate, detailed knowledge of this
breadth poses difficult problems. A corpus-based statistical approach can
potentially overcome this bottleneck by tapping into the linguistic
information inherently embedded in large texts.  I will describe the
Nitrogen system, a large-scale (~200,000 words and concepts), flexible and
robust sentence generator that is easy to use.