ARDA AQUAINT

1
A New Mathematical Framework for Language
Representation, Association, Processing and
Understanding Fair Isaac Corporation

• OBJECTIVES
• Demonstrate automated extraction of semantic
  structure of English, Chinese, Arabic, and
  Spanish
• Develop a context-sensitive system for
  determining word syntax (meaning and usage) in
  context of surrounding text
• Apply above capabilities to sentence meaning
  comparison for question answering
• Promulgate capabilities throughout AQUAINT
  community

A three-level cortronic architecture for sentence
representation

• IMPACT
• Unsupervised learning based on large language
  corpora.
• Automatic creation of semantic hierarchies
  (example The Semantic Map of English) from this
  we derive generally useful semantic distance
  metrics similar to WordNet distance.
• Language independent algorithms can be applied
  quickly to Chinese, Arabic, and Spanish.
• Our entirely novel approach can be applied to
  many other types of data (multilingual text,
  audio, video, etc.).
• Impact of our base sentence-meaning-comparison
  system will be to finally have a language
  understanding program, eliminating further search
  for this capability.

AUGUST 2003
Principal Investigator Robert Hecht-Nielsen,
Fair Isaac