Learning by Reading: Two Experiments

  • Rutu Mulkar ,
  • Jerry R. Hobbs ,
  • Eduard Hovy ,
  • Hans Chalupsky ,

This paper addresses the challenge of learning information by reading natural language text. The major aim is to map natural language input into logical expressions anchored upon concise and specific theories underlying the domains, in such a way that a reasoning engine can be used to answer questions about the input. We define a 3-step procedure, including parsing and abduction, and explore different implementations for the steps. Experiments were conducted in two domains, chemistry and biology, and the versatility of the approach suggests that extension to other domains is possible when the underlying theories are suitably specified.