Understanding Written Procedures

Gary C. Borchardt, Sue Felshin, Boris Katz, Pat Langley

Keywords: natural language understanding, semantic interpretation, procedural descriptions

Abstract: In this paper, we pose a new problem for AI researchers: to develop systems that understand written instructions for complex procedures. We state the problem, the challenges it poses, and why existing techniques cannot yet solve it. We review prior research on language processing that could be useful, including its limitations, and analyze the computational elements that solutions to the problem must incorporate. Finally, we consider the use of controlled experiments to evaluate systems that learn procedures from instructions and to gain insight into their behavior.

Paper Track: Technical paper

Submission Number: 16