Resolving Noun Compounds with Multi-Use Domain Knowledge

In this paper we describe a system for semantic interpretation of noun compounds that relies on world and domain knowledge from a knowledge base. This architecture combines domain-independent compounding rules with a task-independent knowledge representation, allowing both components to be flexibly reused. We present examples from Scientific American text, on which the system was developed, and then describe an exercise that tests the portability of the architecture to a new domain: email text on the topic of conference planning.