Parameterizing Menu Based Natural Language Interfaces with Location Models

Abstract

Menu based natural language interfaces (MBNLI) use cascaded menus and completion dialogs to keep users from straying beyond the bounds of language coverage that an underlying target system can understand. Domain restricted English subsets can be parsed and translated into the native language of the underlying system, thus resulting in an effective means for humans to communicate with a computer. By acting as a sort of "universal remote" to many agents in a system, an MBNLI can enable a user to command and query these agents using natural language. This paper explores parameterizing an MBNLI grammar with a location model - a set of rules having to do with spatial information that can be reused across many agent systems and enables rapid development of MBNLI-equipped agent systems