Text Realization for Dialog
Susan McRoy, Songsak Channarukul, Syed Sumera Ali · 2000
We present a natural language realization compo-nent, called YAG, that is suitable for intelligent tutoring systems that use dialog. Dialog imposes unique requirements on a generation component, namely: dialog systems must interact in real-time; they must be capable of producing fragmentary output; and they may be re-deployed in a number of different domains. Our approach to real-time natural language realization combines a declara-tive, template-based approach for the representa-tion of text structure with knowledge-based meth-ods for representing semantic content. Possible text structures are defined in a declarative lan-guage that is easy to understand, maintain, and re-use. A dialog system can use YAG to realize text structures by specifying a template and content from its knowledge base. Content can be specified in one of two ways: (1) as a sequence of proposi-tions along with some control features; or (2) a set of feature-value pairs. YAG’s template real-ization algorithm realizes text without any search (in contrast to systems that must find rules that unify with a feature structure).