Abstract:
This paper presents the reasoning behind the selection and design of a parser for the Lingo project on natural language interfaces at MCC. The major factors in the selection of the parsing algorithm were the choices of having a syntactically based grammar, using a graph-unification-based representation language, using Combinatory Categorial Grammars, and adopting a one-to-many mapping from syntactic bracketings to semantic representations in certain cases. The algorithm chosen is a variant of chart parsing that uses a best-first control structure managed on an agenda. It offers flexibility for these natural language processing applications by allowing for best-first tuning of parsing for particular grammars in particular domains while at the same time allowing exhaustive enumeration of the search space during grammar development. Efficiency advantages of this choice for graph-unification-based representation languages are outlined, as well as a number of other advantages that acrue to this approach by virtue of its use of an agenda as a control structure. We also mention two useful refinements to the basic best-first chart parsing algorithm that have been implemented in the Lingo project.