Grail: an Interactive Parser for Categorial Grammars

Since early work by Lambek [Lambek 58], linguists and logicians have been trying to specify grammar s as logical theories (see e.g. [Morril l 94] or [Moortgat 97] for a good overview). An advantage of this approach over more traditional approaches to linguistics is that we can prove our grammar s have abstract well-behavednes properties likesoundness , completenes sand consistency. Grail is a tool which allows you to specify such logical theoriesand functionsasa parser for the resulting grammar. I wil l first give a brief introduction to categorial grammar and then show how proof search for this logic is implemented in Grail. Theparser isdivided in two parts: we useproof nets to indicate, roughly, the functor-argument structure and term rewriting to implement the structural constraint s imposed by the language we are describing. On both levels, parsing is fully interact ive. When there are multiple ways of continuing the computation, the user can select which oneto try first. Benefitsof thisare that the user can guide the computation to a point in which he is interested, which is useful when debuggingagrammar andsidestepsthecomputational complexity of the parser when it operates fully automated. Grail is currentl y used at the University of Utrecht asaresearch tool and ascoursewarefor introductory to advanced level courses in computational linguistics. It is implemented in a combination of SICStusProlog and Tcl/Tk.