CHR Grammars with multiple constraint stores

CHR Grammars (CHRG) are a recent constraint-based grammar formalism added on top of CHR analogously to the way Definite Clause Grammars are defined and implemented over Prolog. A CHRG executes as an error robust bottom-up parser, and the formalism provides several advantages for natural language analysis reported elsewhere [4–6]. A notable property is CHRG’s inherent ability to handle abduction in a way that requires no metalevel interpreter which otherwise is a common way to implement abduction. The technique, first noticed in [3], consists of a straightforward reformatting of grammar rules so that abducibles are passed through the constraint store, and this allows other CHR rules to serve as integrity constraints. The principle works best for grammars without local ambiguity but this is of course too limited as our main target is natural language analysis. To do abduction with an ambiguous grammar, additional techniques are necessary to avoid mixing up different sets of abducibles that represent different abductive interpretations. The last thing does not fit properly with the current paradigm of CHR, and our current implementation, which is based on impure facilities of the SICStus Prolog version of CHR [12], is not very efficient. In the present paper we propose an architecture aimed at efficient execution of abductive CHR Grammars. It is based on a multiple constraint store model which also generalizes to committed choice languages with disjunction such as CHR∨ [2]; it is indicated how constraint solving in this model can be optimized using methods developed for data integration.