Term Encoding of Typed Feature Structures

I explore in this paper a variety of approaches to Prolog term encoding typed feature structure grammars. As in Carpenter [3], the signature for such grammars consists of a bounded complete partial order of types under subsumption (⊒) along with a partial function Approp:Type×Feat ⇀ Type. The appropriateness specification is subject to the constraint that feature-value specifications for subtypes are at least as specific as those for supertypes: if Approp(t, f) = s and t subsumes t, then Approp(t, f) = s for some s subsumed by s. Previous approaches to term encoding of typed feature structures ([1], [2], [7]), have assumed a similar signature plus additional restrictions such as: limitations on multiple inheritance, or exclusion of more specific feature-value declarations on subtypes. The encoding presented here is subject to no such restrictions. The encoding will ensure that every feature structure is well-typed (Carpenter [3]), i.e., for every feature f on a node with type t, the value of this feature must be subsumed by Approp(t, f). And furthermore, the encoding will ensure, as required by hpsg, that every feature structure is extendible to a maximally specific well-type feature structure.