Information flow in tabular interpretations for generalized push-down automata

Abstract

This paper presents a general framework for deriving tabular algorithms for a very large class of stack-based computations, not only in context-free parsing but in logic programming as well and more generally for all kinds of “information” domains (abstract domains, constraint domains). Tabular algorithms store traces of computations in a table to achieve computation sharing, which is most useful when dealing with non-deterministic computations. By considering what can be naively described as partial information on stack elements, we interpret these traces as stack fragments. Tuning the exact amount of information present in these traces as stack fragments. Tuning the exact amount of information present in these traces allows us to improve tabular evaluation of stack-based computations, both by increasing the sharing of partial computations and by unifying different tabular algorithms within the same framework.