Earley Deduction

This paper first reviews Earley Deduction, a generalization of the Earley Parsing Algorithm to the execution of Horn Clause Logic Programs. Earley Deduction is both sound and—unlike the standard Prolog interpreter—complete; proofs of this are included. For functor-free programs, the method is also guaranteed to terminate. (The functor-free subset of Prolog is called DATALOG and can be used to compute relational join, selection, transitive closure, etc.) Earley Deduction is elegant but, unfortunately, the price paid for completeness is efficiency. We conclude by describing implementation techniques that make Earley Deduction practical for DATALOG programs. Review of Earley Deduction The traditional Prolog interpreter performs a depth-first search for a proof and consequently suffers from some of the problems associated with depth-first parsing algorithms. For example, if a program contains left-recursive rules the interpreter may go into an infinite loop, failing to find a proof even though one exists, just as a top-down parser may fail to find a parse when given a grammar containing left-recursion. In 1970, Jay Earley described an algorithm called Earley Parsing that overcomes some of the problems of both top-down and bottom-up parsing by combining aspects of both [Earley, 1970]. In computational linguistics, various incarnations of this algorithm have been studied under the name Chart Parsing. 1 Although written long ago, when I was a Ph.D. student at Oregon Graduate Center / Oregon Graduate Institute / OHSU, this paper was reformatted in May, 2009, and posted to the web at this time. Other than repairing typos, no substantive changes have been made.