Improvements to Earley’s Context-Free Parser

Summary

This paper is devoted to the presentation of a two-parameter family M ^t_k of parsers for general context-free grammars; the algorithms have a top-down structure, in which all the possible candidate parses are investigated in parallel. Backtracking is avoided by keeping track of the stage reached in all parses in a set of “states”. The integer parameters t and k describe tests performed on strings of terminals, of length t and k respectively, to the right of the point currently reached in the analysis of an input string. The tests involving parameter k enhance the performance of the parser on grammars with LR(k) characteristics, whereas those involving parameter t are most suited for grammars showing LL(t)-type conditions. However, the use of parallelism enables the algorithm to work on any CF-grammar.