A New Parallel Algorithm for Generalized LR Parsing

Tomita's parsing algorithm [~Ibmita 86], which adapted the LR parsing algorithm to context fl'ee grammars, makes use of a breadth-first strategy to handle LR table conflicts. As the breadth-first strategy is compatible with parallel processing, we can easily develop a parallel generalized LR parser b~ed on Tomita's algorithm [Tanaka 89]. However, there is a problem in that this algorithm synchronizes parsing processes on each shift a,:tion for the same input word to merge many s t~ks :into Graph Structured Stacks (GSS). In other words, a process that has completed a shift action must wait until all other processes have ended theirs --a strategy that reduces parallel performance. We have developed a new parallel parsing algorithm that does not need to wait for shift actions before merging many stacks, using stream communication of a concurrent logic programming language called GIIC [Ueda 85]. Thus we obtain a parallel generalized LR parser implemented in GHC. 1 I n t r o d u c t i o n To provide an efficient parser tbr natural language sentences, a parallel parsing algorithm is desirable. As Tomita's algorithm is compatible with parallel processing, we can easily develop a parallel generalized LR parser [Tanaka 89]. However, with respect to the performance of the parallel parsing, one of the defects of Tomita's algorithm is that it forces many parsing processes to synchronize on each shift action for the same input word. A parsing process that has completed a shift action must wait until all other processes have con> pleted their shift actions as well; such a synchronization strategy reduces the performance of parallel parsing. In this paper, we will present a new parallel parsing algorithm which is a natural extension of Tomita's [Tolnita 86]. Our algorithm can achieve greater performance in parallel parsing for natural language sentences. There are two major differences between Tomita's algorithm and ours. Initially, the new algorithm does not make parsing processes wait for shift actions to merge many stacks with the same top state. The process that has finished a 'shift N' action first can proceed to the next actions until a reduce action needs to pop the element 'N' from the stack. If some other parsing processes carry out the same 'shift N' actions, their stacks will be merged into the position in which the first process has placed an element by the 'shift N' action. Secondly, to avoid duplications of parsing processes the new algorithm employs rl~ree Structured Stacks (TSS) instead of Graph Structured Stacks (GSS). The reason why we do not use GSS is because it is rather complicated to implement the GSS data structure in the framework of a parallel logic prograrnming language ,inch as GltC. The merge operation of the stacks is realized by a GttC stream communication mechanism. In section 2 we explain generalized LR parsing, in section 3 give a brief introduction to GtIC, and in section 4 decribe our new parallel generalized LR parsing algotihm. In section 5 we compare tile parallel parsing performance of our algorithm with Tornita's.