Exploiting Parallelism in Unification-based Parsing
Marcel P. van Lohuizen · 2000
Because of the nature of the parsing problem, unication-based parsers are hard to parallelize. We present a parallelization technique designed to cope with these diculties. Parallel parsing of natural language has been researched extensively. In [6] we can nd an overview of parallel chart parsing. Most attempts, however, were not very successful [8, 3]. Only recently two NLP applications were successfully parallelized [7, 5]. However, the former focussed on Prolog and the latter exploits coarse-grained parallelism of the kind that proved unusable for our Deltra system 1 or other systems [3]. We present a more widely applicable approach, not limited to Prolog. Most unication-based parsers have characteristics that make them particularly hard to parallelize. Typically, unications account for the bulk of the processing time in unication-based parsing. However, parallelizing this operation is dicult [1] and does not speed up the CF part. Therefore, most research has focussed on ex...