USC: description of the SNAP system used for MUC-4

The main goal of the SNAP project is to build a massively parallel computer capable of fast and accurat e natural language processing [3] . Under NSF funding, a parallel computer was built in the Parallel Knowledg e Processing Laboratory at USC and software was developed to operate the machine [2] . The approach i n designing SNAP was to find a knowledge representation and a reasoning paradigm useful for natural languag e processing which exibits massive parallelism . We have selected marker-passing on semantic networks as a way to represent and process linguistic knowledge . The work for MUC-4 started at the end of January 1992 . Prior to this we had implemented on SNAP a simple parsing system accompanied by a small knowledge base . When we started the MUC-4 effort we ha d only vague ideas how to implement large semantic network knowledge bases for linguistic processing . Thus for us MUC-4 was a major undertaking and soon we realized the difficulties and the time limitation . Faced with the MUG4 challenge, our group consisting of one faculty and five graduate students spent approximately 1450 hours to engineer a large system .