SBIR-STTR Award

We will assess various approaches to data-flow modeling, programmability, reliability, communication and performance by surveying existing data-flow work. We will experimentally test the feasibility of data-flow architecture for a numeric and non-numeric application. The problem of syntactic parsing in natural language is the selected non-numeric application. We will define a research strategy and experiments for a larger multiprocessor system for phase ii of this project.

This Phase II SBIR proposal on programming parallel architectures for knowledge-based system applications continues work carried out by Language Systems, Inc. for Rome Laboratory/COES under a Phase I SBIR contract (Contract No. F30602-90-C-0053), which explored the applicability of the data-flow model to non-numeric processing. The proposed work involves requirements definition, design, and implementaion of a data-flow development environment as well as validation of the implemented design via experimental development of selected segments of knowledge-based and natural language processing (NLP) systems. Multiprocessor KB/ NLP applications will be considerably simpler to develop in such an environment, and should show substantial improvement in speed of execution, based upon research performed under the Phase I SBIR effort and related work by one of the co-principal investigators, reported in Appendix A.