Natural Language Multiprocessing: A Case Study

This paper presents two case studies of parallelization of large Natural Language Processing (NLP) applications using a parallel logic programming system (called "ACE") that automatically exploits implicit parallelism. The first system considered is Artwork, a system for semantic disambiguation, speech act resolution, and temporal reference resolution. The second system is ULTRA, a multilingual translation system. Both applications were originally developed in Prolog without any consideration for parallel processing. The results obtained confirm that NLP is a ripe area for exploitation of parallelism. Most previous work on parallelism in NLP focused primarily on parallelizing the parsing phase of language processing. The case studies presented here show that parallelism is also present in the semantic and discourse processing phases, which are often the most computationally intensive part of the application.