Massively Parallel Spoken Language Processing using a Parallel Associative Processor IXM2

This paper reports experimental results of a natural language processing scheme for spoken inputs implemented on a massively parallel machine IXM2. We have carried out performance tests on language models on IXM2 including a word-pair grammar and a memory-based parsing. Experimental results demonstrate that use of IXM2 as a language processing engine for certain language models is a promising approach in attaining a real-time response. Syntactic recognition completes in a few milli-seconds. Plus, response time grows only linearly to the length of the input sentences, attaining approximately of order o(n).