|
Abstract : |
Summary: Spoken language translation requires both (1) high accuracy and (2) a real-time response which are difficult to achieve using conventional technologies. To fulfill the first requirement, we have adopted an Example-Based Approach. It generates a target sentence by combining partial translations ob-tained by mimicking best-match partial translation examples. To fulfill the sec-ond requirement, this paper proposes using a Heterogeneous Computing Platform consisting of Multiple Instruction Multiple Data (MIMD) and Single Instruction Multiple Data (SIMD) parallel machines. Example-Based Approach is dominated by two processes, each of which is optimally accelerated by utilizing MIMD and SIMD, respectively, a) to build the source structure, and b) to retrieve the best-match examples. Experimental results show that Example-Based Approach is drastically speeded up with the Heterogeneous Computing Platform and has a performance sufficient for real-time response, even with a large vocabulary and a highly ambiguous sentence. 1., |
