Automatic Extraction of Chinese/Japanese Translation Patterns Using Prefix Span

In late years, a large number of translation patterns are required for the pattern based machine translation. We propose an efficient method to extract the Japanese/Chinese translation patterns from the corpora using Prefix Span. They performed chunking on the sentence pairs of the parallel corpora, collected the candidate translation patterns from them using Prefix Span, and narrow down the candidates using two criteria: the point wise mutual information (PMI) and the degree of confidence for the threshold values. The proposed method achieved precision 85% when the PMI is 1.0 and the degree of confidence is 0.15.