Corpus-based Automatic Compound Extraction with Mutual Information and Relative Frequency Count