Pattern Matching Machine for Text Compressed Using Finite State Model

The classical pattern matching problem is to nd all occurrences of patterns in a text. In many practical cases, since the text is very large and stored in the secondary storage, most of the time for the pattern matching is dominated by data transmission of the text. Therefore the text compression can speed-up the pattern matching. In this framework it is required to develop an e cient pattern matching algorithm for searching the compressed text directly without decoding. In 1992, Fukamachi et al. proposed a method of constructing pattern matching machine that runs on Hu man coded text, based on the Aho-Coracick algorithm. However, since the Hu man code is optimal only under the assumption of the memoryless source model, the compression ratio is not very high. On the other hand, it is known that English text can be highly compressed by the compression method based on the Markov model. In this paper, we focus our attention on the nite-state model, which subsumes the Markov model as an important special case, and show an algorithm for constructing pattern matching machine for text compressed under the assumption of this model. We also give a proof of the correctness of the algorithm.