Fast string matching in stationary ergodic sources

A connection is made between the theory of ergodicity and the expected complexity of string searching. In particular, a substring search algorithm is introduced which, when applied to searching in text that has been produced by an appropriate stationary ergodic source, has an expected running time of O((N/m + m)logm), for a text string of length N and search string of length m. Similar expected complexity results have been obtained before, but the analysis is performed in a significantly more general framework, which models with greater accuracy the statistics of many types of strings, including natural language. The analysis also sheds light on the performance of the Boyer-Moore algorithm and the Sunday algorithm when applied to natural language.