1 On the role of autocorrelations in texts

The task of finding a criterion allowing to distinguish a text from an arbitrary set of words is rather relevant in itself, for instance, in the aspect of development of means for internet-content indexing [1] or separating signals and noise in communication channels [2]. The Zipf law is currently considered to be the most reliable criterion of this kind [3]. At any rate, conventional stochastic word sets do not meet this law. The present paper deals with one of possible criteria based on the determination of the degree of data compression. The most natural approach to solving the above problem is, no doubt, a study of autocorrelations in sequences of words forming a document. Data compression used in various file compression systems, in particular, gzip archiver, can be proposed as a mechanism for study of similar autocorrelations. Really, the compression method it employs is based on coding repeated patterns [4], which, undeniably, in case of text compression, depends, among other things, on the order of words: the more stable patterns are there in the data set, the more efficient is the compression. It can be supposed that changing the order of words arrangement in a coherent text an