Balancing and clustering of words in the Burrows-Wheeler transform

摘要

Compression algorithms based on Burrows-Wheeler transform (BWT) take advantage of the fact that the word output of BWT shows a local similarity and then turns out to be highly compressible. The aim of the present paper is to study such "clustering effect" by using notions and methods from Combinatorics on Words. The notion of balance of a word plays a central role in our investigation. Empirical observations suggest that balance is actually the combinatorial property of input word that ensure optimal BWT compression. Moreover, it is reasonable to assume that the more balanced the input word is, the more local similarity we have after BWT (and therefore the better the compression is). This hypothesis is here corroborated by experiments on "real" text, by using local entropy as a measure of the degree of balance of a word. In the setting of Combinatorics on Words, a sound confirmation of previous hypothesis is given by a result of Mantaci et al. (2003) [27], which states that, in the case of a binary alphabet, there is an equivalence between circularly balanced words, words having a clusterized BWT, and the conjugates of standard words. In the case of alphabets of size greater than two, there is no more equivalence. The last section of the present paper is devoted to investigate the relationships between these notions, and other related ones (as, for instance, palindromic richness) in the case of a general alphabet.