Interactive Compression of Books

In this paper we study interactive data compression and present experimental results on the interactive compression of textual data (books or electronic newspapers) in Italian or English language. The main intuition is that when we have already compressed a large number of similar texts in the past, then we can use this previous knowledge of the emitting source to increase the compression of the current text and we can design algorithms that efficiently compress and decompress given this previous knowledge. By doing this in the fundamental source coding theorem we substitute entropy with conditional entropy and we have a new theoretical limit that allows for better compression. Moreover, if we assume the possibility of interaction between the compressor and the decompressor then we can exploit the previous knowledge they have of the source. The price we pay is a very low possibility of communication errors. Key-Words: Data Compression, Interaction, Dictionary based compression, Fingerprinting.