Tree Languages Arithmetic Compression

In this paper, we explore the appli ability to ompression tasks of the algorithms for regular language inferen e from sto hasti samples. We ompare two arithmeti en oders based upon two di erent kinds of formal languages: string languages and tree languages. The experiments show that tree-based methods outperform the predi tive apability of stringbased methods when they are applied to les ontaining stru tural information and, then, they allow for better ompression rates.