Most text compression algorithms perform compression at character level. If the algorithm is adaptive, it slowly learns correlations between adjacent pairs of characters, then triples and so on. The algorithm rarely has a chance to take advantage of longer range correlations. If text compression algorithms were to use larger units (words) than single characters as the basic storage element, they would be able to make the most of the longer range correlations and, perhaps, achieve better compression performance. Faster compression may also be possible by working with words. On the other hand, PPM is one of the most promising lossless discrete-data and character-based compression algorithms, which uses Markov models of order k.