The paper does not cite [1], which describes prediction by partial match algorithm with unbounded context length. [1] https://www.researchgate.net/publication/

The paper does not cite [1], which describes prediction by partial match algorithm with unbounded context length. [1] https://www.researchgate.net/publication/2473004_Unbounded_Length_Contexts_for_PPM https://www.researchgate.net/publication/2473004_Unbounded_L... That is from 2003 and actually quite interesting. It shows, for example, that these models were applied to rather small texts, because of the presence of context with length of -1 (uniform distribution). When text is large the need for such context vanishes, but for small texts it can give an advantage.