If the actual model of our tree source is not known, we can use a
context tree to compute an appropriate coding distribution.
A context tree (see figure 4) consists of nodes that
correspond to contexts s up to a certain depth D.
The root 
of the context tree corresponds to the empty
context.
Each node s in the context tree is associated with the subsequence of
source symbols that occurred after context s.
Example:
Figure 4: The context tree splits up the source sequence.
Suppose that the source generated the sequence 0100110 while the
past symbols were 
. Then the source symbols are
partitioned by the context tree, see figure 4.