Perry, N.,Binder, P. M.
We propose a measure of complexity for symbolic sequences, which is based on conditional probabilities, and captures computational aspects of complexity without the explicit construction of minimal deterministic finite automata (DFA). Moreover, if the sequence is obtained from a dynamical system through a suitable encoding and its equations of motion are known, we show how to estimate the regions of phase space that correspond to computational states with statistically equivalent futures (causal states). [S1063-651X(99)11707-0].