Markov conducted the first [[n-gram language model]] experiment by counting the frequencies of character bigrams in a novel by Pushkin. He assumed that the probability of the next word is independent of some of its earlier history, thus this is called a Markov independence assumption.