The minimum word length in Li ’ s article

Li [1] does not explicitly assume that the minimum word length that his random text model generates is one. The fact that he shows examples of random texts (p. 1842 of [1]) with sequences of more than one blank in a row at the beginning of his article can be confusing because there is evidence later in the same article that the author is implicitly assuming that words have a minimum length of one. In Eqs. 3 and 15 of [1], summations are restricted to lengths greater or equal than one. Further evidence of the absence of empty words in the simulations comes from Fig. 1 of [1]. There, the plots of the rank spectrum for different alphabet sizes of a random text with equal character probabilities start with plateaus of the number of characters in the alphabet (excluding the space), confirming the absence of empty words. However, the manner in which the parameters of the simulations with unequal character probabilities are presented is confusing. We assume that we have N characters other than space and p1, ..., pi, ..., pN are the probabilities of each these characters in Li’s model and pb is the probability of blank. The presentation by Li of these probabilities allows one to interpret although incorrectly from what we have noted above that the probability of a blank does not depend upon the number of characters that have already been placed for the current word. In contrast, if the current word does not have any characters the probability of a blank is actually zero and the probabilities for other characters are no longer valid. For this reason, it would be more accurate to state that pb, p1, ..., pN are the character probabilities when the word being constructed has more than one character. If not, then