Characteristics of Sentence Length in Running Text

This article reports on the examination of the distribution of sentence lengths in a number of running texts. More specifically, the question we try to find an answer to is what the relation is between the lengths of adjacent sentences. We start our examination on the basis of the hypothesis that authors may aim at an alternation of long and short sentences. The results of autocorrelation scores of sentence lengths, however, show that in none of five corpus texts that we examined alternation of sentence lengths is a consistent feature. Rather the opposite is true: two adjacent sentences are more likely to have similar lengths. This is especially true in the fiction texts. We then discuss two different ways of assessing the degree to which adjacent sentences in running texts are more likely to have equal lengths than any random selection of two sentences, namely the mean squared jumps (i.e. a measure derived from the difference in length between two adjacent sentences and the autocorrelation of jumps. Finally, we test whether within a text there are fragments where alternation of sentence lengths might occur even though the text as a whole was characterized, by the other tests, as displaying no alternation. This is done by looking at consecutive fragments of fifteen adjacent sentences, and relating the number of fragments with a certain degree of negative autocorrelation with the number of fragments that can be expected to have a negative autocorrelation under randomness. The two fiction texts are shown to have more fragments with a negative autocorrelation than can be expected under a random sequence hypothesis. A number of patterns for these fragments are established. They are exemplified in the last section.