Artificial speech generation has existed for many decades and recent applications produce quite convincing results as evidenced by a number of widely known commercial applications. However, only recently have some of these applications begun to exhibit typical signs of spontaneous speech with the inclusion of disfluencies, despite the fact that these applications are most definitely NOT producing spontaneous speech. While the idea of inserting disfluencies has been tested in some ways for the past decade, there is not much controlled investigation of this trend. A corpus study was performed to look at patterns of disfluency use in a cross-linguistic manner. The experiment used the Crosslinguistic Corpus of Hesitation Phenomena (Rose, 2013) and extracted a wide variety of acoustic features of disfluencies exhibited in this speech corpus. The acoustic information was processed through a clustering algorithm to find common disfluency trends among sub-groups of the 35 speakers in the corpus. Surprisingly, the clustering method yielded a large number of groupings, suggesting that speakers are quite unique in their disfluency methods and cannot be easily categorized. One conclusion is that “mean” disfluency trends might be used in generated speech, for example as “nudges” to help learners achieve certain learning outcomes more readily.