As we speak, we mostly don't generate novel sentences from scratch. Instead, we often refer back to already holistically stored sequences. However, what constitutes such sequences is highly debated. An investigation into the occurrence of hesitations can offer some insight into the way language is composed and into the quality and role of 'chunks'.
I hypothesise that the frequency of the chunks in question plays a much greater role than commonly assumed. Thus, speakers tend not to interrupt high frequency chunks, while low frequency sequences might be interspersed with hesitations.