In naturally occurring conversation, the emerging utterance takes shape through the production of intonation units. The intonation unit may be informally defined as a spurt of speech uttered under a coherent intonation contour. As a fundamental unit of spoken and signed language, the intonation unit’s importance is two-fold: (a) The intonation unit is universally recognizable by the prosodic cues that mark its boundaries, even in a language you’ve never heard before (Troiani, Du Bois & Gries 2022); (b) By hypothesis, the intonation unit represents a prosodic-cognitive workspace tied to working memory, whose affordances and limitations shape the development of the emerging utterance. Different research traditions have identified different units as organizing one aspect or another of the speech production process. Three recognized key units are the utterance, turn, and intonation unit. Now, it is easy enough to see why we need utterances and turns – to package our messages and coordinate our interactions, respectively – but the function of the intonation unit is less clear. To address this question, we present an exploratory statistical analysis of the 67,000 manually transcribed intonation units in the Santa Barbara Corpus of Spoken American English (Du Bois et al. 2000-2005). Taking words as proxies for linguistic function, we present heatmaps showing the distribution of a given word type over the time course of the intonation unit. We further present a clustering model based solely on intonation unit properties (the size of the intonation unit in words, and the position of a word relative to intonation unit boundaries). What emerges from the model is that some words/functions show a well-defined behavior in relation to the time course of the intonation unit, while others vary in less predictable ways. Interjections are the clearest part-of-speech-like category to emerge, suggesting that interjections show their deepest connections to prosody, but not to syntax. Among function words, certain other part-of-speech-like groupings emerge with reasonable clarity, including some connectives, discourse markers, and prepositions. In contrast, content words seem less likely to converge. Taken together, the results suggest that intonation units have a psychological reality closely tied to working memory insofar as it shapes the utterance production process, rather than to the conceptual content that shapes the internal phrasal structure of the message. We conclude with a more speculative discussion of recent cross-disciplinary research that is suggestive of a possible evolutionary precursor of the intonation unit.