This paper describes the distribution of pitch patterns and communicative types in the interpausal units (IPUs) preceding pause or gap silences extracted from a corpus of spontaneous speech as part of our work towards automatic prediction of turn-taking in dialogue interaction. IPUs preceding speaker change (‘Gaps’) and IPUs preceding silence where the same speaker continues talking (‘Pauses’) were selected in the course of automatic extraction of pause/gap silences in dyadic dialogue interactions. A listening test was conducted to estab-lish ‘human predictable’ pause/gap data sets which were sub-sequently manually annotated in terms of pitch patterns and communicative types. Overall, the Gaps and Pauses subsets show differentiation in terms of both their communicative types and pitch tunes. Declaratives and Questions are mainly found in Gaps, whereas in Pauses we mainly find Hesitations and Incomplete Declaratives. Gaps are generally characterised by falling or rising pitch patterns, whereas in Pauses a large proportion of speech samples are realised with level pitch. Classification experiments reveal strong discrimination of pauses and gaps for both prosodic and functional annotation labels.