Embedded in a larger study that examines the form-function relationship for French positive feedback markers this study examines the form of ‘oui’ and ‘ouais’ with respect to their relative placement to the own speaker’s and the interlocutor’s speech production, and with respect to acoustic features including F0, aperiodicity and intensity. Previous corpus-based research on these items has shown the importance of positional features for determining the choice of a specific item (Gravano et al., 2012). Neiberg et al., (2013) has shown the importance of phonological operations in Swedish (similar to our ‘oui’ vs. ‘ouais’ opposition) for communicating different attitudes. Other studies took a qualitative approach towards the phonetics of response tokens in naturally occurring talk in interaction. Gardner (2001) shows that the response token ‘mm’ in English has specific acoustic features (fall-rising, falling or rise-falling pitch) when it is used in specific sequences of talk (continuation, acknowledgement, assessment). Ogden (2010) found that specific phonetic-prosodic properties (loudness, F0 height, F0 span and voice quality) of complaints have specific interactional consequences. In French, ‘oui’ vs. ‘ouais’ is assumed to be opposed in terms of level of language (‘ouais’ being more colloquial). However, their simultaneous presence in most interactive situations as well as some recent descriptive work (Péroz, 2009) tend to suggest different communicative functions for these two items. Here, we apply basic machine-learning techniques on a large dataset of these feedback items. The driving research question is how the various positional and acoustic properties interact and produce distinct properties of feedback items. The hypothesis is that independent from the lexical content (‘ouais’ vs. ‘oui’), the selected features can be used to train a classifier that distinguishes between different speech-exchange situations, e.g. conversational vs. task-oriented dialogue. The material used for this study contains two corpora. The CID corpus (Bertrand et al., 2008) are face-to-face conversations that have the mere instruction to talk about “particular” events. The Aix Maptask Corpus (Gorisch et al., 2014), which is the French version of the famous Edinburgh corpus. Two conditions are recorded: face-to-face vs. remote interaction.