The paper addresses the differences between ecological data and laboratory elicited data for the study of the semantics/pragmatics of prosody. A new experiment design -Rep Task- is presented. It is based on a reduplication task: subjects are asked to reenact a conversation that has been recorded beforehand and turned into a script. Thus, the data consist in several renditions of the same conversation: one (the original) has been produced in a natural setting while the other(s) are read off the script of the original conversation. Those renditions may be compared in order to discover the extent to which and how they differ or match. A pilot study is reported: it is based on a 12 minutes' extract of a dialogue taken from the CID corpus (Bertrand et al. 2008). Two case studies are developed to illustrate the type of evidence Rep Task may contribute. One is the use of a particular pitch contour identified by Portes et al. 2007 under the label "rising of list". The other is the restriction on pitch range and pitch contour associated with the use of reprise declaratives in the interactive management of topic flow. We conclude that subject's choices in the lab converge with speakers' choices in everyday interactions. This conclusion about prosodic choices is in line with Bresnan's 2007 conclusions about lexicosyntactic choices.