Despite the proliferation of corpus-based studies focusing on the complex category of discourse markers in the recent years, consensus is yet to be found regarding the most reliable yet informative model to describe their behavior in authentic data. The major and most widely spread frameworks include the Penn Discourse TreeBank (Prasad et al. 2008), Rhetorical Structure Theory (Mann & Thompson 1988), Segmented Discourse Representation Theory (Asher & Lascarides 2003) and the Cognitive approach to Coherence Relations (Sanders et al. 1992). These models disagree both on the top levels (number and type of generic annotation levels, if any) and the specific relations included in them. It is precisely the relation between top levels and corresponding sublevels that will be discussed in this paper, starting from a recent proposal of functional taxonomy applied to the French-English spoken corpus DisFrEn (Crible, in press) and its revision in the framework of the LOCAS-F corpus (Degand, Martin & Simon 2014). In DisFrEn, four top-level functions or "domains" are distinguished, from the revision of existing proposals for both speech and writing (Cuenca 2013, González 2005, Halliday & Hasan 1976, Zufferey & Degand in press): ideational (objective relations), rhetorical (subjective, metadiscursive functions), sequential (structuring functions) and interpersonal (speaker-hearer relationship). In the original model, these four domains include a total of thirty functions, each of them belonging to one – and only one – domain. For instance, the function labeled "CAUSE" is always ideational, while "MOTIVATION" is always rhetorical, etc. This interdependent system is intended to maximize the informativity of the annotation labels (one label for one function, vs. combinations of labels) while giving the opportunity to filter the distribution from thirty to four values, hence more efficient for quantitative purposes. The annotation procedure doesn't specify which decision should be made first, the domain or the function, and it is assumed that both orders are possible although not equally relevant depending on the specific function at stake and/or the research question, annotator's expertise, etc. (see Crible & Degand 2015). An on-going project (Degand & Simon 2015) is currently working on a revision of the DisFrEn taxonomy aiming at reducing the number of options and enhancing the reliability and cognitive validity of the model. The main difference between this revision and the original is that domains and functions are no longer interdependent. On the contrary, it is assumed that most – if not all – functions can be assigned more than one domain, roughly following Sweetser’s (1990) discourse domains. For instance, a [CONTRAST] can be ideational (1), rhetorical (2), and even sequential (3) or interpersonal (4), a proposal which still needs careful investigation and discussion. (1) I wasn’t looking forward to doing it but I am now (DisFrEn EN-phon-01) (2) a rebate is when they send the money back // yes but how do you define it in economic terms (DisFrEn EN-clas-02) (3) (after a digression on the industries in Bristol area) but Bristol itself is a large metropolis (DisFrEn EN-intf-05) (4) I think the Marks is better // actually I’m not sure it is (DisFrEn EN-conv-01) In this presentation, we will address several methodological implications of the interdependence vs. independence of domains and functions, paying particular attention to issues of inter-rater reliability and overall validity and consistency of the taxonomy. In addition, quantitative results will be presented to illustrate the comparison of the two versions on samples of spoken French from the LOCAS-F corpus annotated with both systems.