Three independently motivated types of information are usually assumed to influence prosodic boundary placement and to play a role in their relative strength: the morpho-syntactic structure, the information structure and the metrical complexity. The phonetic realization associated with the different boundary types (in particular IP and ip) is also assumed to vary. Based on data of clitic left-dislocations in French, we argue here that differences in the relative strength of the prosodic boundary occurring at the end of the dislocated XP (i.e. an intermediate (ip) or an intonational phrase (IP) boundary) cannot be derived in a straightforward manner from these three types of information. In a production experiment, where the syntactic and information structure were controlled, while the metrical complexity was varied, it appeared that the strength of the boundary occurring at the right edge of the dislocated object NP displayed a high degree of variability. In addition, the results indicate a lack of correlation between metrical complexity and boundary strength. The results lead us to argue that a sort of phonological neutralization occurs in certain textual contexts. This neutralization does not allow for distinguishing between intermediate and intonational phrase boundaries in all cases.