Résumé
This article identifies a number of criteria that allow us to determine whether the “barriers” that prevent phonological processes from applying across word boundaries are of representational (hash-marks # in SPE, empty CV units here) or procedural (restrictions on the cycle, the PIC in current syntactic theory) nature. One result is that process-specific sandhi-blocking (a given morpho-syntactic division may block this, but not that process), a hard fact of sandhi phonology, cannot be due to representational intervention: it must be a consequence of a PIC condition, which may or may not be associated to individual phonological processes. The same generalisation holds on the representational side: parametric variation is studied regarding the question what the initial CV can be initial of. Two cases are identified: the initial CV heads words in some languages, utterances in others (Corsican illustrates the latter pattern). Given that the initial CV always heads computational domains, it turns out to be phase-initial in all cases. Its distribution, though, is just like the distribution of the PIC: phases may or may not be headed by CV units on a language-specific basis. That is, units that the initial CV can be initial of are only phases, but not every phase is headed by an initial CV. It is shown that this parametric perspective on footprints which are left by phase-based spell-out in phonology may be a way to account for evidence that has been adduced against symmetric spell-out (when a phase head is spelt out, it is sent simultaneously to both PF and LF, not just to one or the other interpretational device).