Clements addresses himself to one of the oldest issues in linguistics – namely, how syllables are formed out of constituent phonemic segments – and in doing so, makes a noteworthy step forward toward a rigorously formulated theory of sound patterns. What I can contribute to this endeavor, I hope, is a somewhat different perspective on the issue. My suggestions may be too radical to be immediately useful, but if they manage to direct phonologists' attention to some emerging findings in experimental phonetics, particularly articulatory studies, that are relevant to the very basis of nonlinear phonological theories like Clements's, and if we can thereby acquire a better understanding of the basic assumptions underlying the current theoretical development both in abstract and concrete representations, it will justify my participation in this forum of discussion.

It is crucial, from my point of view, to note that Clements's idea of a demisyllable's internal organization is different from the idea I have been developing over the years. For Clements, a well-formed string of (phonemic) segments constitutes a demisyllable; the demisyllable is the domain over which sonority constraints on segment organization work. My approach, in contrast, is to try to use demisyllables as minimal integral units in place of phonemes. From this point of view, I defined a demisyllable as a set of (unordered) feature specifications (Fujimura 1979: 474), even though originally my way of thinking was still more in terms of phonemic sequences.