Ah, thanks for clearing that up. So phonemic theories assert that realizations are dependent on the contrast exhibited and the environment around the contrast, but we could just as easily conceive of the contrast set as depending on environment, or of the contrast segment and its environment having no clear boundaries, or of any other relationship between realization, contrast, and environment.

>Theories within a phonemic framework usually assume that the basic building blocks of phonemes (features, elements, atoms, etc) are drawn from a universal set. This universal set is claimed to be sufficient to express the contrasts found in all languages of the world. So a theory that dispenses with a universal feature set in favor of language-specific parochial phonological properties can be non-phonemic.

Is close vs. loose contact an example of a language-specific feature/element/atom? also, is that part of what's going on with the short-long-overlong contrast going on in Estonian?

>Theories within a phonemic framework assume a single set of contrasts for the whole language. Restrictions on the distribution of segments are stated as phonotactic constraints (e.g., English prohibits /ŋ/ word-initially). So a theory that admits for multiple sets of contrasts without special phonotactic restrictions may be non-phonemic.

So this would be like defining one sound inventory for one environment, and then defining another sound inventory that is roughly parallel to the first, for use in a different environment. That is, the sound inventory specific to the word-initial environment simply doesn't include /N/, but the sound inventory specific to the syllable-final environment does. Am i understanding that correctly?

>It's a purely abstract phonological category, but it has some interesting phonetic exponency. ... While this particular contrast isn't necessarily non-phonemic, it wouldn't have occurred to me to use it if I thought of sound structure as merely a string of phonemes.

Ok, this is exactly the kind of thing i'm interested in. My project is intended to have a polysynthetic verb, and i really like the disjunct/conjunct boundary of Athabaskan languages. I know (and probably so do you) that the forms in the conjunct domain have been subject to centuries of sound changes, but morphologically what they had was working for them, so the sound changes had to improve ease of pronunciation while preserving individual exponents as recognizable motifs (i.e. remaining agglutinative, not becoming fusional).

I want to emulate this feature naturalistically, but if i filled in the cells in the paradigms right now, at my current level of morphophonological understanding, all the exponents would be linearly segmentable affixes but for the heaps of wildly unmotivated metathesis i would throw in to make the conjunct domain look "old". See, i don't know how to design an agglutinative system without linearly segmentable affixes, because if i start to add too much nonlinearity, i worry i'm just making a fusional language with irritatingly long affixes. I feel like an understanding of sound structure as more than a string of phonemes would make it easier to tell the difference. Is that misguided? It would certainly simplify my task a lot, compared with my current approach of evolving a polysynthetic verb from an analytic one.