The Kagoshima dialect is perhaps one of the most unique dialects of the Japanese mainland. It earns this title since the dialect features some very heavy phonological changes that have warped most of its vocabulary and made it incomprehensible to other surrounding regions. One sound change in particular involving the reduction of high vowels has led to some interesting challenges on the phonotactic level as well as the phonological level.
With obstruent-vowel, tap-vowel and nasal-vowel sequences, the process is pretty straight-forward: with a high vowel, the two first reduce to a moraic obstruent (denoted by the symbol 'Q'), while the last reduces to a moraic nasal (denoted by 'N'). This only applies to word-medial and -final syllables, never the first one. A syllable containing a high vowel can also not reduce if the one on its immediate right is already reduced. Following these rules, */kutu/ becomes /kuQ/ 'shoes' and */mimi/ becomes /miN/ 'ears'.
The dialect also experienced a collapse of long vowels to short ones. This change appears to have come after the reduction of high-vowel syllables, so that a syllable containing a historically long high vowel will not undergo further reduction to a moraic sound, even if the vowel is now shortened. Under this premise, */kinu/ becomes /kiN/ 'silk', while */kinuu/ becomes /kinu/ 'today' (standard Japanese: /kinoo/).
Devoicing and Sibilants
One issue that has not been studied regarding the dialect is whether or not it still features high vowel devoicing. As long vowels have now become short, this leaves the possibility for the dialect to acquire the phenomenon once anew. For instance, in the regional word /tutu/ tsutsu or /tjutju/ chuchu 'butterfly' (cognate to Japanese /tjootjoo/ chouchou), could one or both the vowels be subject to devoicing considering an environment of voiceless consonants?
The reason for needing to know this is also related to how sibilant consonants react to high vowels. No official study has been done on the phenomenon, but it has been put forward that sibilant-vowel sequences reduce to merely a sibilant. Under this analysis, [ɸuto] 'person' (a variant of /hito/) would reduce to [ɸto]. What remains to know is whether this is actually true, or is the vowel devoicing in the sequence as [ɸu̥to]? If full deletion is happening, the second question then becomes: are these sibilants moraic/syllabic (i.e. are we looking at something like [ɸːto] or [ɸ ̍to]), or does the consonant cluster with the following syllable, or does it attach itself to the preceding syllable (even if not part of the word)?
It would also be of importance to know whether sibilant-vowel sequences have the same constraints as the obstruent and nasal sequences previously mentioned, and it would be good to find out whether any distinction occurs when followed by a voiced or nasal syllable. The next thing to wonder is how these sibilants interact with historically long vowels that are now short. Do these devoice at all? How do they differ from sequences with historically short vowels?
The syllable structure
Given all the above-mentioned issues, it is currently impossible to determine with precision the exact syllable structure possible in the Kagoshima dialect. If reduced sibilant syllables in Kagoshima are proven to no longer contain any vowel, then Kagoshima would be the only known Japanese dialect to feature a more complex syllable structure, possibly comparable to that of Miyako. But whether clusters exist or whether these sounds become moraic or syllabic remains to be known.
Assuming just devoicing, then the syllable structure would permit /CV/ sequences as well as the two moraic consonants /N/ and /Q/. Assuming full deletion, then /CV/, /CVS/ and /SCV/ sequences would be possible along with the moraics /N/ and /Q/. Assuming the syllabic or moraic status of sibilants, then the structure would become /CV/ with /N/, /Q/ and /S/. Whether palatalized and labialized sounds should be analyzed as clusters or independent syllables is a matter of preference, but this would only augment the sequences to /CjV/ and /CwV/ (with restrictions to the latter).
It's worth nothing that /S/ may encapsulate up to six different sibilant sounds that are more or less stable, unlike /N/ and /Q/ which have predictable realizations and consistently assimilate to the following sound. Given this, and if the vowel does happen to be deleted, Kagoshima may effectively contrast such sounds a /h/ and /ɸ/ or /ɕ/ and /s/. This would further give the Kagoshima dialect one of the largest consonant inventories in the Japonic strata.