Ah, glottalisatioin in Australian languages. One of my favourite topics.
Let me add to Mark’s query that the glottal stops in question (at least in Yolŋu languages) have a weird distribution even within their restricted distribution, and an odd realisation too.
Intervocalically, they are a glottal stop. Word finally (after a vowel), they usually disappear entirely, although not for all speakers.
In complex codas, they are most commonly realised as glottalisation on the preceding consonant, rather than a separate stop. This is especially true with l, ḻ (retroflex l) and n. I also heard more nasalisation on the preceding vowel than when there was no glottal, at least when the syllable is stressed. e.g. buny’tjun “smoke” versus gandjarr “careful”- way more nasalisation on the u. This stuck out for me because Bardi speakers never nasalise their vowels that I heard. In contrast, in a word like runu’-ŋa ‘on the island’ there’s no checking of the vowel.
Wonderfully different things happen at morpheme boundaries in different Yolŋu languages. There’s a paper (I’m being vague because I can’t remember if I’m allowed to quote from it or not) that says in some Yolŋu varieties the glottalisation is a property of the whole syllable rather than a segment, and it should perhaps be analysed as an accent feature. Someone (I think in the same paper) points out its interesting behaviour at morpheme boundaries, and how it metathesises. That doesn’t happen in Yan-nhaŋu though.
Finally, glottals appear in reduplication, but it’s morphologically conditioned. For example, garama ‘go’ reduplicates to garagarama, but manapan reduplicates to mana’manapaṉ. If the base is glottalised, the reduplicant retains the glottal, but there is only one glottal in the word: buny’tjun ‘smoke’ > buny’bunydjun or buny’bunytjun, for example.
Anggarrgoon 