[I was sure there was a thread like this once, but I can't find it]
[[the info in this post is mostly from various articles by Robert Blust, as well as a chapter he authored in Linguistic Change and Reconstruction Methodology]]
[[[also note that in some places in this post I've written [j] as "y", but these cases shouldn't cause any confusion]]]
Most people here probably know that the regular reflex in Armenian of PIE initial *dw is erk (*dw → erk /#__, as in erku "two" from PIE *dwo-), since it's basically the paradigm example of weird correspondences and/or sound changes (though the individual steps the change went through were probably not terrible weird; there's some evidence it went something like *dw- > *dg- > *tk- > *rk- > erk-). But there's plenty more examples! A number of Austronesian languages have had weird or even crazy sound changes. I (Ø → ŋ /#__V) in the correspondence library yesterday (the same change also occurred in some Samoyedic languages as well). But there's others!
For example, in Rennellese, *l became ŋɡ.
The Hawai'ian dialectal unconditioned change of *t → k is well known, but the *t → k change in fact happened at least twenty separate times in different Austronesian subgroups and languages. Most often this occurred following the loss of earlier *k, which freed up space for greater variation in the pronunciation of the phoneme /t/.
In several languages of western Manus Island, word final glides (*w and *j) became p -- for example, Levei op "you" (< Proto-Manus *koe > *kow) and ip "3sg" (< Proto-Manus *ia > *iy). Also puep "crocodile" (Proto-Manus *puaya), peʔep "shark" (Proto-Manus *paʔiwa), etc. These languages have several other odd(ish) changes as well, such as Drehet *t, ndr → kʰ /#__ (in nouns only, because it was conditioned by a preceding noun-marker which was later lost), e.g. kʰa "blood" (Proto-Manus *ndra); kʰu "dugong" (Proto-Manus *ndruyu); etc. Incidentally, the resulting phoneme /kʰ/ is the only aspirated consonant in Drehet (I'll post a phoneme inventory in one of the "interesting natlang features" threads or something).
In some (but not all) words, Sundanese reflects Proto-Malayo-Polynesian (PMP) *w and *b as ʧ initially and nʧ medially (i.e., *w, *b → ʧ /#__; *w, *b → nʧ /V__V). For example, PMP *sawa "python" > Sundanese sanʧa; PMP *badas "gravel, stony ground" > S. ʧadas; PMP *laban "oppose, opponent" > S. lanʧan; etc.
Robert Blust reports on a change in Northern Batak and Berawan of *b, *d, *ɡ → m, n, ŋ /__#, which he considers quite unusual. It didn't seem that odd to me, though (partly since it's similar to changes I've used in conlangs before). His main point is that final position tends to encourage devoicing, and that it's odd that the voiced stops didn't change in other environments and blah blah blah etc. I don't really buy it, but anyway if anyone was thinking of including this change in a conlang or whatever but were worried it was unnatural, you have a natlang precedent so that's good I GUESS? But for instance, Karo Batak deleŋ "mountain" (Proto-Batak *deleɡ).
Another change that's fairly unusual even if it's not totally bizarre, also occurs in Berawan languages: the onset consonant of open final syllables was geminated (*C → [+long] /$__V#), e.g. Long Terawan bittoh "stone" (PMP(?) *batu), binnəh "husband" (PMP(?) *bana), and sikkoh "elbow" (PMP(?) *siku); compare gitoh "a hundred" (PMP(?) *ʀatus), tana "earth" (PMP(?) *tanaq), or tukon "prop, support" (PMP(?) *tukud).
On the other hand, Berawan also reflects intervocalic *b and *ʀ (the latter may have been /ʀ/ or /r/) as k(k), for example: Long Jegan bikuj "pig" (PMP *babuj); Long Terawan akkuh "ash" (PMP *qabu); Long Jegan tukkəj "derris fish poison" (PMP *tuba); and Batu Belah dukkih "thorn" (PMP *duʀi) (the result is -kk- rather than -k- in some instances due to the gemination rule described above). This was evidently a several step change: first *ʀ and *b merged to *ɡ, which subsequently devoiced (*ʀ → *ɡ,  *b → *ɡ /V__V,  *ɡ → k /V__V) -- note however that this is still really bizarre on several levels: first, the change from labial to velar POA (but see below), and second, intervocalic devoicing!: *b and *ɡ/*ʀ remained voiced in initial position (in final position they became nasals, as described above); it's only intervocalically that they become k(k).
In a couple Austronesian languages, there has also been a change (independently in each language) whereby obstruents were devoiced following nasals, but not elsewhere (*O → [-voice] /N__), as in Murik lintem "dark" (Proto-Kayan-Murik *lindem). Though it's odd, similar changes have happened in some Bantu languages (e.g. Tswana, see ).
In a number of Austronesian languages (independently), a prothetic /j/ was added to words beginning with /a/ (*Ø → j /#__a). In several cases, the phoneme *j (< both inherited */j/ and this prothetic /j/) in the given language experienced further changes, creating some odd cognate lists; e.g., in Motu, *j → l, so you've got e.g. lau "I" (Proto-Oceanic [POc] *aku) and lahi "fire" (POc *api).
In Seimat, synchronically, nasalized vowels can occur only after /w/ and /h/. Nasalizing /w/ comes from earlier *mʷ, and nasalizing /h/ from earlier *d, hence, cf. kawã- "forehead" (POc *damʷa) vs. awa- "mouth" (POc *qawaŋ), or hũa "two" (POc *dua) vs. hua "fruit" (POc *puaq). The first part isn't a particularly odd change, though the second is; but in either case the result is a very unusual synchronic nasality system.
In a number of languages, the glides *w and *j underwent fortition -- not a particularly unusual change, but combined with some others (e.g. the loss of unstressed syllables) and the fact that this fortition usually included the automatic glides inserted phonetically in some vowel combinations, leads to some bizarre looking reflexes. For example, Tunjung rəɡa "two" is from Proto-Austronesian (PAn) *dusa, via: *dusa > *duha > *dua > *duwa > *duɡa etc.; and Long Terawan kəʤin "durian fruit" is from PAn *duʀian via *duʀian > *duʀijan > *ʀijan > *ɡijan > *ɡiʤan etc. In the case of Narum, however, only the phonetic glides underwent fortition, and not the inherited phonemic glides! That is, inherited /w j/ remained, but the phonetic transitions between a high vowel and a following unlike vowel became /b/ and /ʤ/, so compare dəbeh "two" (Proto-North-Sarawak [PNS] *dua) and ləʤeəh "ginger" (PNS *lia) with pawaat "fruit bat" (PNS *pawat) and bayeəh "crocodile" (PNS *buaya).
Kiput (I wrote about it in the Correspondence Library [as did Rorschach earlier, but he got some things wrong]) exhibits a number of these bizarre and/or unusual and/or interesting changes. It's got glide fortition, for one. The North Sarawak languages, though, also evolved a set of true voiced aspirates -- i.e., the stop begins voiced, but there is a delayed VOT following the stop (they can also be interpreted as stops that begin voiced and end voiceless, since apparently the aspiration is optional). These derive from older geminate stops, which in turn derive from clusters or from stops that directly followed a stressed schwa. The interesting part here is that there is no corresponding voiceless aspirate series, so in those North Sarawak languages which retain the voiced aspirate series, the stop system is: /p t c k/, /b d ɟ ɡ/, /bʰ dʰ ɟʰ ɡʰ/ [look familiar?]. But anyway, Kiput merged all of *bʰ, *dʰ, *ɟʰ into s, which gives us another fun set of reflexes, such as səiʔ "water" from PNS *əbʰak or təsəw "sugarcane" from PNS *təbʰu. Evidently, *bʰ first became *f, which later merged into s. Finally, Kiput also shows intervocalic devoicing as well: intervocalic *v, *ɟ, and *ɡ become f, c, and k respectively; this includes *v and *ɟ from earlier glide fortition, so Kiput shows some instances of intervocalic -f- and -c- for PNS hiatuses, as in cəiʔ "good" (< PNS *diaq > *diyaʔ > *diɟaʔ > *dicaʔ etc.) and dufih "two" (PNS *dua).
I'm tired of writing, now. Share your own and so on blah blah blah blah, I'll post some Plains Algonquian stuff tomorrow
Last edited by Whimemsz on Thu Aug 23, 2012 9:20 am, edited 6 times in total.