One of the things I've discovered in puttering about lately in my Sprachbundesgarten of little-known languages in Papua New Guinea is that Otto Dempwolff, the granddaddy of historical and comparative Austronesian linguistics, did not recognize the possibility of labialized labial phonemes (/pʷ/, /bʷ/, /mʷ/), despite how common they are among Oceanic languages. Since Dempwolff was the chief linguistic adviser of most of the German Lutheran missionaries in New Guinea, his theoretical insights as well as limitations influenced many of the new writing systems devised by those missionaries for evangelical and teaching purposes.
I had long been aware of his influence on Jabêm, a Lutheran mission and school lingua franca in Morobe Province, PNG, where I did fieldwork in 1976. (My host father had been a teacher in Jabêm schools.) Dempwolff spent the last months of his life completing a grammatical description of Jabêm, working with a missionary, Heinrich Zahn, who was no mean linguist himself. Dempwolff died in 1938 and the grammar appeared in 1939, a rather inauspicious year that helped condemn that work to undeserved obscurity.
In Jabêm orthography, labialized velars, that is, velar consonants with secondary rounding, are written as kw, gw, ŋgw, but labialized labials are written with an intermediate round vowel before the vowel that forms the nucleus of the syllable. So [mʷa] is written moa, [pʷa] is written poa, [bʷa] is written boa, and [mbʷa] is written mboa. This seems inconsistent to me, but presents no major hurdle for people writing Jabêm. (A much greater nuisance stems from the decision to distinguish the two sets of mid vowels by marking the much more frequent member of each pair with a circumflex: upper-mid ô, ê are far more ubiquitous than lower-mid e, o.)
Jabêm's closest relative is Bukawa, which has been so long overshadowed by Jabêm's prestige that its literate speakers wrote in Jabêm rather than in their own far more varied and numerous village dialects. Now, however, a linguist from SIL International has published a grammar of Bukawa, based on a dozen years residing among its speakers. In Bukawa orthography, labialization is uniformly indicated by -w-, whether it follows a labial, velar, or even alveolar consonant (/dʷ/). (Bukawa also has a voiceless lateral, written lh, and voiceless semivowels, written yh and wh. Fascinating, and rather exotic within its Sprachbund.) In other respects, the new Bukawa orthography follows its Jabêm predecessor.
I've only recently discovered that the Sio language on the north coast of the Huon Peninsula suffered a far worse orthographic fate. The Sio community should have been assigned to the Jabêm church circuit, which included mostly Austronesian-speaking communities along the southern half of the Huon Peninsula and along the south side of the Huon Gulf. Instead, Sio was assigned to the Kâte circuit, which used a Papuan lingua franca. Worse yet, the orthography of Siâ (as it is written) was based on that of Kâte, which was also greatly influenced by Dempwolff. The dedication page of the Lutheran missionary Pilhofer's 1933 grammar of Kâte reads Herrn Professor Dr. Otto Dempwolff / in Dankbarheit und Verehrung / Ehrerbietigst Gewidmet.
Both Kâte and Sio have a set of "labiovelar" stops that are written as (voiceless) q and (voiced) q. (My boldfaced q stands for a curly q with hooked serifs that I cannot properly render here.) Each language also has a prenasalized "labiovelar" that is written ŋq in Kâte and mq in Sio. Sio also has a "labiovelar" nasal, written ɱ. Most of the German-era orthographies represent the velar nasal with ŋ and people still seem quite comfortable with it, calling it the 'long en'.
Michael Stolz, the missionary who first reduced Sio to writing, translated and compiled a book of Bible stories, catechisms, and hymns in the language, which was edited and published posthumously by his successor, Hans Wagner. After Stolz died in 1931 (after 20 years in the field), Dempwolff used his materials to write up a very rough sketch of Sio grammar, which was never published, but was transcribed by "L. Wagner" (perhaps the wife of Hans) in 1936. Dempwolff retained the "labiovelar" class of consonants.
In 1985, an SIL couple, Stephen and Dawn Clark, arrived to work among the Sio people, who soon asked about reforming their orthography to better match the conventions of Tok Pisin and English, with which most villagers were now more familiar. The Clarks discovered that the "labiovelars" were all pronounced as labialized labials ([pʷ], [bʷ], [mbʷ], [mʷ]), even by the oldest villagers they could find. (Judging from his fieldnotes, a colleague of mine discovered the same thing when he collected survey data on Sio in 1976.) The word for 'snake', for instance, was spelled ɱâta and pronounced [mʷɔta]. Its cognates are pretty widespread in Oceanic languages.
So the Sio people readily abandoned their old symbols for the labiovelars (the two varieties of q and the long ɱ, in favor of the usual labial consonants with a superscript ʷ. Feeling strongly that the labialized labials were unit phonemes, they at first insisted on writing the labialization with a superscript, but after several years they got used to writing pw, bw, mbw, and mw instead of troubling with superscripts.
So now I'm wondering, could the "labiovelars" in Kâte also be reanalyzed as labialized bilabials? Pilhofer (1933) says quite clearly that his q and curly q are both labiovelar stops, in which kp and gb are coarticulated and simultaneously released. But now I'm suspicious. I wouldn't question Pilhofer if Kâte were an African language, but I haven't encountered such coarticulated stops in New Guinea. Then again, I haven't looked at the phonologies of many Papuan languages.
References and further details on the above are now available in Wikipedia. Earlier disgruntled musings on linguists and Oceanic orthographies can be found here and here.
UPDATE: According to the World Atlas of Linguistic Structures, Eastern New Guinea is one of only two areas of the world with labiovelar stops. The other is Central and West Africa. Kâte is included in their very small sample of such languages, based on a Kâte dictionary published in 1977 (which I have never seen). So Pilhofer appears to have been correct, and Sio appears to have been doubly ill served, first by adopting a mismatched Papuan language for its orthographic model, and second by Dempwolff's failure to recognize labialized labials.