complementary distribution

[txmmj]

Could two separate phonemes be in complementary distribution? From everything I've read or heard in classes, phonological complementary distribution is brought up when talking about two allophones belonging to one phoneme. But would two distinct phonemes be able to be in complementary distribution? Is it a simple yes/no or is there another term for this occurance?

For example, from what I know (anyone feel free to correct me if I'm wrong) about Mongolian phonology, both /g/ and /ɢ/ are separate phonemes written with the same letter in Mongolian's cyrillic alphabet. The former only occurs in words containing +ATR/"front" vowels and the latter only occurs in words contaiming -ATR/"back" vowels. If I saw this in a corpus, I would assume both are allophones of most likely /g/ in complementary distribution, but apparently both sounds are distinct.

If anyone could clear this up for me, it'd be greatly appreciated!

[zompist]

Sure. The classic example is /h/ and /ŋ/ in English.

From your description it sounds like Mongolian /g ɢ/ could be allophonic, but I don't know the language and there may be good reasons they're not.

[gach]

The Mongolian (Khalkha) phonemes /g ɢ/ are distinct and their distribution isn't fully determined by the ATR feature of the surrounding vocalism. Svantesson gives the minimal pairs /bag/ "team" vs. /baɢ/ "small" and /dzʊrga/ "picture-REFL.POSS" vs. /dzʊrɢa/ "six" (p. 226) and further examples under Ex. (10) not all of which have a predictable choice between velar and uvular stops.

Mostly it all looks historically very transparent still which fits the picture that the velar-uvular opposition is a young one. From the previous minimal pairs /baɢ/ is a case of /g/ > /ɢ/ before a word final /a/ that later got dropped while /bag/ preserves the stop as velar as it's an originally consonant final word. In /dzʊrɢa/ the final /a/ following /ɢ/ is still present while in /dzʊrga/ it's a separate morpheme that's suffixed on an originally consonant final word putting there /g/ before an /a/. I don't know Mongolian too well but I presume that all this is only possible in -ATR words since in +ATR words there isn't anything that could have caused the /g/ > /ɢ/ backing.

A small side note on notation. I'm going by the convention of writing the weak stops and affricates with "voiced" symbols (<b dz> etc.) despite the fact that the labial and coronal members of this series are pretty much always pronounced unvoiced. The corresponding "unvoiced" symbols are reserved for the strong series witch has an aspiration feature.

[vec]

Sure. The classic example is /h/ and /ŋ/ in English.

Are there no places where /h/ and /ŋ/ contrast? Does this mean they could be analyzed as the same phoneme?

[gach]

Are there no places where /h/ and /ŋ/ contrast? Does this mean they could be analyzed as the same phoneme?

I'm sure there are no minimal pairs but doesn't behind the singers have both intervocalic /h/ and /ŋ/ in most dialects at least putting them into the same environment.

But even if the distribution of /h/ and /ŋ/ were absolutely complementary with no exceptions, it would be silly to think of them as one and the same phoneme. There aren't any special functional connections between the two sounds that would support such an analysis.

Here's another similar situation to think about. In a language that has both word final /ʔ/ and empty onsets alternating freely between a phonetic [ʔ] and a fully vocalic onset, are these two different [ʔ] better to be analysed as phonemically identical or not? An example from Finnish (additionally marking the normally unwritten glottal stop) would be

syöʔ eväät ("eat the skacks") = [syøʔ:evæ:t]
ne eväät ("those snacks") = [neʔevæ:t] ~ [ne.evæ:t]

[KathTheDragon]

but doesn't behind the singers have both intervocalic /h/ and /ŋ/

Eh, /h/ is unambiguously the onset of the second syllable, while /ŋ/ wants to be the coda of the first syllable, causing it to geminate. So, for me /h/ and /ŋ/ are still in complimentary distribution (as /h/ can't ever be geminated)

[din]

You have geminate consonants in your dialect? Aren't you sure it's not just [Ng]?

[Salmoneus]

I assume she'd know if that were the case - /Ng/ in 'singers' is very marked and dialectical. But like you I'm surprised by the idea of a geminate /N/ there. English does have geminates (/long consonants/etc), but normally only at morpheme boundaries - 'pennines' and 'pen-nines' would normally be different, but if she geminates intervocalic nasals they would merge (and the /n/ is in the coda there just as much as the /N/ is in /singer/).

[Grunnen]

but doesn't behind the singers have both intervocalic /h/ and /ŋ/

Eh, /h/ is unambiguously the onset of the second syllable, while /ŋ/ wants to be the coda of the first syllable, causing it to geminate. So, for me /h/ and /ŋ/ are still in complimentary distribution (as /h/ can't ever be geminated)

But wouldn't /h/ normally precede the stressed vowel, while /ŋ/ would follow it? If that's so, they'd still be in complementary distribution. So would anyone know it this is indeed the case?

[KathTheDragon]

I assume she'd know if that were the case - /Ng/ in 'singers' is very marked and dialectical. But like you I'm surprised by the idea of a geminate /N/ there. English does have geminates (/long consonants/etc), but normally only at morpheme boundaries - 'pennines' and 'pen-nines' would normally be different, but if she geminates intervocalic nasals they would merge (and the /n/ is in the coda there just as much as the /N/ is in /singer/).

Well, I'm sure it's not /ŋg/ - I can deliberately produce that, and it's not how I normally talk. What I think is happening is that the sequence /sɪŋəz/ is first being syllabified as /sɪŋ.əz/ due to the general principle that /ŋ/ isn't a valid onset, then the /ŋ/ is being geminated so that we don't have a closed syllable followed by an onsetless syllable, giving /sɪŋ.ŋəz/.

[Salmoneus]

Interesting! That would violate what's normally a fairly basic rule, that /N/ can never be syllable-initial. Do you also do this gemination with other consonants? Eg does 'sinner' have the gemination? 'Setter'? 'Sizzle'? 'Talon'?

'Closed syllable followed by an onsetless syllable' is a very common condition in English - I'm curious how far you've gone to avoid it.

[KathTheDragon]

I think 'sinner' has it, 'sizzle' and 'talon' might, but I'm not sure (less sure that 'talon' has gemination), and 'setter' probably doesn't.

[finlay]

Sure. The classic example is /h/ and /ŋ/ in English.

Are there no places where /h/ and /ŋ/ contrast? Does this mean they could be analyzed as the same phoneme?

I would say that when they occur medially, h only occurs before a stressed syllable (behind, sahara), and ng only after (singer, etc). But I'm pretty sure that this only happens because there just isn't a word that fits the other pattern, not because it's impossible. A marginal case might be names like Dogherty, which is sometimes pronounced with an h sound, post stress.

This, I would say, is the downfall of the whole idea of phonemes - linguists had to add in a third condition that two sounds, if no minimal pairs could be found and they are nominally in complementary distribution, would be considered separate phonemes if they were not "phonetically similar", pretty much just to disclude the idea of h and ng in English being one phoneme. This to me looks like a copout. I think to some extent something like a phoneme must exist, because we know instinctively that two sounds are different, but I don't think this idea is exactly right. However, it beats optimality theory in my opinion.

[Zju]

This, I would say, is the downfall of the whole idea of phonemes - linguists had to add in a third condition that two sounds, if no minimal pairs could be found and they are nominally in complementary distribution, would be considered separate phonemes if they were not "phonetically similar", pretty much just to disclude the idea of h and ng in English being one phoneme. This to me looks like a copout. I think to some extent something like a phoneme must exist, because we know instinctively that two sounds are different, but I don't think this idea is exactly right. However, it beats optimality theory in my opinion.

Interesting. What do you propose?

[Pabappa]

Hmm? there's plenty of English words with post-tonic /h/, even if they are all loans and/or compounds. "Freehand" is typical of compounds and "Baja" of loans. Surely no one would take you seriously if you argued they should be "freengand" and "banga"? Loans are as much a part of English as any other word so you cant just say "well that doesnt count because it's a loan". New Zealand English probably has some words with pre-tonic /ŋ/, but Im not sure if any of them are common nouns as opposed to names.

[zompist]

Pre/post stress doesn't work; an example with ŋ would be "include". (the AHD insists there's an /n/ there, but that's a clear spelling bias to me— they don't claim there's an /n/ in "improper".)

Just to show that this isn't just a quirk of English— another classic phonological puzzle is Mandarin, where the palatals /j q/ are in complementary distribution to three sets of consonants: /z c/, /zh ch/, and /g k/ (using pinyin representations throughout). The palatals occur only before /i ü/; the others occur everywhere else.

[Herr Dunkel]

I'm not sure I've read this correctly (plus it's late) but if I have... are you people seriously proposing that /h ŋ/ are actually one phoneme?

[Grunnen]

I'm not sure I've read this correctly (plus it's late) but if I have... are you people seriously proposing that /h ŋ/ are actually one phoneme?

I didn't have the impression anyone here is trying to argue they are.

[Herr Dunkel]

Vecfaranti introduced the idea (albeit in the form of a question) and I kinda took that the rest was about that. Heh sorry

[zompist]

I'm not sure I've read this correctly (plus it's late) but if I have... are you people seriously proposing that /h ŋ/ are actually one phoneme?

No, that strikes everyone as absurd; the problem is, as finlay said, that there's really no principled reason why that is. I don't think we have to throw away the idea of phonemes, but it's just kind of an embarrassing discrepancy.

[finlay]

Another problem with phonemes is that field linguists often use them for pissing contests, like "my language has fewer phonemes than your language". Or you get people describing languages (e.g. Rotokas, Pirahã) as lacking nasals, when in fact they do nothing of the sort, it's just that they've noted a complementary distribution with voiced plosives and theorized the nasals out of existence. It's a bit more of a prosaic point I guess, but it ties into a wider point that tribes like the Pirahã have been exploited a lot by the European linguists and missionaries that they've dealt with.

[vec]

It's interesting; almost no Icelandic grammar book lists only the minimal phoneme inventory of Icelandic. Several sounds which only occur in complimentary distribution with other sounds are always listed in Icelandic sound tables, such as the one on Wikipedia. The table seen there is half-way between phonemic and phonetic. It lists more sounds than just the "phonemes" but it certainly doesn't list all the allophones. I don't know why this is, but it's become convention.

[Herr Dunkel]

Isn't it supposed to be sth. like:

/p pp b bb t tt d dd k kk g gɡ/
/f v θ (ð) s ss h/
/m mm n nn ŋ/
/r rr l ll j/

/i ɪ ʏ u/
/ɛ œ ɔ/
/a/

/aɪ aʊ ɔʊ ɔʏ ɛɪ/

The allophony is sick tho

Edit: added /ð/ in brackets to account for several compounds, not sure if that works but hey <Alþingi> has a medial /θ/. Does vowel length count as minimally/potentially phonemic as there are things like <alltaf> and <Alþingi> that both have [ɬ] but the first element differs in length as: [a] vs [a:] (according to a certain Ölvir well at least)

[Zaarin]

Pre/post stress doesn't work; an example with ŋ would be "include". (the AHD insists there's an /n/ there, but that's a clear spelling bias to me— they don't claim there's an /n/ in "improper".)

Just to show that this isn't just a quirk of English— another classic phonological puzzle is Mandarin, where the palatals /j q/ are in complementary distribution to three sets of consonants: /z c/, /zh ch/, and /g k/ (using pinyin representations throughout). The palatals occur only before /i ü/; the others occur everywhere else.

Maybe it's dialectal. I definitely have /n/ in "include" with a strong syllable break between the prefix and the root; but I have occasionally heard /ŋ/.

[vec]

Isn't it supposed to be sth. like:

/p pp b bb t tt d dd k kk g gɡ/
/f v θ (ð) s ss h/
/m mm n nn ŋ/
/r rr l ll j/

/i ɪ ʏ u/
/ɛ œ ɔ/
/a/

/aɪ aʊ ɔʊ ɔʏ ɛɪ/

The allophony is sick tho

Edit: added /ð/ in brackets to account for several compounds, not sure if that works but hey <Alþingi> has a medial /θ/. Does vowel length count as minimally/potentially phonemic as there are things like <alltaf> and <Alþingi> that both have [ɬ] but the first element differs in length as: [a] vs [a:] (according to a certain Ölvir well at least)

Hm, I wonder if its necessary for the geminates to be phonemic, but if they are /ff/ is also phonemic. Unvoiced n, /n̥/ is certainly phonemic, and so are unvoiced r /r̥/, unvoiced l /l̥/ or /ɬ/ and unvoiced j /ç/. In my idiolect, unvoiced /ŋ̊/ contrasts with /n̥/ before /tʰ/ so I guess in some cases that is also phonemic. The velar fricatives /x~ɣ/ are definitely phonemic, but they are probably only one phoneme. They might also be allophones of /g/, but it depends on how you analyze the stop system. The non-geminated stops don't contrast intervocalically, but I guess they are phonemic initially. The diphthong <au> is [œy], not [ɔʏ].

I think /þ/ and /ð/ are probably phonemic, with words like Aþena, Maraþon, as well as words like maðkur [maθkʏr̥] 'worm', which could contrast with a word like maðgur [maðkʏr̥] (if one existed). The closest near minimal pair is mæðgur [maiðkʏr̥] 'mother and daughters'. There's no 100% minimal pair though. [ð] might also be an allophone of /d/.