Schwa-schwi merger

[ol bofosh]

["b6I.@s] for me.

I'm discovering that I have a mixed pronunciation of "schwi words". Some are all schwi, or all schwa, or vary between the two. Sometimes I think it's somewhere between [@] and [1].

[Travis B.]

Random question:

If I have only one phoneme for schwa-schwi, with the default realizations of [ə̝] initially, [ɨ̞] medially, and [ə] morpheme-finally, what should I consider its form as a phoneme? Should I call it /ə/ or /ɪ/?

[finlay]

Answer: who cares? Phonemes are just meant to be convenient.

[Rory]

Another question masquerading as an answer: What evidence do you have that it is a phoneme? (What evidence do we have that anything is a phoneme?

[Melteor]

Answer: who cares? Phonemes are just meant to be convenient.

Can we call it a grapheme please?

Another question masquerading as an answer: What evidence do you have that it is a phoneme? (What evidence do we have that anything is a phoneme?

We could just hash everything. (Not that I'm a programmer or anything. No, really, I'm not.)

[Ser]

Random question:

If I have only one phoneme for schwa-schwi, with the default realizations of [ə̝] initially, [ɨ̞] medially, and [ə] morpheme-finally, what should I consider its form as a phoneme? Should I call it /ə/ or /ɪ/?

That's when you go neutral and use hearts instead.

/♥/

[Travis B.]

Another question masquerading as an answer: What evidence do you have that it is a phoneme? (What evidence do we have that anything is a phoneme?

The general evidence I have that it is a single phoneme (I am not getting into a discussion of "well, are phonemes of any sort real in the first place?") is as follows:

I cannot form actual contrasts between [ə] and [ɨ̞], even though I have both phones. Where I do appear to have contrasts can be neatly explained away by invoking environment in the form of morpheme boundaries (e.g. I still contrast roses and Rosa's, but morphology explains that away).

The historical distribution of (unstressed) /ə/ versus /ɪ/ has been completely scrambled, resulting in a new distribution that includes [ə], [ə̝]*, [ɨ̞]**, and syllabification/vocalization of the following consonant as possibilities which are only predictable from environment (including morpheme boundaries) and which obscures their original distribution.

Consequently I found I have no intuitive sense of what the historical distribution of (unstressed) /ə/ versus /ɪ/ was in the first place; places in which I seem to have an idea of how these were distributed seem to be more due to the influence of orthography.

I have never gotten around to forming a full table of all the different possibilities, because it would be rather large to cover them exhaustively, and it would be hard finding example words for many of the theoretical possiblities. However, there seem to be a number of general rules (not necessarily exhaustive or including corner cases), ordered in terms of general priority.

{C, 0}V{p, b, w} takes [ə]
{C, 0}Vm takes [ə] or syllabification of the following consonant
(C, 0)V{r, l} where the {r, l} falls on a syllable coda takes syllabification/vocalization of the following consonant
{C. 0}V{r, l} where the {r, l} falls on a syllable onset takes [ə] or syllabification/vocalization of the following consonant
{C, 0}V# where that preceding the boundary is not a prefix takes [ə]
rV{f, t, d, s, z, tʃ, dʒ, k, ɡ, ŋ} takes [ə̝]
rVn takes [ə̝] or syllabification of the following consonant
rVv takes [ə]
lV{f, t, d, s, z, tʃ, dʒ, k, ɡ, ŋ} takes [ə̝]
lVn takes [ə̝] or syllabification of the following consonant
lVv takes [ə]
CV{f, v, t, d, s, z, tʃ, dʒ, k, ɡ, ŋ} takes [ɨ̞]
tVn takes syllabification of the following consonant
CVn takes [ɨ̞] or syllabification of the following consonant
Initial V{f, v} takes [ə]
Initial V{t, d, s, z, tʃ, dʒ, k, ɡ, ŋ} takes [ə̝]
Initial Vn takes [ɨ̞] or syllabification of the following consonant

This might need some corrections, but this will give you a general idea of the overall distribution at work here.

* Close-mid central unrounded vowel, can also be called [ë]
** Near-close central unrounded vowel, can also be called [ɪ̈]

[Bristel]

I don't have the merger, interesting.

[Rory]

I am not getting into a discussion of "well, are phonemes of any sort real in the first place?"

That's probably a good idea.

I cannot form actual contrasts between [ə] and [ɨ̞], even though I have both phones.

So, another question: how do you know that those are different phones? By which I mean, of course I am sure that these two symbols you have written here represent two sounds. But what evidence do you have that each of the sounds that comes out of your mouth can be neatly categorized into one of these two discrete categories, rather than lying on some continuous cline?

(I want to be clear that I'm not trying to prove any particular point or anything like that; I'm just posing questions to make you consider your assumptions, if you haven't already.)

[... some distributions ...]

This might need some corrections, but this will give you a general idea of the overall distribution at work here.

I might have misread, but I don't see any evidence that they're actually phonemically distinct - you just listed some (very comprehensive) allophony rules. So maybe we don't need to worry about different symbols?

[Rory]

Random question:

If I have only one phoneme for schwa-schwi, with the default realizations of [ə̝] initially, [ɨ̞] medially, and [ə] morpheme-finally, what should I consider its form as a phoneme? Should I call it /ə/ or /ɪ/?

That's when you go neutral and use hearts instead.

/♥/

If that kind of reasoning/symbology appeals to you, you should read Hale and Reiss's Phonological Enterprise (digital copy of the uncorrected proofs available here).

[kodé]

Random question:

If I have only one phoneme for schwa-schwi, with the default realizations of [ə̝] initially, [ɨ̞] medially, and [ə] morpheme-finally, what should I consider its form as a phoneme? Should I call it /ə/ or /ɪ/?

That's when you go neutral and use hearts instead.

/♥/

If that kind of reasoning/symbology appeals to you, you should read Hale and Reiss's Phonological Enterprise (digital copy of the uncorrected proofs available here).

I have that sitting on my bookshelf right now, waiting for the unlikely possibility that I will have time to read it in the next three years...

[Ser]

Random question:

If I have only one phoneme for schwa-schwi, with the default realizations of [ə̝] initially, [ɨ̞] medially, and [ə] morpheme-finally, what should I consider its form as a phoneme? Should I call it /ə/ or /ɪ/?

That's when you go neutral and use hearts instead.

/♥/

If that kind of reasoning/symbology appeals to you, you should read Hale and Reiss's Phonological Enterprise (digital copy of the uncorrected proofs available here).

In order to present the resulting phonological inventory, we must confront
a representational difficulty (one which is quite significant in the context of
this chapter). It is common in phonological circles to use the symbols of the
International Phonetic Alphabet (IPA) in a systematically ambiguous manner,
and, indeed, for the most part we have followed that practice so far in this
book. On the one hand, a symbol such as i is used to represent a bundle
of features (both at the start of a derivation—i.e. features in an underlying
representation—and at the end of a derivation, i.e. in a “phonetic” representation);
on the other, this same symbol is used to represent the acoustic
impression (or articulatory realization) of a segment. Since issues in these
domains form a central concern of this chapter, and since in our view the
constantly shifting use of a single symbol for these three distinct purposes
(“phonological” input representations, “phonetic” output representations,
and phonetic realizations) has created some confusion in the literature, we
must try to clarify the matter. The highest vowel of Marshallese, in underlying
representations, is neither front nor back, round nor unround, so there is in
fact no appropriate IPA symbol which may be used to represent it. In order
to keep this clear in the mind of the reader, we will use arbitrary non-IPA
symbols for each of the Marshallese underlying vowels: the [+hi,+ATR] vowel
will be indicated by , the [+hi,−ATR] vowel by ☎, the [−hi +ATR] vowel by
☯, and the [−hi,−ATR] vowel by .

A coffee mug for the high +ATR vowel, a telephone for the high -ATR vowel, a yin-yang symbol for the low +ATR vowel, and a soccer ball for the low -ATR vowel.

theyactuallypublishedthisinallseriousness seriouslyyoucan'tbeserious

This is fukken awesome.

[Corundum]

I cannot form actual contrasts between [ə] and [ɨ̞], even though I have both phones. Where I do appear to have contrasts can be neatly explained away by invoking environment in the form of morpheme boundaries (e.g. I still contrast roses and Rosa's, but morphology explains that away).

So, how is this morpheme boundary pronounced? What phonetic aspect of the morpheme boundary affects the vowel?

[Travis B.]

I cannot form actual contrasts between [ə] and [ɨ̞], even though I have both phones. Where I do appear to have contrasts can be neatly explained away by invoking environment in the form of morpheme boundaries (e.g. I still contrast roses and Rosa's, but morphology explains that away).

So, how is this morpheme boundary pronounced? What phonetic aspect of the morpheme boundary affects the vowel?

Roses has [ɨ̞] after a morpheme boundary, following the typical rules for a CVz syllable as specified above; Rosa's has [ə] before a morpheme boundary, where the preceding morpheme is not a prefix, as specified above.

To my knowledge there is nothing phonetic about the morpheme boundary that affects the realized vowel; it is just an arbitrary morphology-sensitive phonological rule.

[Travis B.]

Random question:

If I have only one phoneme for schwa-schwi, with the default realizations of [ə̝] initially, [ɨ̞] medially, and [ə] morpheme-finally, what should I consider its form as a phoneme? Should I call it /ə/ or /ɪ/?

That's when you go neutral and use hearts instead.

/♥/

That is a nice idea, but the problem is that the X-SAMPA (and CXS) to IPA converter I use does not spit out hearts, and I would much rather not dig through a huge Unicode character chart to look for a heart whenever I want one and do not have one to copy-and-paste on hand...

[Melteor]

Random question:

If I have only one phoneme for schwa-schwi, with the default realizations of [ə̝] initially, [ɨ̞] medially, and [ə] morpheme-finally, what should I consider its form as a phoneme? Should I call it /ə/ or /ɪ/?

That's when you go neutral and use hearts instead.

/♥/

That is a nice idea, but the problem is that the X-SAMPA (and CXS) to IPA converter I use does not spit out hearts, and I would much rather not dig through a huge Unicode character chart to look for a heart whenever I want one and do not have one to copy-and-paste on hand...

Then just use $&@+€£¥ etc.

[Travis B.]

From thinking about it, I will probably use "/ə/", simply because it feels easier to say that it includes [ɨ̞] as a realization than it is to say that "/ɪ/" includes [ə] as a realization (even though it could, of course), and also because "/ə/" has already been used elsewhere to denote vowels that are not [ə], e.g. in French.

It still feels weird to mark -tion, for instance, as -/ʃən/, though, when I have been so used to marking it as -/ʃɪn/...

[kodé]

A coffee mug for the high +ATR vowel, a telephone for the high -ATR vowel, a yin-yang symbol for the low +ATR vowel, and a soccer ball for the low -ATR vowel.

theyactuallypublishedthisinallseriousness seriouslyyoucan'tbeserious

This is fukken awesome.

Hale and Reiss are phonological renegades. If OT is the Galactic Empire, these guys are Luke and Han. I went to a conference they hosted once, which was sort of like a meeting of the Rebel Alliance. There were some cuh-razy ideas there, let me tell you...

[Nortaneous]

This is fukken awesome.

Yeah, until you try to read the thing and forget for the fifteenth time which height value is coffee mug and which one is soccer ball because they tried to be ~cool~ instead of just using a ɜ ɘ ɨ like any reasonable person would. Shit, that would be less brain-dead than the convention for Caucasian VVS langs (where they write the mid and high vowels as a and ə respectively because they like being difficult) and the convention for Caucasian VVS langs is not particularly difficult to read. Not even the Soviets go out of their way to be difficult! Until they're designing writing systems. (Why didn't anyone tell Molodtsov that Cyrillic -- Cyrillic of all systems! -- does not actually need radical expansions to support palatalized consonants?)

[Ser]

This is fukken awesome.

Yeah, until you try to read the thing and forget for the fifteenth time which height value is coffee mug and which one is soccer ball because they tried to be ~cool~ instead of just using a ɜ ɘ ɨ like any reasonable person would. Shit, that would be less brain-dead than the convention for Caucasian VVS langs (where they write the mid and high vowels as a and ə respectively because they like being difficult) and the convention for Caucasian VVS langs is not particularly difficult to read. Not even the Soviets go out of their way to be difficult! Until they're designing writing systems. (Why didn't anyone tell Molodtsov that Cyrillic -- Cyrillic of all systems! -- does not actually need radical expansions to support palatalized consonants?)

Oh, c'mon, gimme a break. You can't deny these transcriptions are just beautiful.

The front–back position of the tongue and the degree of lip-rounding
during the duration of the vowel are simply “transitions” from the values of
the adjacent consonants (Choi 1992), not unlike the transitions from e.g. i
to g in English “league” (such transitions exist in any VC or CV sequence,
in any human language, of course). This can be seen from the derivations
in (61a–c).

(61)

1. C_lightVC_round: /nʲkʷnʲkʷ/ > [nʲkʷnʲkʷ] > ♂nʲi͜ukʷnʲi͜ukʷ♀ ‘clothing’
2. C_lightVC_heavy: /nʲ☯tᵚ/ > [nʲ☯tᵚ] > ♂nʲe͜ʌtᵚ♀ ‘clothing’
3. C_lightVC_light: /tʲtʲ/ > [tʲtʲ] > ♂tʲɛtʲ♀ ‘Lutjanus Flavipes’

Examples are given in (61) for vowels left-flanked by light consonants—
parallel examples for vowels with heavy and round consonants on their left
can easily be constructed (for a systematic list, see Choi 1992: 30).

This analysis not only accounts for the typologically bizarre “tied” vowels
of Marshallese but also provides interesting insights regarding the “normal”
Marshallese surface vowels, such as i and o. Although these vowels sound
like the vowels usually designated [‏i] and [o] in the phonetics literature, they
have a decidedly different status with respect to the phonological system visà-
vis the similar vowels of e.g. English. For example, the Marshallese word for
“bark” (of a dog) is derived as follows: /rʷ☯rʷ/ > [rʷ☯rʷ] > ♂rʷorʷ♀. By
contrast, the English word roar shows the following derivation: /ror/ > [ror]
> ♂rʷorʷ♀. The vowels present in the bodily output are roughly the same in the
English and Marshallese examples. The English examples, however, represent
the bodily realization of phonetic representations for which the front–back
position of the tongue and the degree of lip-rounding are specified. In the
Marshallese examples, by contrast, the bodily output o is the result of the
same “transition” phenomenon found with the odd “tied” vowels—it just
happens to be the case in this instance that the transitions are, for the example
of o, from a back and round articulation to a back and round articulation.
Such a “transition” gives rise to the mirage of a steady-state vowel in the
output

[finlay]

I'm a bit skeptical of this marshallese claim, if only because to me it feels obvious that you could produce a similar transcription, where the rounding and frontness information is transferred from the vowels to the consonants, for any language. There were some transcriptions on wikipedia which sounded nothing like the real thing because they'd obsessively inserted offglides for all the vowels.

Also, is OT really the mainstream now? I mean I know I'm just basically out of that loop but from my casual thought experiments it has never seemed supportable.

[gach]

I'm a bit skeptical of this marshallese claim, if only because to me it feels obvious that you could produce a similar transcription, where the rounding and frontness information is transferred from the vowels to the consonants, for any language.

Unless you can do tests where changing the consonantal environment of the vowels (eg. by adding or removing affixes) productively changes the vowels' surface values rather than the other way around. Of course, if such cases are not very typical, the speakers' mental analysis of the sound system is certainly unstable. I think it's in Ludian where previous palatalisation of consonants in front vocalic environments has changed in the speakers' minds into allophonic vowel fronting when adjacent to phonemically palatalised consonants (transfer of palatality from vowels to consonants).

[Rory]

Also, is OT really the mainstream now? I mean I know I'm just basically out of that loop but from my casual thought experiments it has never seemed supportable.

It's been the mainstream for a long while. Everyone acknowledges that (classical) OT doesn't work, but the majority opinion is that it's better than the previous formalisms. ("All models are wrong; some models are useful.") If you go to a phonology conference, the presentations using OT will probably be in the majority. There are also a lot of developments on OT that allegedly improve on the classical formulation - such as Stratal OT, Harmonic Serialism, or Maximum-Entropy Weighted Stochastic OT. Personally, I'm not convinced. (Nor am I convinced by Hale & Reiss, either.)

[Corundum]

I cannot form actual contrasts between [ə] and [ɨ̞], even though I have both phones. Where I do appear to have contrasts can be neatly explained away by invoking environment in the form of morpheme boundaries (e.g. I still contrast roses and Rosa's, but morphology explains that away).

So, how is this morpheme boundary pronounced? What phonetic aspect of the morpheme boundary affects the vowel?

Roses has [ɨ̞] after a morpheme boundary, following the typical rules for a CVz syllable as specified above; Rosa's has [ə] before a morpheme boundary, where the preceding morpheme is not a prefix, as specified above.

To my knowledge there is nothing phonetic about the morpheme boundary that affects the realized vowel; it is just an arbitrary morphology-sensitive phonological rule.

My point was that using morpheme boundaries as condition for allophony makes as much sense as using semantics as condition. You have a minimal pair where [ɨ̞] and [ə] contrast. Maybe [ɨ̞] and [ə] belong to the same archphoneme or something, but they are different phonemes in my book.

[Travis B.]

I cannot form actual contrasts between [ə] and [ɨ̞], even though I have both phones. Where I do appear to have contrasts can be neatly explained away by invoking environment in the form of morpheme boundaries (e.g. I still contrast roses and Rosa's, but morphology explains that away).

So, how is this morpheme boundary pronounced? What phonetic aspect of the morpheme boundary affects the vowel?

Roses has [ɨ̞] after a morpheme boundary, following the typical rules for a CVz syllable as specified above; Rosa's has [ə] before a morpheme boundary, where the preceding morpheme is not a prefix, as specified above.

To my knowledge there is nothing phonetic about the morpheme boundary that affects the realized vowel; it is just an arbitrary morphology-sensitive phonological rule.

My point was that using morpheme boundaries as condition for allophony makes as much sense as using semantics as condition. You have a minimal pair where [ɨ̞] and [ə] contrast. Maybe [ɨ̞] and [ə] belong to the same archphoneme or something, but they are different phonemes in my book.

Morpheme boundaries having phonological effects is not controversial. One good example I can think of is apparent vowel length contrasts in Scots and Scottish English which are actually conditioned by morpheme boundaries (in addition to environment), which are best analyzed as lacking phonemic vowel length.