Allophony and Orthography

[Hlewagastiz]

Normally, an allophone of a given phoneme is a special "version" of the phoneme in question under special (but regular) circumstances of phonemic environment; e.g. the k-sound heard in, say, kilt slightly differs from the k-sound heard in, say, cone; the difference lies in the fact that the former is slightly "palatalized", due to the front vowel following it.
Now, the question is the following: does the way a word is written (i.e. its orthography) affects the decision on whether a "sound" is considered to be an "allophone" (of a given "phoneme") or an independent "phoneme"?
I'll take an example from a language which uses "historical" orthography, in particular Modern Greek.
Modern Greek speakers say /'mila/ = apples; speak! (imperative); they also say /'miλa/ = miles (I know, the "λ" should be reversed... In any case, it stands here for a palatalized version of /l/). Those lexemes are respectively written "μήλα" (or "μίλα!") and "μίλια". Generally, Modern Greek linguists agree that the said palatalized /l/ sound is an allophone of the usual Greek /l/ rather than an independent "phoneme". However, I think this judgement is based on the way this palatalized sound is written (i.e. it is always followed by an "iota" or some similarly pronounced phoneme); the main characteristic of an "allophone" is that its mere presence in a given lexeme is not sufficient to change the meaning of the latter (i.e. to create a new lexeme, a new word, simply put). But doesn't this /λ/ changes the meaning in the above examples, supposed that Modern Greek were not written in a historical orthography system (that is if the above words were written without the "iota") ?
If anyone has examples from other languages, I'd like to know...

[Radagast]

If there are minimal pairs then they cannot be construed as allophonic, at least not in that particular dialect.

In the example that you give it seems that there could be two interpretations however that would lead to two differeing phoneme inventories.
One is interpreting only one l, but an i that is interchangeable with j in certain positions, then we would have not a palatalised l phoneme but sequence of l followed by j, which is auditively almost indistinguishable from a palatalised l. The interpretation would thus be /milia/ but with the i pronounced as j, which is probably a regular allophone.

The other interpretation would be interpreting two l phonemes, a standard l and a palatalised one. Thus eading to the phonemic interpretation /miλa/.

I suppose that in Greek there are historical reasons for prefering the first explanations, the i sound may be known fom earlier stages in teh languages development and also it is generally preferred to have as few phonemes as possible, particularly when they can be explained like this.

Segmentation is always a very difficult task in phoneme analyses, and poses many interesting problems and quetions in an language. Affricates for example can almost always be interpreted both as a sequence of two phonemes or as a single unit with dual articulation, the histpry of the language and the phonotax of the language in question can give hints to which interpretation is preferable.

Orthography does not generally have anything to do with the phonemisation of a language. However often the orthography because it is generally conservative gives hints as to language history, and thus may reflect earlier stages where the phonemic system may be more transparent. (Sometimes also more obscure.)

[Soren]

The normal position is that a phonemic analysis should be based only on the phonetic data, without appeal to orthography or the historical development of the language.

Your examples would normally establish contrast between those phonemes. Unless you take a generative phonology approach, where you would analyse "μίλια" as /milia/ or /milja/ or something and then have a rule generating the phonetic realisation from that.

An example in Japanese. The phones [s] and [S] can contrast iwhen preceding /M/, /O/, and /a/, but before /i/ only [S] appears, and before /e/ only [s] appears. Traditionally though, [S] is considered an allophone of /s/ before , and where [S] appears elsewhere, it is analysed as /sj/, with a generative rule saying that /sj/ is [S] phonetically. So we have:

/si/ - [Si]
/sM/ - [sM]
/sjM/ - [SM]
etc.

This analysis exactly mirrors the way that these syllables are written in Kana script. To me, it looks very much like a case of letting orthography interfere with an analysis. An alternative would be to posit phonemic /s/ and /S/, and say that contrast is lost before /i/, where only /S/ appears. As far as I can see, this captures the modern phonetic data just as well.

So I suppose that the answer is that although a phonemic analysis should in theory be based on the phonetic data, some linguists do make appeals to orthography or to the historical development of the language. I suppose it is a matter of opinion which kind of analysis you want to take.

Incidentally, the reversed lambda in IPA isn't a palatalised /l/, but a palatal lateral approximant. The former has a secondary articulation at the palate, the latter has its primary articulation at the palate. Confusingly though, the verb "to palatalise" is often used for "to become palatal" as well as "to become palatalised".

[Radagast]

I disagree with that. I think that interpreting two phonemes is the best solution to that example. A similar example in Danish exist in words like "sjov" (fun) (with minimal pair like "sov" (slept)) pronounced with an initial palatal sibilant, this is normally phonematised as four segments while it could easily be interpreted as three. The reason is partly that there are other initial consonant clusters with consonant j, like kj, tj etc. we would then have to introduce an entire palatalised series of consonants if we wanted to interpret it that way. However it is easier and more economic to interpret them as clusters. Also there are historical arguments for not seeing them as one phoneme.

I also disagree with not letting historical knowledge of a language influence phonemisation when it is possible, it is a very valid way of basing your interpretation on something tangible.

It must be remembered that phoneme inventories and phonemic analyses are not definitive. They are always based on interpretations and there can be several valid interpretations of phonemic inventories for any language.

[gsandi]

Normally, an allophone of a given phoneme is a special "version" of the phoneme in question under special (but regular) circumstances of phonemic environment; e.g. the k-sound heard in, say, kilt slightly differs from the k-sound heard in, say, cone; the difference lies in the fact that the former is slightly "palatalized", due to the front vowel following it.
Now, the question is the following: does the way a word is written (i.e. its orthography) affects the decision on whether a "sound" is considered to be an "allophone" (of a given "phoneme") or an independent "phoneme"?
I'll take an example from a language which uses "historical" orthography, in particular Modern Greek.
Modern Greek speakers say /'mila/ = apples; speak! (imperative); they also say /'miλa/ = miles (I know, the "λ" should be reversed... In any case, it stands here for a palatalized version of /l/). Those lexemes are respectively written "μήλα" (or "μίλα!") and "μίλια". Generally, Modern Greek linguists agree that the said palatalized /l/ sound is an allophone of the usual Greek /l/ rather than an independent "phoneme". However, I think this judgement is based on the way this palatalized sound is written (i.e. it is always followed by an "iota" or some similarly pronounced phoneme); the main characteristic of an "allophone" is that its mere presence in a given lexeme is not sufficient to change the meaning of the latter (i.e. to create a new lexeme, a new word, simply put). But doesn't this /λ/ changes the meaning in the above examples, supposed that Modern Greek were not written in a historical orthography system (that is if the above words were written without the "iota") ?
If anyone has examples from other languages, I'd like to know...

This is a very good point, Hlewagastiz, and I for one agree with you. Orthography can influence the phonemic system one develops for a language - but this does not bother me, because I have always argued that there is not one phonemic system for a language, but potentially quite a few. Any phonemic system that is consistent with the outcome (the real language we are trying to analyze) is as valid as any other - although one may be more elegant or easier to use, of course.

Japanese is a good example. The Hepburn and Kunreishiki systems of transliteration approximate to a perfect phonemic system for standard Japanese equally well, but it's the Kunreishiki system that has an almost one-to-one relationship to the Japanese kana syllabaries, so it's not surprising that the Japanese government has favoured it over Hepburn.

The differences between the two are quite marked. The same sequence of sounds can be considered as a sequence of two units (ch-a) in Hepburn, and of three units (t-y-a) in Kunreishiki (and kana). The contrast between shi and sa in Hepburn is reflected both as contrastive consonants and vowels, while in Kunreishiki there is a contrast only in the vowel: si / sa.

-----------------------------

Which is why I and Chomskyites will never get on together...

[Soren]

I disagree with that. I think that interpreting two phonemes is the best solution to that example. A similar example in Danish exist in words like "sjov" (fun) (with minimal pair like "sov" (slept)) pronounced with an initial palatal sibilant, this is normally phonematised as four segments while it could easily be interpreted as three. The reason is partly that there are other initial consonant clusters with consonant j, like kj, tj etc. we would then have to introduce an entire palatalised series of consonants if we wanted to interpret it that way. However it is easier and more economic to interpret them as clusters. Also there are historical arguments for not seeing them as one phoneme.

Good point...the traditional Japanese analysis does reduce the number of phonemes by removing /S/, /tS/, /dZ/.

I also disagree with not letting historical knowledge of a language influence phonemisation when it is possible, it is a very valid way of basing your interpretation on something tangible.

Interesting to hear that. When I first was learning phonetics, I read in quite a few places that an analysis should be based only on the phonetic data. I used to think that this was absurd, because ignoring historical development would often lose some interesting generalities. Evidently I wasn't wrong.

I remember posting about this here a long time back. I was looking at Norwegian words like "gi", and considering whether to analyse it as /ji/ or /gi/. The latter analysis would capture the generality of [g] not appearing before /i/, because [g]>[j] in this environment. But it does require analysing [j] in the phonetic data differently depending on its history. Would you say this latter analysis is OK then?

[Radagast]

I just think that since phonemisation is not an exact science there is no reason to not allow oneself to take hints from where they might come. When one proposes a phonemic inventory for a language, arguments must be used to give weight to ones interpretation. If we know that in related languages there are arguments to be found I don't see why we shouldn't use them, as long as they don't contradict the phonetic level of course.

[Nae]

I feel that a linguist should analyse a languages' phonemic structure by, yes, looking at the phonetic reality in question, and interpret it that way, and not ignore the historical development that has occured in that language (which the "calling 'shu' 'syu'"-thing IS, imho)... Of course, analysing a language by it's historical structure (and its contemporary structure) works in some cases, but not by ignoring the last step in the development of the phonemic system.



While laymen (like language teachers and whatnot) can do whatever they please. Which is yet another thing for linguists to study.

[Radagast]

I don't like the latter analysis. Mainly because there is evidence in the same language that the [j] sound in [ji] "give" is different from the [j] sound in [ja] "yes" because when the verb /gi/ is put in the past tense it is pronounced [ga] "gave" and not *[ja]. There is a differentiation between the two phonemes right there: there is clearly both a [g] and a [j] sound only that the contrast beteen them is neutralised before i. When furthermore we know that it comes from a germanic root with g, analysing two phonemes is really rather well founded.

[chris-gr]

[quote="Hlewagastiz"]Modern Greek speakers say /'mila/ = apples; speak! (imperative); they also say /'miλa/ = miles (I know, the "λ" should be reversed... In any case, it stands here for a palatalized version of /l/). Those lexemes are respectively written "μήλα" (or "μίλα!") and "μίλια"./quote]

Since your examples were culled form the greek lexicon, you should remember that in practice there's no problem:
If you pronounced /'milia/ (3 syllables) instead of the correct /'miλa/ (2 syllables) (=miles), you would be understood; it would sound funny, but it would be intelligible.
I agree with the rest.

[Nae]

I don't like the latter analysis. Mainly because there is evidence in the same language that the [j] sound in [ji] "give" is different from the [j] sound in [ja] "yes" because when the verb /gi/ is put in the past tense it is pronounced [ga] "gave" and not *[ja]. There is a differentiation between the two phonemes right there: there is clearly both a [g] and a [j] sound only that the contrast beteen them is neutralised before i. When furthermore we know that it comes from a germanic root with g, analysing two phonemes is really rather well founded.

So, I should analyse Finnish <vesi> as <vete> because the < i > is only an allophone of <e> at the end of a word, and <s> is only an allophone of <t> before word-final <e>? Oh, ok. Sure. That sounds logical.






sarcasm

[Hlewagastiz]

I agree almost with everything stated about the topic (especially with gsandi and radagast). What Narddyr said about the method a linguist should use is true, in any case. As of my examples taken from Modern Greek, I'd rather choose the interpretation of [λ] being attributed phonemic value, especially if one takes into consideration that not only is [λ] never uttered as [l+j] (as is the case in Scandinavian tongues), but it has become a single [j] in the Cypriot dialect of Modern Greek. I think this could be suggesting...

[Radagast]

So, I should analyse Finnish <vesi> as <vete> because the < i > is only an allophone of <e> at the end of a word, and <s> is only an allophone of <t> before word-final <e>? Oh, ok. Sure. That sounds logical.

I didn't comment on Finnish because I know nothing about it, I only stated what is the case in the germanic languages that I know well, so your sarcasm doesn't really affect me. I do however think that you probably misunderstood my explanation. If you try and phrase your apparent disbelief in my analysis as a question based on the situation in finnish then maybe I could explain to you what I meant. Alternatively you could explain to me why the model of analysis that I proposed for Norwegian cannot be used in Finnish.

[Soren]

I don't like the latter analysis. Mainly because there is evidence in the same language that the [j] sound in [ji] "give" is different from the [j] sound in [ja] "yes" because when the verb /gi/ is put in the past tense it is pronounced [ga] "gave" and not *[ja]. There is a differentiation between the two phonemes right there: there is clearly both a [g] and a [j] sound only that the contrast beteen them is neutralised before i. When furthermore we know that it comes from a germanic root with g, analysing two phonemes is really rather well founded.

I don't quite follow. Your discussion here suggests to me that you would prefer to analyse the [j] in "gi" differently from that of "ja" (i.e. analyse "gi" as /gi/), but you also said that you don't like this analysis. Am I misunderstanding something?

[Radagast]

To S?ren: I must have been unclear, I apologise for that.

I meant that I would not advocate an analysis in Norwegian that proposed that the first sound in [ji] be analysed phonemically as /j/. I would however say that this [j] sound is merely an allophone of /g/ which is found before /i/ and which happens to coincide phonetically with another phoneme in the language namely /j/.

(finds own mistake: I wrote latter when I meant first. Sorry again. Basically I should have written "Yes, the latter analysis is dead on in my opinion")

[Hlewagastiz]

This is interesting, Radagast. You're suggesting two functions for a single sound, depending on the context. My personal opinion is that one could use such a kind of distinction. It seems to work, at a first glance...

[Radagast]

It is not really exotic at all. It is merely that sounds and phonemes are two different things.

Written like rules it would look like this:

/g/> [g]/_V, with exception > [j]/_/i/
/j/ > [j]/_V

I don't understand why this seems so outlandish.

What I am suggesting is not two functions for a single sound, it is two phonemes a that are realised as the same sound in some enviroments.

[Soren]

To S?ren: I must have been unclear, I apologise for that.

I meant that I would not advocate an analysis in Norwegian that proposed that the first sound in [ji] be analysed phonemically as /j/. I would however say that this [j] sound is merely an allophone of /g/ which is found before /i/ and which happens to coincide phonetically with another phoneme in the language namely /j/.

(finds own mistake: I wrote latter when I meant first. Sorry again. Basically I should have written "Yes, the latter analysis is dead on in my opinion")

thanks

[Hlewagastiz]

No, it doesn't seem outlandish at all (at least to me ). I just didn't had in mind such a distinction...

[Radagast]

Ahh. I've found a way to clarify myself.

In Norwegian we can se that we have two different phonemes synchronically because we have one [j] sound that alternates with [g] before some vowels, and another [j] sound that doesn't. That leads the cunning linguist to think "what causes this alternation?" Then he looks for rules and finds out that in fact all the [j]'s that alternate with [g] are found exclusively before . Then we have a rule /g/ > [j] _i. The reverse rule is much more complex because that would be some [j]'s are realised as [g] before all other vowels than /i/. Clearly the first rule is simpler and therefore preferable.

[Nae]

So, I should analyse Finnish <vesi> as <vete> because the < i > is only an allophone of <e> at the end of a word, and <s> is only an allophone of <t> before word-final <e>? Oh, ok. Sure. That sounds logical.

I didn't comment on Finnish because I know nothing about it, I only stated what is the case in the germanic languages that I know well, so your sarcasm doesn't really affect me. I do however think that you probably misunderstood my explanation. If you try and phrase your apparent disbelief in my analysis as a question based on the situation in finnish then maybe I could explain to you what I meant. Alternatively you could explain to me why the model of analysis that I proposed for Norwegian cannot be used in Finnish.

Simple: Finnish went through a sound-change that changed word-final /e/ into /i/, and /ti/ into /si/, so the nominative form of *vete > /Pesi/... But the /te/ survived in other cases and forms of the word, for example /vete:n/. You wanted to analyse a phonemic /ji/ as a /gi/ for phonohistorical reasons (or, rather, as /j2i/). That doesn't really work because there is nothing else to indicate the historical relationship than the different conjugational form, and because if phonological change would accumulate into sets of identical phonemes instead of merging with new phonemes, the human mind would overload and explode. And Occam's razor and all that.

And on a side-note, my sarcastic example was a bit uncomplete, now that I think of it - instead of claiming that <vesi> should be analysed as <vete>, I should have claimed that the /s/ and /i/ should be analysed as different phonemes than those other /s/es and /i/s.

[Miekko]

So, I should analyse Finnish <vesi> as <vete> because the < i > is only an allophone of <e> at the end of a word, and <s> is only an allophone of <t> before word-final <e>? Oh, ok. Sure. That sounds logical.

I didn't comment on Finnish because I know nothing about it, I only stated what is the case in the germanic languages that I know well, so your sarcasm doesn't really affect me. I do however think that you probably misunderstood my explanation. If you try and phrase your apparent disbelief in my analysis as a question based on the situation in finnish then maybe I could explain to you what I meant. Alternatively you could explain to me why the model of analysis that I proposed for Norwegian cannot be used in Finnish.

His example is actually pretty good. s <> t is a regular shift in several words, as is i <> e in that position. That's why you in Finnish get vesi > veden > vett? .

Similarly, in Germanic, /g/ > /j/ is pretty common, and vowel shifts condition it a bit:
giva (/jivA/)
gav (/gAv/).
IMHO, the g/j shift is encoded in a different way. I think (producing) allophony is closer to the 'performance' level, while morphophonetic shifts are (originally performance-related stuff that has been lifted to the) competence level.

[Radagast]

You wanted to analyse a phonemic /ji/ as a /gi/ for phonohistorical reasons (or, rather, as /j2i/).

No, not really I wasn't using phonohistorical reasons, solely synchronic alternation patterns. And the rule that you seem to imply that i am implying is the hypercomplex rule, which, as you say, Occams razor would cut into shreds in an instance. The reverse rule is the case: all g's become j's sometimes (and we can account for which times namely before i). Not that some j's are actually different j's than the other j's because they become g's before all other vowels than i.

It does work and it is, I believe, the currently accepted phonemisation of Norwegian. (i might be wrong about that)

[Nae]

You wanted to analyse a phonemic /ji/ as a /gi/ for phonohistorical reasons (or, rather, as /j2i/).

No, not really I wasn't using phonohistorical reasons, solely synchronic alternation patterns. And the rule that you seem to imply that i am implying is the hypercomplex rule, which, as you say, Occams razor would cut into shreds in an instance. The reverse rule is the case: all g's become j's sometimes (and we can account for which times namely before i). Not that some j's are actually different j's than the other j's because they become g's before all other vowels than i.

It does work and it is, I believe, the currently accepted phonemisation of Norwegian. (i might be wrong about that)

What I tried to imply that you implied was that you implied that some /ji/ are an allophone of /gi/.

[Radagast]

His example is actually pretty good. s <> t is a regular shift in several words, as is i <> e in that position. That's why you in Finnish get vesi > veden > vett? .

Similarly, in Germanic, /g/ > /j/ is pretty common, and vowel shifts condition it a bit:
giva (/jivA/)
gav (/gAv/).
IMHO, the g/j shift is encoded in a different way. I think (producing) allophony is closer to the 'performance' level, while morphophonetic shifts are (originally performance-related stuff that has been lifted to the) competence level.

True. It is difficult to discern allophones from obsolete morphophonetic patterns. I would agree with you that the allophony is characterised by being still productive, and thus on the performance level, while the latter is not.