Do phonemes exist? (Travis bait)

[zompist]

This has come up a few times, and I'm interested in having a discussion about it. Here's a posting from the Eddy thread where Travis mentions his doubts about phonemes:

Or do you consciously think about all the realizations of phonemes (to use a parallel example) in your speech?

Sometimes. I'm sure Travis probably does.

Actually, I do not. I am generally consciously aware of what phones I am realizing when I choose to be, but I am not necessarily consciously aware of what phonemes they are supposed to be except when deliberately pronouncing words as in a language variety I do not natively speak. (I should note that I highly doubt the reality of purported phonemic awareness in individuals' native speech varieties, mine own included.)

This does make for a lot of interesting realizations, though, when I do try to analyze how I actually pronounce things versus what they "should" be in conventional phonemic analyses of North American English, precisely because my own pronunciations do not necessarily match what they "should" be from what I know about NAE diachronics, showing the very lack of intrinsic phonemic awareness and incompleteness of both conventional analyses and my own analyses.

For instance, my normal everyday pronunciation of starting to, [ˈs̻t̻ʌ̂ʁˤnːə(ː)] (1) actually baffled me for a long time, as I had no idea why it was so despite it being my natural pronunciation thereof. Particularly, I was confused as to why there was a falling pitch accent and a non-syllabic long nasal present; diachronically I expected *[ˈs̻t̻ʌʁˤ(ɾ̥)ɨ̃nə(ː)] (2) or *[ˈs̻t̻ʌʁˤ(ʔ)n̩ə(ː)] (3) therefor, with neither a falling pitch accent nor a non-syllabic long nasal, but I clearly did not actually normally pronounce it these ways unless I thought about it. Eventually I figured out that it must have undergone the following series of changes: [ˈs̻t̻ʌʁˤʔn̩ɾ̥ə(ː)] (4) > [ˈs̻t̻ʌʁˤʔn̩ə(ː)] (5) > [ˈs̻t̻ʌʁˤʔn̩nə(ː)] (6) > [ˈs̻t̻ʌʁˤn̩nə(ː)] (7) > [ˈs̻t̻ʌ̂ʁˤnːə(ː)] (1). The thing is that, even for myself, this sequence of changes is not obvious or transparent. Furthermore, it makes me wonder just what the actual underlying form of [ˈs̻t̻ʌ̂ʁˤnːə(ː)] (1) is, as the chain between it and what one could consider it to be phonemically in conventional analyses is essentially opaque.

(1) ["s_mt_mV_FR_?\n:@(:)], (2) *["s_mt_mVR_?\(4_0)1~n@(:)], (3) *["s_mt_mVR_?\(?)n=@(:)], (4) ["s_mt_mVR_?\?n=4_0@(:)], (5) ["s_mt_mVR_?\?n=@(:)], (6) ["s_mt_mVR_?\?n=n@(:)], (7) ["s_mt_mVR_?\n=n@(:)]

Now, I think that people are aware of both phones and phonemes... more or less.

Some informal evidence for the reality of a phonemic level:
1. The success of alphabetic scripts. No general writing system is at the phonetic level, especially not the sort of detail favored by Travis.
2. As a corollary, we can learn the pronunciation of a new word from a spelling or simple phonemic representation. (An educated speaker may learn most new words this way.) The phonetic details seem to be produced automatically.
3. After a bit of adjustment, we can understand people speaking other dialects and can even imitate them. We clearly learn rules on how to map their sounds to ours; we don't have to learn each word separately.
4. Imitating other dialects, most people will not reverse mergers. That is, they're going to rely on their own dialect + transformation rules, not on remembered phonetic representations of each dialect word.
5. Language games such as verlan or Pig Latin operate at the phonemic level— e.g. ch is not taken at t + sh; no one is bothered if a rearrangement changes which allophones are realized.
6. People are surprised by phonetic facts, such as the range of realizations of English /t/, or the fact that affricates are composed of several phones, or the fact that actual phonetic utterances of a given word vary in their own speech.

Some evidence for the reality of the (or a?) phonetic level:
1. Learning alphabetic writing is not difficult, but not immediate either. It seems like a skill that has to be trained.
2. We're perfectly aware that other dialects sound different, to say nothing of the wide range of other things we can do with voice— utterance-specific intonation, singing, identifying individual speakers, etc.
3. People tend to be sticklers about their lects' rules, disdaining variations or offering corrections. So to some extent the phonetic level is not only noticed but valued.
4. With minimal training we can produce phonetic representations— i.e. it's certainly not true that only the phonemic level is present.

People without much exposure to other dialects will have a phonemic representation based on their own dialect, not whatever abstraction linguists use for the standard. This should be obvious, but I think we can forget it a little too easily— e.g. Travis comes close to this above when considering what a representation "should be" in NAE.

Some people have suggested that phonemes are really a projection of alphabetic writing, and there may be something to this. I think it's striking how seemingly poorly a phonemic representation works for Mandarin, for instance: there's little value in assigning phonemes to each vowel. What works far better is the division the Chinese have been using for centuries: initial consonant vs. final. You virtually have to learn each final (e.g. -e, -uo, -ian, -üe) as a unit.

But if that's the case, I'd also wonder if the phonetic representations Travis uses are also to some extent a matter of trained abstraction. I mean, why stop at the phonetic level? The 'real' representation could be 'everything captured in a spectrogram'. (NB, this isn't Yet Another Complaint about Travis's phonetic representations. But if he can doubt the reality of phonemes, I can at least wonder about his phones.)

Anyway, I hope Travis will explain his thoughts on phonemes further, and if anyone else wants to offer arguments on what the brain actually does, that's great too.

[zompist]

I should also note that I think Chomsky & Halle "deep phonetics" are completely barmy and a projection from etymology. That's mostly due to the acqusition problem: a child hearing [neʃn] has no reason to phonemicize it as /nation/.

[faiuwle]

This kind of reminds me of how people seem to tell a lot of stories about small children who speak AmE dialects where /t/ is realized as [tS_h] before /r/ who misspell <tree> as <chree>. I never did this, and I never realized that the [tS_h] wasn't [t] until after taking an interest in linguistics. I suppose you could say that they internalize it as /tSri:/, but I don't think there is any /tSr/ as an onset cluster in English. Or maybe it's that there is simply is no real reason to analyze it as /tri:/ rather than /tSri:/ if you don't know the spelling already.

[äreo]

This kind of reminds me of how people seem to tell a lot of stories about small children who speak AmE dialects where /t/ is realized as [tS_h] before /r/ who misspell <tree> as <chree>. I never did this, and I never realized that the [tS_h] wasn't [t] until after taking an interest in linguistics. I suppose you could say that they internalize it as /tSri:/, but I don't think there is any /tSr/ as an onset cluster in English. Or maybe it's that there is simply is no real reason to analyze it as /tri:/ rather than /tSri:/ if you don't know the spelling already.

I myself have known such spellings. I remember seeing a girl in my kindergarten class spelling 'dream' as <jream>. She'd got the <eam> right, which seemed funny to me. And I remember thinking that her spelling made sense (cos I certainly have [tS_hr] there) despite being incorrect.

[linguoboy]

Some people have suggested that phonemes are really a projection of alphabetic writing, and there may be something to this. I think it's striking how seemingly poorly a phonemic representation works for Mandarin, for instance: there's little value in assigning phonemes to each vowel. What works far better is the division the Chinese have been using for centuries: initial consonant vs. final. You virtually have to learn each final (e.g. -e, -uo, -ian, -üe) as a unit.

You might change your mind if you saw the extremely elegant phonemic analysis of Standard Mandarin Pulleyblank gives in his Middle Chinese: a study in historical phonology. (If anything, it's almost too neat--particularly the "vowelless Chinese" variant discussed at the end.)

[bulbaquil]

I don't know if Pig Latin is the best example to use here, since it moves the entire onset - e.g. scram becomes am-scray, even though /skr/ (or, if you insist, /skr\`/) is not a single phoneme by any means.

Nevertheless, English speakers clearly tend to think of <ch> as a separate sound from <t> and <sh>, though they are essentially the same thing. Kindergarten and first-grade teachers teaching pronunciations and spellings, for instance, treat it as separate.

I think what's probably the case (disclaimer: I've done absolutely no research to corroborate my hypothesis, so it may be flawed):

Phonemic realizations are consciously learned, but we learn them very early, and often without or before schooling, to the point that they're so internalized that they become a reflex. (Add 4+3. Did you even have to think?) An alphabetic (or systematic-syllabic, like Japanese kana) writing system serves to enhance the phonemic reflex. To the best of my knowledge, there is no English word "snitter", but you probably knew how to pronounce it anyway. Likewise, we know how to spell "refudiate", a word which strictu senso did not exist ante Palinem.

Phonetic realizations are subconsciously learned, and probably are connected to the "monkey brain" somehow. We acquire them during initial language acquisition, which means they're in our dialect. You have to make a conscious effort to not speak in your dialect; my strongly-velarized-in-all-positions-/l/ is an example. If I say a word with /l/ in it, it automatically comes out [lˠ] unless I specifically make an effort to say [l].

[Nortaneous]

I don't think there is any /tSr/ as an onset cluster in English.

There is now.

I'd say the coronal affrication thing should be described as t d > tʃ dʒ / _r, not any sort of allophonic process applied to a phonemically unchanged /t d/. So, in those dialects, /tʃr dʒr/ are allowed onsets, but /tr dr/ aren't. It definitely looks odd, but keep in mind that this just extends an already existing restriction. Before, alveolar fricatives weren't allowed syllable-initially before /r/; /sr/ isn't a valid onset in any variety of English that I know of, but /ʃr/ is (<Sri Lanka> /ˈʃri ˈɫaŋkə/). Now, alveolars in general aren't allowed in that position.

[Radius Solis]

/sr/ isn't a valid onset in any variety of English that I know of, but /ʃr/ is (<Sri Lanka> /ˈʃri ˈɫaŋkə/). Now, alveolars in general aren't allowed in that position.

I've never had any problem pronouncing the [sri] in "Sri Lanka" and I can hardly conceive of it having an [ʃ] for it. Though I did have a Social Studies teacher in high school who had "shree lanka" - I remember kids making fun of her for it.

We tend not to have the pre-/r/ affrication rule up in this corner of the country either, which may be related, though it's not nearly as noticeable when someone affricates "tree" and "truck" as it would be if they said "shree lanka". I can't speak for distribution on either question beyond my home turf though. I never thought to listen for it when I used to live in Arizona.

To the best of my knowledge I don't have initial /sr/ in any other word though.

[Nortaneous]

/sr/ isn't a valid onset in any variety of English that I know of, but /ʃr/ is (<Sri Lanka> /ˈʃri ˈɫaŋkə/). Now, alveolars in general aren't allowed in that position.

I've never had any problem pronouncing the [sri] in "Sri Lanka" and I can hardly conceive of it having an [ʃ] for it. Though I did have a Social Studies teacher in high school who had "shree lanka" - I remember kids making fun of her for it.

Really? I've usually heard it with [ʃ], including from someone who was actually from Sri Lanka. And it's listed as a variant on Wikipedia.

[Radius Solis]

More on topic, I think these days we're using only two levels again, underlying representations and surface representations. (Generative Phonology and Optimality Theory have that much in common, at least.) No middle "phoneme" level. Though some seem to have reassigned that term to refer to underlying segments rather than the traditional Phonemic Theory concept which relied entirely on surface contrasts. Personally I try to avoid the term.

[Nortaneous]

More on topic, I think these days we're using only two levels again, underlying representations and surface representations.

Wait, what was the third level?

[Kai_DaiGoji]

6. People are surprised by phonetic facts, such as [...] the fact that affricates are composed of several phones [...]

Side note: I was under the impression that affricates were a single phone, i.e. that [tS] was not a [t] followed by a [S] but a single sound, with a stop followed by a fricative release. The fact that certain affricates contrast with the equivalent (or whatever) consonant clusters would seem to support this. I'm not calling anyone out here, just asking for clarification, as I am deeply confused.

[chris_notts]

I should also note that I think Chomsky & Halle "deep phonetics" are completely barmy and a projection from etymology. That's mostly due to the acqusition problem: a child hearing [neʃn] has no reason to phonemicize it as /nation/.

That's what I've been thinking recently with a lot of the phonology stuff I've been reading, as I said in the other thread.

But it's not just limited to English. There are analyses of a number of languages that involve extremely complicated synchronic rules to derive surface forms from underlying forms. IIRC, one of the best examples I've seen is the analysis of Kabardian (a language supposedly with only two underlying vowels and big consonant clusters) by Colarusso. I find it hard to believe that all these rules are actually applied in real-time by speakers of the language.

The whole motivation for it seems to be to avoid redundancy. But as any computer programmer knows, there's often a trade-off between time and memory usage. Who's to say that the human brain doesn't prefer to store something close to the actual output forms redundantly instead of doing a lot of complicated rule application everytime it decides to speak?

[Legion]

1. Learning alphabetic writing is not difficult, but not immediate either. It seems like a skill that has to be trained.

For what I have seen, this is mainly because it is much easier to get aware of *syllables* than of individual phonemes, so that syllabaries, for all their drawbacks, are actually easier to learn than alphabets.

[AnTeallach]

This kind of reminds me of how people seem to tell a lot of stories about small children who speak AmE dialects where /t/ is realized as [tS_h] before /r/ who misspell <tree> as <chree>. I never did this, and I never realized that the [tS_h] wasn't [t] until after taking an interest in linguistics. I suppose you could say that they internalize it as /tSri:/, but I don't think there is any /tSr/ as an onset cluster in English. Or maybe it's that there is simply is no real reason to analyze it as /tri:/ rather than /tSri:/ if you don't know the spelling already.

I don't find it at all unreasonable that some people internalise it as /tʃr/ while others internalise it as /tr/. Similarly with other cases of neutralisation: a friend of mine once said something which seemed to imply he thought of what are normally analysed as /st sp sk/ in English as /sd sb sg/.

I agree with most of what zompist said, but especially:

People without much exposure to other dialects will have a phonemic representation based on their own dialect, not whatever abstraction linguists use for the standard. This should be obvious, but I think we can forget it a little too easily— e.g. Travis comes close to this above when considering what a representation "should be" in NAE.

I always find it a bit bizarre when people seem to want a phonemic (or underlying) representation to apply to all dialects simultaneously.

[Jipí]

he thought of what are normally analysed as /st sp sk/ in English as /sd sb sg/.

The plosives are unaspirated in that position. Since AFAIK English voiced stops aren't fully voiced, the impression of voiced stops arises. In my Intro to Linguistics class last week exactly this happened to a couple of people when we were supposed to give the pronunciation of a word with <sp> /ʃp/ in IPA. They transcribed it as /ʃb/ because that's what it sounded like to them. The instructor then went to demonstrate the difference between [p pʰ] by example of French using [p], not [pʰ] like German does. We didn't do the hold-your-hand-in-front-of-your-mouth thing, wtf?!

I would've expected Travis to have bitten by now.

[Travis B.]

I would've expected Travis to have bitten by now.

I have seen this thread, but I have waited to see what others would write and to formulate more coherent responses than if I just responded immediately.

[Travis B.]

I don't think there is any /tSr/ as an onset cluster in English.

There is now.

I'd say the coronal affrication thing should be described as t d > tʃ dʒ / _r, not any sort of allophonic process applied to a phonemically unchanged /t d/. So, in those dialects, /tʃr dʒr/ are allowed onsets, but /tr dr/ aren't. It definitely looks odd, but keep in mind that this just extends an already existing restriction. Before, alveolar fricatives weren't allowed syllable-initially before /r/; /sr/ isn't a valid onset in any variety of English that I know of, but /ʃr/ is (<Sri Lanka> /ˈʃri ˈɫaŋkə/). Now, alveolars in general aren't allowed in that position.

My dialect is weird here, in that it traditionally affricates /t/ and /d/ before /r/, but it does not merge them with /tʃ/ and /dʒ/ before /r/ (historically only found medially) because the resulting affricates are retroflex and begin as alveolar while (medial) /tʃ/ and /dʒ/ before /r/ are not. However, some but not all younger people such as myself have lost this distinction, and can thus be said to have had the shift you describe.

[Travis B.]

I agree with most of what zompist said, but especially:

People without much exposure to other dialects will have a phonemic representation based on their own dialect, not whatever abstraction linguists use for the standard. This should be obvious, but I think we can forget it a little too easily— e.g. Travis comes close to this above when considering what a representation "should be" in NAE.

I always find it a bit bizarre when people seem to want a phonemic (or underlying) representation to apply to all dialects simultaneously.

It certainly is bothersome, because it makes it so that one has to try to jury-rig some scheme to get a phonology to fit in with an expected phonemic system that may very well not make sense or seem contrived from a synchronic perspective (e.g. require complex chains of transformations that are unlikely to actually take place synchronically). And if one does not, then one is seen as an Octavià-type kook, as I know some do see me.

In my own case, in much of the English I have come into contact with - not just my own dialect, but at least most Inland North and even many more progressive General American-like varieties I have come into contact with - it seems that new vowel length (derived from historical vowel length allophony) and vowel nasalization are already actually part of the underlying forms (except in final syllables that never had obstruents in their codas, where then vowel length is unset), whatever they are. The reason why I think this is that vowel length and nasalization are too consistent to be derived from synchronic allophony; i.e. once a vowel's length and nasalization has been set, the only way either change is if a vowel loses syllabicity, with changes in environment never changing either except in final syllables' vowels' whose vowel lengths are unset. Were vowel length and nasalization to be synchronically allophonic, one would expect at least sporadic cases of vowels "forgetting" what length and/or nasalization they "should" have - especially in cases where the diachronic path by which a certain vowel's length has come to be is obscure, such as in some contracted forms such as gonna, wanna, and it'd, in which one would expect the vowel to easily change length (in these cases from short to long) in the process of language acquisition due to a lack of an internalized path by which a vowel's length could turn out to be different than what its synchronic environment "should" condition.

(Note that this view of things also requires that things can be part of underlying forms yet be consciously inaccessible, as individuals cannot neologize new forms with unexpected vowel lengths, and can only set arbitrary vowel nasalization in interjections.)

[Travis B.]

I should note, even though I will only get into detail later, that I do not believe that there is no such thing as underlying forms. Rather, I doubt conventional phonemic-type models and the synchronic reality of purported generative-type phonologies, especially "deep" ones. Likewise, I doubt that all information in underlying forms is necessarily consciously accessible or arbitrarily manipulable (in processes like borrowing and neologism) without being "cued" by something like an orthography.

Particularly, it seems to me that typical purported generative-type phonologies heavily confuse diachronic phenomena for being synchronic phenomena, that is, they presume phenomena to be happening in the process of speech production that actually happened diachronically, with the results of such processes being acquired directly. Likewise, the purported synchronic results of said generative-type phonologies, especially "deep" ones, could be far more plausibly treated in terms of a combination of allomorphy and analogy. One key reason why I think this is that sound change can be shown to actually take place in terms of lexical diffusion and is often quite lexically (and even syntactically!) sensitive and often takes place upon multiple words at a time which are in frequent close contact, things that all directly contradict generative-type phonological models but which are very compatible with phonological models that are synchronically "shallow" and which heavily rely upon allomorphy and analogy to replace "deep" phonological processes.

[Richard W]

I'd say the coronal affrication thing should be described as t d > tʃ dʒ / _r, not any sort of allophonic process applied to a phonemically unchanged /t d/. So, in those dialects, /tʃr dʒr/ are allowed onsets, but /tr dr/ aren't.

They may still occur in words like history - I have ["hIstri] alternating with ["hIstSrI], whereas word-initially [tr] for [tSr] is rare. I've never been able to persuade myself that the former word-initial is [t`r].

Incidentally, the coronal affrication is also present in England.

[äreo]

They may still occur in words like history - I have ["hIstri] alternating with ["hIstSrI],

Yeah, I have that distinction too, but in such words (like history) it results from schwa reduction.

[Radius Solis]

Code: Select all

[quote="Kai_DaiGoji"][quote="zompist"]
6. People are [i]surprised[/i] by phonetic facts, such as [...] the fact that affricates are composed of several phones [...]
[/quote]
Side note: I was under the impression that affricates were a single phone, i.e. that [tS] was not a [t] followed by a [S] but a single sound, with a stop followed by a fricative release.  The fact that certain affricates contrast with the equivalent (or whatever) consonant clusters would seem to support this.  I'm not calling anyone out here, just asking for clarification, as I am deeply confused.[/quote]
Whether a certain articulation is one or two phones is an matter of how you definition "phone", not an empirical question with a solid answer. Some linguists will call [tS)] one phone, some may still call it two, depending what they mean when they say it. The traditional answer is to call affricates two phones; the IPA is structured on that assumption, e.g.. Whereas in articulatory terms, the difference between a plosive and an affricate is how suddenly or gradually the closure is released - a more gradual release makes more audible the portion of the release that resembles a fricative, but that portion is there in any plosive. It's just a matter of sudden burst versus a bit drawn out so you can hear it better. But for lightweight discussion that doesn't need to be exact, it can still be a convenient approximation to discuss affricates in terms of stop phones followed by fricative phones.

It's worth noting that very few languages contrast affricates with stop-fricative clusters, and that in the ones that do (e.g. Montana Salish; perhaps Roninbodhisattva can comment further?), the clusters involve a full release of the plosive followed by an independently initiated fricative - that is, no different than clusters of [kS] or [pS].

[zompist]

But it's not just limited to English. There are analyses of a number of languages that involve extremely complicated synchronic rules to derive surface forms from underlying forms. [...]

The whole motivation for it seems to be to avoid redundancy. But as any computer programmer knows, there's often a trade-off between time and memory usage. Who's to say that the human brain doesn't prefer to store something close to the actual output forms redundantly instead of doing a lot of complicated rule application everytime it decides to speak?

That was exactly my reaction, as a programmer, to much of linguistics when I first studied it. It was reinforced by noticing things like spelling errors by native French speakers, which made it clear that they were missing 'obvious' rules (e.g. "spell all /-e/ infinitives with er"). Because linguists can recover a rule does not mean that the speakers do.

Michael Tomasello is very good on this in syntax; his idea, based on acquisition studies, is that children learn verbs and constructions individually, and only generalize to ideas like 'direct object' far later. Even things like the question transformation are historical, especially as children often hear and master questions first, before statements.

[Travis B.]

They may still occur in words like history - I have ["hIstri] alternating with ["hIstSrI],

Yeah, I have that distinction too, but in such words (like history) it results from schwa reduction.

Not necessarily. In my dialect there are a number of words where historical /stər/ turned into, depending on the speaker, /stʃər/ or /ʃtʃər/, without any schwa elision taking place at all. Rather consistent examples thereof include history (but not other related words such as historic), mystery, and restaurant. There are also words where this has happened inconsistently in some idiolects (or particular subdialects) but not consistently or broadly, such as sister and faster; I for instance outside formal speech inconsistently have it sister, even though most people speaking roughly the same dialect as myself lack it therefor, and I have heard people have it for faster even though no one I know from back in Wauwatosa has it there.