"Canadian raising": expert opinions needed

[Basilius]

A dialogue quoted from another thread:

To give an example: the raising of /aI/ to /VI/ in some north american dialects has taken place AFTER the neutralisation of the intervocalic T-D contrast that conditions that change! On the surface level, there's just a flap, but speakers are able to recognise the underlying phoneme and raise or not raise the preceding vowel accordingly.

This is slightly off-topic, but it may have been somewhat different. I am used to there still being a remaining preceding vowel length difference between unstressed intervocalic /t/ and /d/, indicating that the two phonemes are not truly neutralized, even when unstressed intervocalic /t/ is fully voiced.

I'm not.

I was not necessarily saying that there is synchronically a vowel length distinction in all present dialects that split /ai̯/ into /aɪ̯/ and /əɪ̯/, just that such may have existed at the time the split originally occurred, and that it would explain away any need to remember historical forms without surface realizations.

Of course this would be harder to explain if we could find dialects that both have this split and preserve historical vowel length, i.e. have never developed allophonic vowel length in the first place. Then we would have to go back to what Sal was saying.

The topic may be worth some in-depth discussion. Some obvious questions:

(1) For retrogrades like me, the idea that a shift in pronunciation may depend on (telepathically transmitted?) underlying forms is still too odd. What can be considered a strong evidence favoring it, in this case? Is there indeed any evidence of that sort?

(2) At a glance, the difference in vowels with pairs of words like writer vs. rider looks just faithfully historical, i. e. reflecting a positional development whose conditioning factors (ultimately associated with voicedness of the consonant) were lost relatively recently. Is there anything strongly contradicting such analysis?

(3) On the other hand, forms like writer and rider can be easily affected by analogy. How much material is there on "Canadian rising" in unanalyzable forms with flapped [t ~ d]?

(4) And, of course, phonetic detail. There must have been tons of research in instrumental phonetics around "Canadian raising" - what's going on with vowel length (like Travis suggested), phonations, intensity curves, whatever?

[Nortaneous]

Note that certain unanalyzable words with <d> -- spider and cider are the two I can think of, but I think there are more -- have [ʌi], at least IMD.

Another question: How are words with [ai] followed by <t> borrowed?

(1) For retrogrades like me, the idea that a shift in pronunciation may depend on (telepathically transmitted?) underlying forms is still too odd.

Spelling pronunciation isn't telepathic transmission.

[Basilius]

Spelling pronunciation isn't telepathic transmission.

Yep. The question is, then, how they can be distinguished from the rest.

[Travis B.]

Note that certain unanalyzable words with <d> -- spider and cider are the two I can think of, but I think there are more -- have [ʌi], at least IMD.

I remember a whole discussion about this, that there is a whole class of words ending historically in -/aɪ̯Lər/, where L is a lenis obstruent (not just /d/; I have raising in tiger) and where this sequence is not split by any morpheme boundaries, that underwent raising to /əɪ̯/ in many American English dialects (but incidentally not Canadian English ones). This is also why I analyze this as a true phonemic split, at least in American English dialects with this, in that not all words that could have undergone raising according to this pattern actually did; e.g. fiber never underwent raising in my dialect.

(My dialect also has raising in some other sequences of historical /aɪ̯də/, /aɪ̯dɔː/, and /aɪ̯doʊ̯/ not split by a morpheme boundary, resulting in words like idle, Idaho, Ida*, Midol*, Fido, and bridle* also having raising to /əɪ̯/.)

* These words seem to be able to go both ways, both with and without raising.

Another question: How are words with [ai] followed by <t> borrowed?

At least in my own dialect, /aɪ̯/ cannot be followed by a fortis obstruent unless split from it by a morpheme boundary, so hence [ai] followed by <t> is uniformly borrowed as /əɪ̯/.

[Radius Solis]

I suspect there so few monomorphemes with /ai/ followed by historical /t/ that no general conclusions about its outcome may be reachable. The sole example I can think of is mitre, which is little used by people who aren't bishops or carpenters.

Examples with historical /d/ also include idle and sidle.

[Herr Dunkel]

I suspect there so few monomorphemes with /ai/ followed by historical /t/ that no general conclusions about its outcome may be reachable. The sole example I can think of is mitre, which is little used by people who aren't bishops or carpenters.

Examples with historical /d/ also include idle and sidle.

Do <bite>, <quite>, <kite>, <site> (of both or only one kind), <lite> (from ON <hlíta>) count or am I missing the point?

[Travis B.]

I suspect there so few monomorphemes with /ai/ followed by historical /t/ that no general conclusions about its outcome may be reachable. The sole example I can think of is mitre, which is little used by people who aren't bishops or carpenters.

Umm... there are Herr Dunkel's examples, and then there are more, such as light, night, might, height, bright, right, fight, Dwight...

(Well, unless you are only talking about loans...)

[Herr Dunkel]

Maybe Radius meant /ait/ that didn't come from /*ixt/, IDK?
Edit: or about loans, yeah.

[Travis B.]

Maybe Radius meant /ait/ that didn't come from /*ixt/, IDK?

That is a good bit further back than what people normally mean when they speak of historical phonemes in English, that is, typically very late Early Modern English or early Modern English phonemes.

[Basilius]

I think Radius meant words with flapped /t ~ d/, for words where these weren't neutralized aren't problematic in terms of sound changes conditioned by (non-audible) underlying distinctions.

Perhaps a few other words like vital can be relevant here, too, for their relation to forms with unambiguous /t/ isn't transparent enough either semantically or morphonologically (i. e. involves other, more sophisticated alternations).

[Salmoneus]

There's no need to introduce passive-aggressive telepathy in order to give people basic awareness of their own language. Alternative media include:
- memory: where a change has taken place within the speaker's lifetime, they may remember how they used to say it, or at least may remember that they used to say things differently.
- spelling: where the words are written down, the knowledge of how a word is spelled may inform speakers of how it 'ought' to be spoken
- registers: where a speaker speaks two or more registers (or even just related languages, or languages in which the words in question are related), knowledge of one register, where, say, a distinction has not been neutralised, can provide information than can affect their pronunciation of the 'same' word in another register
- dialects: where a speaker speaks, or passively understands, two or more dialects, knowledge of one dialect may affect their speech in the other
- lexical borrowing: where a speaker lacks intimate understand of another dialect or language, they may nonetheless borrow pronunciations on a word-by-word basis from that dialect or language; if the source dialect includes a sound change that is 'impossible' in the target dialect, such borrowing may eventually produce the impossible change in the target dialect
- morphophonemic variation: where a word is paired in the speaker's mind with some other word via an inflectional or derivational process, information about one word may affect their pronunciation of the other word
- morpheme identification: where a speaker is able to analyse certain phonemes as belonging to a certain morpheme, they will be able to apply different rules to those phonemes depending on whether they do or do not belong to that certain morpheme
- contextual distinction: a speaker may come to regard even a homophonous word as being two distinct words (and apply different sound changes to them) if they are able to distinguish one word from another by semantic or pragmatic context
- disambiguative anomalies: where two words are homophonous but the speakers consider them two different words, the speakers may exaggerate any perceived differenced in their pronunciation (due, eg, to spelling or morphology or function vs non-function, or perceived borrowing, or whatever) and end up changing the pronunciation of one word but not of the other.
- lexical stress: where one word tends to be said with greater stress than another, different sound changes may apply to the two words (this often distinguished function and non-function words, or adjectives and nouns, etc).
- frequency: more commonly used words may undergo different sound changes from less commonly used words

All the above can lead to sound change that is not determined by "surface" phones. All of the above probably happen in north american English.

Frankly, though, your approach seems entirely the wrong way around. What people are aware of are phonemes, and morphemes, and lexemes. You want to advance a fairly radical and prima facie improbable theory: that sound change is always and entirely and everywhere determined by surface phones (of which the speaker is often only partially conscious, if at all), and never takes into account any knowledge the speaker may have about phonemes, morphemes, or lexemes. This may well be the case, but it's a very bold theory to advance! On the one hand it's quite counterintuitive (it requires speakers to close off their brains while sound changes just magically happen in a pure state uncontaminated by messy humans), and on the other hand any universal claim is likely to be flawed. So as I say, maybe your theory is true, but I should think you would need to put forward some extraordinary evidence to demonstrate it!

[Put in these terms, the question I answered above is equivalent to 'how can people know about phonemes, lexemes and morphemes other than by the surface phones appearing in their current synchronic speech in a single dialect and register?'. And of course surface phones in the way we say things in our primary dialect and register today are indeed our main source of information about emes... but not our only source]

[Travis B.]

I think Radius meant words with flapped /t ~ d/, for words where these weren't neutralized aren't problematic in terms of sound changes conditioned by (non-audible) underlying distinctions.

Perhaps a few other words like vital can be relevant here, too, for their relation to forms with unambiguous /t/ isn't transparent enough either semantically or morphonologically (i. e. involves other, more sophisticated alternations).

On that note, my dialect has /əɪ̯/ for both mitre and vital.

[Basilius]

Salmoneus, it would be helpful if you'd labeled the items in your list with numbers or somesuch.

OK, I'll do that boring and passive-aggressive job for you:

(1) memory: where a change has taken place within the speaker's lifetime, they may remember how they used to say it, or at least may remember that they used to say things differently.
(2) spelling: where the words are written down, the knowledge of how a word is spelled may inform speakers of how it 'ought' to be spoken
(3) registers: where a speaker speaks two or more registers (or even just related languages, or languages in which the words in question are related), knowledge of one register, where, say, a distinction has not been neutralised, can provide information than can affect their pronunciation of the 'same' word in another register
(3) dialects: where a speaker speaks, or passively understands, two or more dialects, knowledge of one dialect may affect their speech in the other
(4) lexical borrowing: where a speaker lacks intimate understand of another dialect or language, they may nonetheless borrow pronunciations on a word-by-word basis from that dialect or language; if the source dialect includes a sound change that is 'impossible' in the target dialect, such borrowing may eventually produce the impossible change in the target dialect
(5) morphophonemic variation: where a word is paired in the speaker's mind with some other word via an inflectional or derivational process, information about one word may affect their pronunciation of the other word
(6) morpheme identification: where a speaker is able to analyse certain phonemes as belonging to a certain morpheme, they will be able to apply different rules to those phonemes depending on whether they do or do not belong to that certain morpheme
(7) contextual distinction: a speaker may come to regard even a homophonous word as being two distinct words (and apply different sound changes to them) if they are able to distinguish one word from another by semantic or pragmatic context
(8) disambiguative anomalies: where two words are homophonous but the speakers consider them two different words, the speakers may exaggerate any perceived differenced in their pronunciation (due, eg, to spelling or morphology or function vs non-function, or perceived borrowing, or whatever) and end up changing the pronunciation of one word but not of the other.
(9) lexical stress: where one word tends to be said with greater stress than another, different sound changes may apply to the two words (this often distinguished function and non-function words, or adjectives and nouns, etc).
(10) frequency: more commonly used words may undergo different sound changes from less commonly used words

(1) - (4) are about interaction of registers, and ultimately dialects (for written norm can be considered a dialect/register), through borrowing, and in that perspective aren't specific enough in a few important points (let it be "group A"); (5) and (6) are actually the same, and (9) is a subset of the former two (group "B", analogical levelings); (7), (8) and (10) don't exist without the factors in the other two groups.

Also, it seems that you haven't read what other people said on the topic. There was no shortage of haphazard hypotheses, and it would be much more interesting to discuss what the actual data point to.

[Travis B.]

One thing to note is that apparently Canadian Raising was introduced into Canadian English from Scottish English, where the distribution of /əi/ versus /aɪ/ corresponds to the Scottish Vowel Length Rule, with /əi/ being found in short environments and /aɪ/ being found in long environments.

Hence it would not be surprising of Canadian Raising of /aɪ̯/ really were originally being conditioned by environments conditioning vowel length, with the main change in Canadian English vis-a-vis Scottish English being the exact environments that condition vowel length (since allophonic vowel length in NAE does not follow the Scottish Vowel Length Rule), and hence why /aɪ̯/ is found before all lenis obstruents, at least before subsequent sound changes that introduced /əɪ̯/ into various long environments.

[Basilius]

OK, an attempt at an intermediary summary.

It looks like common words with historical /d/ have /əɪ̯/ at least in the dialects most familiar to both Travis and Nortaneous - *except* when there is an obvious connection to related forms with a coda /d/ (rider etc.); on the other hand, there seems to be no /aɪ̯/ before a historical /t/.

This may point to analogy as the source of the current realization of the contrast in rider vs. writer . Also:

(a) this analogy was a stronger factor than spelling pronunciations (else we'd have /aɪ̯/ rather than /əɪ̯/ in spider and the like);

(b) the distribution in morphologically obscure forms suggests that only /əɪ̯/ was originally pronounced in non-final syllables, which may support Travis' hypothesis about vowel length formerly involved in the contrast (for this positional difference in vowel length tends to be strongest before a coda consonant, in English dialects in general);

(c) it seems that in Canada the distribution of /aɪ̯/ and /əɪ̯/ reflects the original difference in adjacent consonants more straightforwardly, which is interesting since the area where the very phenomenon originated from was there, and the merger of (the flapped allophones of) /t/ and /d/ happened later in that area.

Does the above look plausible enough?

(EDIT: didn't see Travis' latest comment while writing mine)

[Travis B.]

OK, an attempt at an intermediary summary.

It looks like common words with historical /d/ have /əɪ̯/ at least in the dialects most familiar to both Travis and Nortaneous - *except* when there is an obvious connection to related forms with a coda /d/ (rider etc.); on the other hand, there seems to be no /aɪ̯/ before a historical /t/.

No, most words with historical /d/ have /aɪ̯/ except in the cases I delineated before, which are more narrow in range than the range of Canadian Raising in general (i.e. /əɪ̯/ in spider is found more narrowly, and /əɪ̯/ in idle is found even more narrowly).

This may point to analogy as the source of the current realization of the contrast in rider vs. writer . Also:

(a) this analogy was a stronger factor than spelling pronunciations (else we'd have /aɪ̯/ rather than /əɪ̯/ in spider and the like);

I doubt that analogy was the source of the distinction, but analogy could have a role in maintaining the distinction (e.g. analogy between ride and rider and between write and writer preventing the formation of stem alternations between /aɪ̯/ and /əɪ̯/) and resisting words switching sets with subsequent sound changes (e.g. maintaining /əɪ̯/ before /t/ in writer even with voicing of /t/ and neutralization of vowel length, and keeping rider from switching to /əɪ̯/ as spider has).

(b) the distribution in morphologically obscure forms suggests that only /əɪ̯/ was originally pronounced in non-final syllables, which may support Travis' hypothesis about vowel length formerly involved in the contrast (for this positional difference in vowel length tends to be strongest before a coda consonant, in English dialects in general);

Properly it seems that /aɪ̯/ is normally found at the end of morphemes, with some exceptions in some dialects (e.g. high school taking /əɪ̯/ in many dialects); it would not be surprising if there was some interaction between vowel length and morphemic structure at some point (compare how vowels that alternate in length according to the Scottish Vowel Length Rule are typically long at the ends of morphemes, even when followed by a consonant that would normally condition a short environment).

(c) it seems that in Canada the distribution of /aɪ̯/ and /əɪ̯/ reflects the original difference in adjacent consonants more straightforwardly, which is interesting since the area where the very phenomenon originated from was there, and the merger of (the flapped allophones of) /t/ and /d/ happened later in that area.

What I wonder about is why the innovations that have occurred with regard to the distribution of /əɪ̯/ have not crossed the border back into Canada, considering the generally high degree of language contact between the northern US and non-Atlantic English-speaking parts of Canada.

[Basilius]

No, most words with historical /d/ have /aɪ̯/ except in the cases I delineated before, which are more narrow in range than the range of Canadian Raising in general (i.e. /əɪ̯/ in spider is found more narrowly, and /əɪ̯/ in idle is found even more narrowly).

Aha.

Then the simplest (although also least instructive) thing to do is to suppose some lexical diffusion... Unless something more interesting can be found about it.

However, /əɪ̯/ before |d| (in spider etc.) is more difficult to explain away than /aɪ̯/. I mean, it can spread as a spelling pronunciation only if already perceived as the default treatment of historical /aɪ̯/ + /d/; as a cross-dialectal loan, it implies that there are some dialects around in which it's phonetically regular, etc.

Can't we say, then, that words with /əɪ̯/ before |d| are a conservative layer in dialects that have them?

(I'll reply to the rest, too, but I need to think a bit... and sleep a bit, too :) )

[Travis B.]

No, most words with historical /d/ have /aɪ̯/ except in the cases I delineated before, which are more narrow in range than the range of Canadian Raising in general (i.e. /əɪ̯/ in spider is found more narrowly, and /əɪ̯/ in idle is found even more narrowly).

Aha.

Then the simplest (although also least instructive) thing to do is to suppose some lexical diffusion... Unless something more interesting can be found about it.

However, /əɪ̯/ before |d| (in spider etc.) is more difficult to explain away than /aɪ̯/. I mean, it can spread as a spelling pronunciation only if already perceived as the default treatment of historical /aɪ̯/ + /d/; as a cross-dialectal loan, it implies that there are some dialects around in which it's phonetically regular, etc.

Can't we say, then, that words with /əɪ̯/ before |d| are a conservative layer in dialects that have them?

(I'll reply to the rest, too, but I need to think a bit... and sleep a bit, too )

The most conservative case is /aɪ̯/ in all cases, and /əɪ̯/ is always the progressive case; rather, we have four layers of sound changes, all of which changed historical /aɪ̯/ to /əɪ̯/, i.e. before fortis obstruents (classical Canadian Raising), before /r/ (before breaking of coda /r/ into /ər/ after diphthongs), before many cases of /Lər/ together in monomorphemes (where L is a lenis plosive), and before cases of /də/, /dɔː/, and /doʊ̯/ in some words.

Of course what complicates this is if complete neutralization of unstressed intervocalic /t/ and /d/ precedes the first change in some dialects with this change, but we have no reason to believe that complete neutralization, i.e. loss of all distinctions including vowel length, actually precedes this step.

[Travis B.]

Now that I think if it, if analogy were the driving force behind the distribution of /aɪ̯/ and /əɪ̯/ before historical unstressed intervocalic /t/ and /d/, we should see discrepancies where the /aɪ̯/ would be found before historical unstressed intervocalic /t/ and /əɪ̯/ would be found before historical outside unstressed intervocalic /d/ (outside the scope of the sound changes outlined above), particularly stem-medially where there is no force of analogy between different words with similar stems to maintain which vowel they receive.

But there are no discrepancies in /əɪ̯/ being found before historical unstressed intervocalic /t/, and all the places /aɪ̯/ is not found before historical unstressed intervocalic /d/ are explainable in terms of straightforward sound change (and any discrepancies in that are explainable in terms of lexical diffusion).

[Basilius]

The most conservative case is /aɪ̯/ in all cases, and /əɪ̯/ is always the progressive case; rather, we have four layers of sound changes, all of which changed historical /aɪ̯/ to /əɪ̯/, i.e. before fortis obstruents (classical Canadian Raising), before /r/ (before breaking of coda /r/ into /ər/ after diphthongs), before many cases of /Lər/ together in monomorphemes (where L is a lenis plosive), and before cases of /də/, /dɔː/, and /doʊ̯/ in some words.

OK, it seems that some remarks are needed to clarify what I meant by "conservative" and what I'd consider conservative in principle, and why.

(1) Ultimately, the more conservative realization of the "long-i" vowel is [əɪ̯], and this should not be discarded especially when we start speaking of possible influences from the dialects of Britain. In other words, it is in principle possible that what we deal with is not "raising" but, rather, a restriction on widening ([əɪ̯] -> [aɪ̯]). To choose between the two hypotheses ("raising" and "restricted widening") we'll have to consider a separate set of arguments, and I didn't mean to touch upon this in my previous comment.

(2) What I meant was a distributional thing: in dialects which already contrast /aɪ̯/ and /əɪ̯/, the "raised" diphthong in spider etc. should reflect the regular phonetic development, while the forms with /aɪ̯/ before historical intervocalic /d/ can be either analogical (rider etc.) or borrowed (including, as individual lexemes).

(3) What I missed is the possibility, to which your other comment seems to point, that /aɪ̯/ could be perceived as a marker of a morphological boundary; in other words, the distinction between /aɪ̯/ and /əɪ̯/ could spread across dialect boundaries as a morphological technique rather than a phonetic feature. One weakness of this hypothesis is that at a glance, it needs the (newly) phonemic distinction to be borrowed together with the new morphological technique, which may look a bit too complicated. Nevertheless, this hypothesis, essentially treating the raising in spider etc. as an analogical innovation in dialects that recently borrowed the very distinction in vowels, should not be rejected offhand.

(4) However, your comment quoted above points to one more possibility, of the sort which I called more interesting than (random) lexical diffusion before: there are several phonetically conditioned environments in which different dialects have different distribution of /aɪ̯/ and /əɪ̯/; we need to analyze the logic of this distribution to understand what happened (or failed to happen) to spider &like. But again, IMO we should not declare some distributions more conservative without a more scrupulous analysis. In particular, looking at linguogeography, I'd consider more elaborate conditioning as a sign of earlier change, and less discriminate distributions as pointing to areas of later spread, but it's not granted at all.

[Drydic]

And you say əI is the most conservative value why?

[Basilius]

And you say əI is the most conservative value why?

Great Vowel Shift.

Around 1700, [əɪ̯] was still a possible realization in the prestige lects of Britain. In many places (e. g. in Ireland) it survived well into the 20th century. So...

[Drydic]

And do you have proof that it didn't finish moving to aI and then partially shifted back (no, parts of Ireland retaining əI is not proof)? Note: Occam's Razor does not count as proof, however sensible it is.

[Basilius]

And do you have proof that it didn't finish moving to aI and then partially shifted back (no, parts of Ireland retaining əI is not proof)? Note: Occam's Razor does not count as proof, however sensible it is.

No.

As I said, it's just a possibility which needs a separate set of arguments to be ruled out. But cannot be rejected just because, for it's not about some prehistoric antiquity at all. It's about urban varieties of English which were quite alive in 1970's (dunno about today).

[Drydic]

And do you have proof that it didn't finish moving to aI and then partially shifted back (no, parts of Ireland retaining əI is not proof)? Note: Occam's Razor does not count as proof, however sensible it is.

No.

As I said, it's just a possibility which needs a separate set of arguments to be ruled out. But cannot be rejected just because, for it's not about some prehistoric antiquity at all. It's about urban varieties of English which were quite alive in 1970's (dunno about today).

Which varieties?