To start off, its clear that our theoretical assumptions are irreconcilable (I might go so far as to say "diametrically opposed"), and that we are not going to agree here, but its important to note that my model is perfectly able to capture your German data.

First, I got most of my linguistics education in German so sorry for my bad example when I was looking for an English one.

It was a great example. There's no such thing as a bad example, because sound change is equally regular in every natural human language.

Yes, the vast majority of theoretical linguists, and practically all historical linguistics, both in America and in Europe (with much of the best European work still coming from Germany and the Netherlands), very much still subscribe to the regularity of sound change, because as far as we can tell, it's an empirical fact.

Also note that it's impossible to prove language relatedness without the regularity of sound change. Regularity of correspondence is literally the only metric we have that can prove relatedness, so if the Neogrammarian Hypothesis were somehow disproven (which is very unlikely), then the scientific underpinnings of the way we group languages into families immediately collapses.

(Also, yes, hypercorrection is another form of analogy, often called "interdialectic analogy".)

This might be a philosophical question tho: Is everything regular but we don't know all the rule or are there "real" exceptions?

This is a great question, and technically it's still unproven (and may never be), but the hypothesis has been borne out in so much data for so many decades, with no convincing counterexamples, that there seems to be no good reason to disbelieve it.

OH! I should include the most important reason why the regularity of sound change is considered by most western linguists to be scientifically reliable - it makes predictions that are borne out by new data.

The Case of the Indo-European laryngeals

(This is an oversimplication of the events, because the data is complex and goes beyond the scope of our discussion here, but the (wikipedia page)[https://en.m.wikipedia.org/wiki/Laryngeal_theory] is fairly good if you're interested.)

Basically, in the late 1800s, scholars working on reconstructing Proto-Indo-European were a bit confused - the reconstructed sound system (which is reconstructible, of course, due to the regularity of sound change) seemed to have two different systems of vowel alternations - a system unheard of in any of the world's languages.

Ferdinand de Saussure (yes, that Ferdinand de Saussure) realized that he could collapse both systems into one by positing an unspecified series of sonorant consonants (his famous coefficients sonantiques) that colored adjacent vowels in specific environments before disappearing entirely in all of the daughter languages. This resulted in a much simpler system that was also more typologically likely.

His contemporaries ridiculed him for reconstructing a proto-sound that disappeared in all of the daughter languages, but, once Hittite was deciphered in the early 1900s, shortly after de Saussure died, every single place de Saussure had predicted his "coefficients sonantiques" to show up in the proto-language, Hittite had an "h".

None of this is possible without the regularity of sound change, and we've seen the theory make predictions that are borne out by new data again and again.

Yes, linguists very much still subscribe to the regularity of sound change, both in the US and in Europe.

a factor you forgot to address is frequency

I didn't forget anything. While frequency is clearly a factor in language change, it's not relevant for sound change since it reduces to a prosody/stress change, which is a describable regular phonological environment that can be acted upon by regular sound rules.

In your "haben" example, for example, the grammaticalization of a main verb to an auxiliary verb clearly establishes a new prosodic pattern, which can be acted upon by regular sound change to the exclusion of other main verbs.

We see similar alternations in English main/auxiliary verb pairs:

I've already eaten. BUT
*I've a cheesburger. (In American English - Brits can do this)
I'm gonna eat a cheeseburger. BUT
*I'm gonna the store.

This is, of course, expected, since the grammaticalization of main verbs into auxiliary verbs results in a different stress/prosodic pattern (which I'm sure you can feel in German with "haben" as well), and so it's a perfect location for a regular and exceptionless sound rule to occur.

And these phenomena (and likely "haben"'s case also, though I'm not familiar with the literature) have been thoroughly treated in generative and historical literature perfectly satisfactorily for exactly this reason.

This is a common mistake made by those trying to "disprove" the Regularity of Sound Change - they don't invest enough time in phonology to understand that phonological domains larger than the word exist. It's actually kind of funny how elementary all of the "counterexamples" critics bring up always are - you'd think people would understand that a field that's over two hundred years old would have come across auxiliary verbs at some point during that time.

Also, you've asserted that "haben"'s change is not due to analogy or interdialectic borrowing, but I'm not sure where your certainty is coming from here. Without looking more deeply into the phenomenon, at this point the data you've presented could easily be described by sound change, analogy, or borrowing, and though I'm not familiar with that data specifically, I have no doubt that one or a combination of the three fully explains the data (because, again, one or a combination of the three fully explains literally all historical data that we've found so far).

You repeat that like a dogma but don't give any logical explanation.

I mean, it's an empirical fact of language going all the way back to de Saussure and Jan Boudoin de Courtenay's insight that phonemes have regular and predictable relationships with their allophones, but luckily there's also a clear physiological explanation for the regularity of synchronic phonology as well. (It's interesting that you're so interested in "explanation" now, but we'll get to that later).

The explanation comes from a combination of the nature of the movement of the articulators, and the fact that (as de Saussure famously noted), language is a regular system composed entirely of contrasts.

Humans articulate language by moving their articulators in a surprisingly small number of regular, precise, complex movements that they have been practicing since they acquired their language in childhood.

These movements eventually become second nature to the speakers, but humans always feel a constant pull between wanting the system to be as simple as possible (leading to regular sound change - our "ease of articulation" here), and wanting the system to have enough contrasts to adequately encode meaning.

That's why phonology is regular. That's it. It's a consequence of the nature of human articulation. Every time an American English speaker pronounces a /t/ between a stressed vowel and an unstressed vowel in a word, that /t/ becomes a flap, because of these millions of practiced, unconscious movements.

Note that this also means that American English speakers literally cannot (without practice) produce a different /t/ allophone in that position in one specific word when speaking fluently. If the pronunciation of intervocalic /t/ were to change in one word, it must change in all of them (unless a different specific environment catalyzes a different regular change), because that sequence of articulator movements functions as a single unit.

Once again, it's an empirical fact that phonology is regular, and the regularity of sound change follows from it.

Also, the fact that synchronic phonology is regular is further proven by the fact that it's difficult to pronounce foreign language sounds. The mechanism is the same: we are only accustomed to pronouncing the relatively small set of regular movements in our native language, and altering those is difficult. It's just as difficult, if not more so, to spontaneously begin pronouncing one word in a way that doesn't conform to a language's phonotactics.

[–] hakase@lemmy.zip 1 points 1 day ago* (last edited 1 day ago)

you said at one point that it is a relic that used to ease the pronunciation but not anymore. Is that a statement you agree to? Because if so, when did it stop to do that and turned into a relic?

Yes, I agree with this statement, and I've already answered this question, but here it is again:

It stopped easing pronunciation as soon as the phonotactic constraints of the language changed to once more allow the sequence that was previously disallowed.

That's the answer.

And, once again, we can test for when this happens by looking for apparent exceptions to the sound rule in question (introduced later by borrowing, analogy, or subsequent regular sound change).

Once apparent exceptions appear, that indicates to us that the phonotactic constraints have changed, and that the sequence is once again being allowed in the language. At that point "easing pronunciation" no longer makes sense as a descriptor of the alternation (as in the case with the a/an alternation).

This does not mean that it is a regular sound shift. It never was. It always only effected this one word

This is empirically incorrect. It also affected my/mine in exactly the same environment, and at the same time (12th cent. to 14th cent.) because, as mentioned, sound change is regular and exceptionless in its environment.

Now, let me ask you a question.

About the a/an alternation, you say that "in every instance it occurs, it demonstrably eases the pronunciation", but you never say how it eases the pronunciation, or what that even means to you. I, on the other hand, have given you thorough explanations and theoretical underpinnings for my position at every turn.

So, if it "demonstrably eases the pronunciation", then please do demonstate it. What's the strict, rigorous, definition of "easing pronunciation" (or whatever we want to call this) that you're using here, and how is it useful? That is, how does it make useful predictions about the data?

Because currently, your definition feels like something like "it feels better to speakers" or some equally un-useful metric. If "it feels better to speakers" is your definition (which I'm not saying it is - that's why I asked), then "I would have eaten the apple" would have "easier pronunciation" than "I eaten the apple", and I think that's a bad result for your position.

My definition would probably be something like: "a process that leads to a repair of some sort (by addition, deletion, etc.) to avoid a sound string that is disallowed by a language's phonotactics".

No other process would be easing pronunciation, because all other strings would be allowed by the language's phonotactics.

And, since the sound sequence represented by the "a/an" alternation is clearly allowed elsewhere by English's phonotactics, this process cannot, by definition, be easing pronunciation.

If your theoretical framework doesn't allow something that happens, isn't that rather bad for the framework than for reality?

I suppose that depends on one's perspective, but since you're a functionalist, it certainly makes sense that you'd see it that way.

If one's framework doesn't allow something that happens, that's a good thing, because it means that the model is falsifiable, and therefore scientific. Since, as you correctly stated, all models are wrong, it should be the case that a good framework doesn't allow something that happens if you're actually doing science.

This is exactly my problem with Role and Reference Grammar, and functionalism in general - it's not falsifiable. Everything they do is descriptive - they just restate their data a dozen times in a dozen different ways and call it a day, without actually explaining anything. Nothing can prove them wrong, because they never actually say anything in the first place.

Of course they would want their models to be able to account for literally everything that could possibly happen, because they need to have room to describe it, whatever it is, and they don't care about making useful predictions.

Unfortunately, a model that is powerful enough to account for everything is, of course, also too powerful to actually do anything useful.

This is exactly why generative models are so specific and constrained - we want our models to be proven wrong by new data, so that we can revise them into better, more accurate models.

Luckily for me, though, none of the data you've brought up in this comment comes anywhere close to creating a problem for either the regularity of sound change, or generative linguistics in general.

A bit about me in return, I suppose.

I received my PhD in Linguistics in the mid-late 2000s focusing on the core subfields (generative phonology, morphology, and syntax) and historical linguistics, and then worked as an assistant professor for around five years, teaching, publishing, and supervising theses, before finally leaving the field for industry about ten years ago (though I try to stay relatively current on research).