Author Topic: sound change rules  (Read 12350 times)

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #15 on: October 29, 2014, 12:23:45 PM »
Well, yeah...of course they predict change. That's what makes them stochastic.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 1700
  • Country: us
    • English
Re: sound change rules
« Reply #16 on: October 29, 2014, 12:28:04 PM »
Do they work using post hoc analyses or can they actually perform without hints? For example, have any successfully explain the great vowel shift or grimm's law?
Welcome to Linguist Forum! If you have any questions, please ask.

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #17 on: October 29, 2014, 12:37:26 PM »
Explaining the great vowel shift or grimm's law would be, by definition, post hoc. But, to answer what I think you're asking...yes, there are non-teleological analyses of both. This is a fairly mainstream thread in contemporary phonology.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 1700
  • Country: us
    • English
Re: sound change rules
« Reply #18 on: October 29, 2014, 12:43:34 PM »
Quote
Explaining the great vowel shift or grimm's law would be, by definition, post hoc.
Not in a close system where the results are not included as training data or as an endpoint for the analysis. If you can feed in (only) Early Middle English and have a algorithm output the results after the Great Vowel Shift, I'd be very impressed. If not, then I don't see how this is "predicting".

I know that a lot of work is done to look at why these things happen. But I don't get the impression it's yet reliable or predictive. Instead, it suggests reasonable post hoc analyses for observations and suggests that some of these changes (or maybe all) are motivated by the properties of the system before the change.

If it's actually beyond that point now, I should update the curriculum for the class I teach, especially regarding the section on sound change.
Welcome to Linguist Forum! If you have any questions, please ask.

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #19 on: October 29, 2014, 12:50:40 PM »
I'm really having a hard time following you. Do you understand what a stochastic model is? You seem to be treating diachronic structure as a deterministic system. Do people actually claim that?

Offline freknu

  • Forum Regulars
  • Serious Linguist
  • *
  • Posts: 397
  • Country: fi
    • Ostrobothnian (Norse)
Re: sound change rules
« Reply #20 on: October 29, 2014, 12:55:35 PM »
Do they work using post hoc analyses or can they actually perform without hints? For example, have any successfully explain the great vowel shift or grimm's law?

Wouldn't unstressed /ai/ → /eː/ already be a "stochastic" (why is it random?) prediction?

  • /ai̯/ is reduced due to degeneration of unstressed syllables (stress)
  • /a/ is the primary component due to /ai̯/ being an off-glide (strength)
  • /i/ is the secondary component, causing mutation of the primary sound (harmony)
  • phonemic length is preserved (duration)

The actual phonetic realisation may vary, but the only contrast needed in this case is back /a/ versus front /e/, diphthong /ai̯/ versus monophthong /eː/.

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #21 on: October 29, 2014, 12:58:11 PM »
(why is it random?)

In this kind of modeling, "random" just means underdetermined. It doesn't mean absolutely random.

Offline freknu

  • Forum Regulars
  • Serious Linguist
  • *
  • Posts: 397
  • Country: fi
    • Ostrobothnian (Norse)
Re: sound change rules
« Reply #22 on: October 29, 2014, 01:22:35 PM »
(why is it random?)

In this kind of modeling, "random" just means underdetermined. It doesn't mean absolutely random.

I see.

So that would then be the answer to my next question, how could you possibly predict which change would occur? But since there are not enough constraints (stochastic) you can merely predict the likely outcome, not the exact outcome?

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 1700
  • Country: us
    • English
Re: sound change rules
« Reply #23 on: October 29, 2014, 01:50:08 PM »
Quote
I'm really having a hard time following you. Do you understand what a stochastic model is? You seem to be treating diachronic structure as a deterministic system. Do people actually claim that?

I understand that it's based on probabilities. And I guess in that there's no categorical distinction between "effective" and "ineffective", unless you have a metric for a specific purpose.

However, my main point is the following: we can't claim to "understand" the past (in the sense of "why") if we can't predict the future (with a reasonable level of probabilistic accuracy).

To rephrase:

We CAN make explanations such as: "if this happened [eg, great vowel shift], then it was probably this reason"
But we can't really make explanations such as: "this is what the language was like, and therefore the next logical step was this [eg, great vowel shift], even with some probability (where that probability allows us to, a fair amount of the time, make the right guess)"
Welcome to Linguist Forum! If you have any questions, please ask.

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #24 on: October 29, 2014, 07:16:48 PM »
I see.

So that would then be the answer to my next question, how could you possibly predict which change would occur? But since there are not enough constraints (stochastic) you can merely predict the likely outcome, not the exact outcome?

Right, that's the nature of non-deterministic systems. A basic model of trait inheritance will be able to tell you why I have a son with red hair and a son with dark hair. Hoping to also explain why it's the older one in particular who got the red hair, however, requires a fairly fundamental misunderstanding of the nature of genetics.

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #25 on: October 29, 2014, 07:31:57 PM »
To rephrase:

We CAN make explanations such as: "if this happened [eg, great vowel shift], then it was probably this reason"
But we can't really make explanations such as: "this is what the language was like, and therefore the next logical step was this [eg, great vowel shift], even with some probability (where that probability allows us to, a fair amount of the time, make the right guess)"

You've got it exactly backwards. Stochastic modeling allows us to make predictive estimations about the next state of a system given its current configuration. It *never* allows us, however, to look back at a transformation and assign it one cause from among many. That's just not how non-deterministic systems work.

I gather you're just not familiar with this work. That's fine, of course, but if you really find the possibility of what I'm describing so hard to believe, there's no reason for you to take my word for it. "Evolutionary Phonology" by Blevins (2004) is probably the easiest place to start; I've pointed you there a few times before during similar conversations, I believe.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 1700
  • Country: us
    • English
Re: sound change rules
« Reply #26 on: October 29, 2014, 08:04:37 PM »
Hmmm... can stochastic models actually answer a "why" question at all though? They seem to simulate and number crunch, but to actually say "why" we'd need to understand. Or is the idea that the algorithm is a kind of understanding?

An interesting case is Hidden Marchov Chains. They work fairly well for implementing certain aspects of language computationally. But they are essentially meaningless and uninformative-- they work with hidden statistical operations and they give a reasonable output. But they don't really help us "understand". Unless you mean that the why question is answered by an algorithm in the general case.
Welcome to Linguist Forum! If you have any questions, please ask.

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #27 on: October 29, 2014, 09:30:29 PM »
Hmmm... can stochastic models actually answer a "why" question at all though?

Think about what you're asking here. "Why" is a question of causation, but the entire point of calling something non-deterministic is to say that its basis of cause cannot be exhaustively specified. That's not a limitation of our knowledge; it's an intrinsic part of how non-deterministic systems function.

For this kind of phenomena, any coherent model of cause is necessarily going to be stochastic. For example, something like Evolutionary Phonology presumes variation as an inherent property of messy, material speech, but then goes on to demonstrate that basic facts of perception and articulation make some transformations more likely and some configurations more stable. It's a Markov model in every sense, but to call it "meaningless and uninformative" strikes me as pretty bizarre. I mean, good gravy, it's a comprehensive, systems-based explanation of phonological change! What more could you possibly be hoping for?

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 1700
  • Country: us
    • English
Re: sound change rules
« Reply #28 on: October 29, 2014, 10:52:19 PM »
Quote
That's not a limitation of our knowledge; it's an intrinsic part of how non-deterministic systems function.
Ok, so then why would that be an answer to the "why" question? :)

In other words, I think I agree.


Quote
It's a Markov model in every sense, but to call it "meaningless and uninformative" strikes me as pretty bizarre.
It depends on what we mean by "meaning" or "informativeness". I'm talking about it from the perspective of the researcher understanding from it. This gets into issues like the Chinese room-- personally I'd rather actually understand Chinese than be able to perfectly imitate it. If it turns out that linguistics (or parts thereof) are just probabilistic, then I'm not too interested in studying them (I wouldn't mind showing it and moving on though). [I won't ever mind doing descriptive linguistics, though, I suppose!]
Welcome to Linguist Forum! If you have any questions, please ask.

Offline MalFet

  • Global Moderator
  • Serious Linguist
  • *****
  • Posts: 282
  • Country: us
Re: sound change rules
« Reply #29 on: October 29, 2014, 11:20:13 PM »
Ok, so then why would that be an answer to the "why" question? :)

In other words, I think I agree.

Huh?

It explains why sounds change (for example, from *e->a) by describing the system pressures that favor the transformation, the system pressures that disfavor it, and the mechanisms of change itself. What more could you possibly hope for?

It depends on what we mean by "meaning" or "informativeness". I'm talking about it from the perspective of the researcher understanding from it. This gets into issues like the Chinese room-- personally I'd rather actually understand Chinese than be able to perfectly imitate it. If it turns out that linguistics (or parts thereof) are just probabilistic, then I'm not too interested in studying them (I wouldn't mind showing it and moving on though). [I won't ever mind doing descriptive linguistics, though, I suppose!]

If a system is built on probabalistic mechanisms, "understanding it" requires building a representation that includes probability. There's nothing particular to linguistics about that. It's just a basic issue of epistemology. The comparison to questions about consciousness is completely missing the point.

Edit: type-o change-o
« Last Edit: October 29, 2014, 11:22:27 PM by MalFet »