Tuesday, May 20, 2008

OT gymnastics? The challenge, pt. deux

OK, I haven't more than skimmed the comments on the earlier opacity post, and don't have time for more than a quick stab at the second challenge right now. And keep in mind that I'm not a real phonologist, so this is just amateur hour start to finish. (I have talked to a couple of professionally-trained phonologists about this.) Still, let's blindly plunge ahead to challenge no. 2:
2. OT requires “lots of gymnastics” to account for opacity, while SPE doesn’t.
The full text of the challenge is pasted in at the bottom of this post, for quick reference. Almost all of the comments that seem to have pissed people off were all paraphrases of things others had said, but the word gymnastics was actually my choice, so I should explain it. Here's part of what Merriam-Webster's 11th says about the word:
1 … b : a competitive sport in which individuals perform … acrobatic feats mostly on special apparatus …
2 : an exercise in intellectual or artistic dexterity *my earlier philosophic study had been an intellectual gymnastic — John Dewey* *mental gymnastics*
3 : a physical feat or contortion …
I guess I had parts of all those things in mind. (By the way, Dewey was talking about being inspired by Hegel's philosophy, "no mere intellectual formula".)

In terms of fundamental architecture, theories with serial derivation can very comfortably handle cases like Canadian Raising or Tiberian Hebrew spirantization, where we have interactions among post-lexical rules. The rules are benignly learned independent of each other by the child, and ordered appropriately, perhaps without the help of opacity. The resultant opacity emerges from the particular rules and the particular ordering. The rules and ordering are boorish at best, but the interaction is sophisticated at worst. (It also mirrors how language change works, as Adam Ussishkin rightly noted.) Again, precisely this kind of pattern is common in language and easily handled, even predicted by a serial rule-based model. Additional conditions on the application of rules like structure preservation and the derived environment condition seem to represent refinements of our understanding of rule application. Maybe these refinements to rule application are gymnastics as Eric suggests but there hasn't been any improvement on the phenomena of structure preservation and derived environment effects produced by OT. So it would seem that we all need to be flexible.

In contrast, sympathy theory, comparative markedness and OT-CC really feel like gymnastics — using some special apparatus, needing dexterity and requiring contortion. To many people these feel like kludges that fail to fix a fundamental part of the basic architecture: 'classic OT' seems inherently ill-suited to garden-variety opacity. As exciting as many of us found OT in its early days, and while it provided new ways to think about some of the cases that rattled phonology through especially the '80s, early versions simply had no clean way of handling opacity — which triggered the string of contortions and acrobatics mentioned earlier. At that point, a number of people feared that we were looking at a great leap backwards: Even if the new approach could address the kinds of problems that arose with, say, late Lexical Phonology, to abandon a basic kind of approach that has served us well (since, what, Panini?), some linguists felt the new approach should have an honest way of dealing with a basic phenomenon the old approach could handle.

I'm not interested in declaring who's wrong or right; I'm just trying to understand how language works. But it seems like the people I paraphrased earlier had a point, even in light of Eric's comments: You just can't really compare some things that were "at some point or other brought to bear on opacity" with wholesale reworking of a theory to handle simple opacity.

Now that I look at challenge #3, it looks like basically the same as #2, so maybe I'll skip to #4. Stay tuned!

Challenge 2:
2. OT requires “lots of gymnastics” to account for opacity, while SPE doesn’t.
But — I hear you object to what I just wrote above — by “monostratal” Mr. Verb must mean “non-serial within a lexical-phonological stratum”, and in that sense he’s basically right. John McCarthy is without a doubt the most visible proponent of OT who accepts that at least some legitimate cases of opacity are not reducible to cross-stratal interactions, and is also the one who has developed most of the proposals to amend ‘classic’ OT to accomodate such cases of opacity, in particular those that Mr. Verb specifically mentioned (sympathy theory and comparative markedness). McCarthy’s most recent proposal, OT-CC (see his book and some of his more recent papers on ROA), is likely to be viewed by critics as just so many more gymnastics (as opposed to a proposal that makes predictions that may be right or wrong).

But is SPE really any different in this regard? What about the transformational cycle, the Elsewhere Condition, structure preservation, the alternation condition, the strict cycle condition, and so on? Each of these developments within SPE was at some point or other brought to bear on opacity, whether to help account for cases of it or to constrain the power of ordered rules to generate particularly outlandish kinds of opacity (’Duke of York’ derivations and absolute neutralization being particularly well-beaten horses). I’d like to see someone explain how any these developments within SPE are not “lots of gymnastics” while some developments within OT are. (And I’m not brushing aside the interesting work done in the 80s that attempted to reduce some of the developments within SPE to others, though I don’t believe many of these attempts were particularly successful.)
Image of Contortion Girl from here. Pretty amazing contortion, I'd say.

10 comments:

Bill Idsardi said...

If you're not a *real* phonologist, do you play one on TV?

Mr. Verb said...

Not even in the blogosphere ...

Eric said...

Please don't skip to #4, Mr. Verb -- #3 is oh-so-different from #2. The point of that is to say that derived environment effects are cases of opacity that can't be easily handled by rule ordering alone. If you're not going to tackle that one, then my challenges were all for naught.

Mr. Verb said...

Well, gee, if you put it that way ...

cassaday rassmussen said...

Yes, we really do need to hear about #3 because Eric's reply just makes my head hurt and confuses me...

Ok, so if I understand Eric's suggestions here and over on Phonoblog, then any sort of contentful theory of rule application should be considered 'gymnastics' and cause us to question the utility of serial ordering of rules. Do I get that right, Eric? I'm dense about these things so please explain this in detail.

I'm assuming that Eric will provide a reasonable explication of the above ideas as to why rule based phonology consisting of a contentful theory of rules and a contentful theory of rule ordering is so unattractive. For bonus points though, it would then need to be explained to me why a contentful theory of CON (constraints for those of you not anointed in the chapel of OT) isn't gymnastics when OT already has a contentful theory of EVAL (ranking of constraints). I think someone needs to take a tour of their own back yard...

We'll leave the questions as to whether a separate contentful theory of GEN and all of the different Correspondence Relations such as Output-Output, Base-Reduplicant, Rose and Walker's Correspondence suggestions take any OT analysis of anything from gymnastics to Jim Rose Sideshow contortionism...

Rule based phonology may not be a 10 but neither was my prom date...

Ed Keer said...

I'm assuming that Eric will provide a reasonable explication of the above ideas as to why rule based phonology consisting of a contentful theory of rules and a contentful theory of rule ordering is so unattractive.
I'm hesitant to speak for Eric, but we did once share a brain so here goes.

In my view, it requires a lot of (willful) ignorance of a huge amount of important work in the 70s and 80s to think that OT doesn’t make significant progress in many areas (duplication, conspiracies, top-down and bottom-up effects, the emergence of the unmarked, …) where SPE essentially foundered.

Eric said...

"Cassaday Rassmussen" said:

Ok, so if I understand Eric's suggestions here and over on Phonoblog, then any sort of contentful theory of rule application should be considered 'gymnastics' and cause us to question the utility of serial ordering of rules. Do I get that right, Eric?

No.

I'm assuming that Eric will provide a reasonable explication of the above ideas as to why rule based phonology consisting of a contentful theory of rules and a contentful theory of rule ordering is so unattractive.

I don't believe I ever said this, so I don't believe I have to provide a reasonable explanation for it (especially to someone hiding behind a fake name).

I think someone needs to take a tour of their own back yard...

My point exactly, so thanks for making it (or rather, for trying to throw it back at me). Here's a bare-bones run-down of the exchange so far, since it appears that you haven't been paying attention.

1. Mr. Verb claimed (and/or quoted someone who claimed) that a "monostratal" theory like OT requires "gymnastics" to handle opacity, while a rule-ordering theory like SPE does not.

2. I countered that there are plenty of similar "gymnastics" that were found to be required in SPE to handle things we expect a phonological theory to handle. (One of these is derived environment effects, which are a cases of opacity.)

3. Mr. Verb has been addressing my points piecemeal. So far, not a single peek into the backyard of SPE to identify the gymnastics.

You all can simply decide that things like the Elsewhere Condition, the Alternation Condition, Structure Preservation, and so on are just "parts of a contentful theory of rules" while analogous devices in OT are "gymnastics", but let's not pretend that you're not just being unfair if you do so. These are all add-ons to a more basic theory of rule-ordering in order to account for things one wants to account for -- nothing wrong with that, but again, let's not pretend here.

cassaday rassmussen said...

I'm still confused by some of Eric's questions...so in a 4 part missive...

First
Eric posits a challenge in the blogosphere to some guy named Mr. Verb who is well known to be anonymous but anonymous comments on the discussion are not welcomed? Furthermore, does the lack of knowledge of the origin of a question make the question not worth answering? Would it be better if I asked questions just listed as anonymous? I wouldn't be able to post the links to cute otters holding hands then...

Second
Yes, let's play fair. Why should Mr. V or anyone who reads this blog or your blog rise to your challenge when you don't respond in kind to other reasonable questions?

I know that Ed is trying to be helpful; but is quoting part of your original post

"In my view, it requires a lot of (willful) ignorance of a huge amount of important work in the 70s and 80s to think that OT doesn’t make significant progress in many areas (duplication, conspiracies, top-down and bottom-up effects, the emergence of the unmarked, …) where SPE essentially foundered."

really answering any questions that have been posed back to you? I think people would be interested in this list of 70's and 80's papers that are being ignored and the particular OT analyses that solve these problems. Come on, from Mr. Verbs splash page you should know that his 'minions' don't like to read...and as you suggest are 'willfully ignorant'.

Third
Why are you riled up about this reply and not my earlier reply? I asked a question there about how to understand Lexical Phonology with respect to SPE. I'd like to know what your answer to that question is because I don't think you have really worked out an answer to this.

I'm not trying to be antagonistic on this point but you suggest that the Elsewhere Condition, Structure Preservation, the Alternation Condition and the Strict Cyclicity Condition are proposed in SPE. I think this is likely just shorthand because anyone who has read SPE knows that these devices are not proposed there.

Here is where Eric can jump up and say 'hurrah', gymnastics have been admitted because these devices are obviously add ons to the basic SPE theory. This is not quite right though. Most if not all of the phenomena that caused people to propose the EC, SP, AC and SCC were all discussed in SPE. Most of the work that these devices do now was originally accomplished via boundary symbols (+, #, $, ;-), =-[, etc.) and the transformational cycle.

This is where 'gymnastics' does not really apply to SPE in accounting for opacity. From the beginning, rule based phonology laid out the complex issues of the interaction of syntax, morphology, phonology and phonetics. The core role of the serial derivation was the claim that intermediate levels of representation were needed to make important generalizations about the nature of language. Rules were the devices that mapped one level of representation to another. Rules were the object of study and obviously SPE is massively incorrect about the nature of rules at some levels. The original SPE mechanisms could grind out an amazing amount of the facts though.

The Derived Environment Condition and other devices actually REFINE proposals from SPE and replace them. The purpose of the Derived Environment Condition is to remove the boundary '+' as a legitimate mental category in the theory. This is the essence of REFINEMENT. SPE accounted for these facts with boundary symbols, the field decided that boundary symbols were a bad idea, removed them and part of this exercise was the proposal of the DEC to account for the data. One thing in and one thing out...

The development of OT appears strikingly different to me though. Derived Environment Conditions CAN NOT be done at all in the original formulation (PS 93, MP 93). It is only when more data was thrown at OT that it became apparent that 'opacity' effects were a problem. This is where 'gymnastics' become involved because 'non-surface true generalizations' are not the core interest of OT. Whether this is formal necessity or convenience is for the particular researcher to decide but the most common implementation of OT make claims that constraints are evaluated at the 'surface'.

This is the litmus test for 'gymnastics' in my mind. SPE and descendants privilege 'underlying' generalizations and this is why 'opacity' does not require gymnastics. OT on the other hand privileges 'surface' generalizations and this is why 'opacity' requires gymnastics.

Another distinction between rule based phonology and OT in dealing with opacity is whether there is pruning of the theory or not. SPE is not Lexical Phonology and consequently you are not allowed to mix boundary symbols and the DEC. Its one or the other and the practitioners have accepted this and generally abide by it. This is rather easy though since no one does SPE anymore (thus I'm a bit confused why there is all this talk of SPE...) There does not appear to have been any pruning in OT formally or in practice. All the devices that have been proposed to do opacity are still acceptable as far as I can tell. To be more specific, to do DEC effects in OT, which is the proposal to use, sympathy, conjoined constraints, strata, co-phonologies, positional faithfulness, alignment constraints, two level constraints? Which is it? I think Odden's paper referenced in the earlier comment says quite a lot about this issue.

Four
This is the blogosphere so there is no room for disappointment. If you're not learning something here from this exchange (regardless of whoever is involved in it) then move on and watch youtube or something. If you are learning something (even if its that 'the evil other side just doesn't understand the hard work I do' regardless of which side you are on) then keep reading and heaven forbid, comment. Don't disappoint me here Eric. I watch enough youtube already. What are the major advances that have been achieved by OT since 93'? We may get lucky and see that there is a lot of common ground between everyone reading.

Ed Keer said...

This is where 'gymnastics' become involved because 'non-surface true generalizations' are not the core interest of OT. Whether this is formal necessity or convenience is for the particular researcher to decide but the most common implementation of OT make claims that constraints are evaluated at the 'surface'.

This is the litmus test for 'gymnastics' in my mind. SPE and descendants privilege 'underlying' generalizations and this is why 'opacity' does not require gymnastics. OT on the other hand privileges 'surface' generalizations and this is why 'opacity' requires gymnastics.


Non-surface true generalizations are central to OT--that is the point of 'emergence of the unmarked'.

Constraints are evaluated 'on the surface' only inasmuch as they evaulate possible output candidates. However, constraints can assess the wellformedness of a candidate purely based on the surface form (Markedness) or in relation to the input (Faithfulness).

It's been a long time since I've thought about sympathy theory, but that seems like a case of using this basic Markedness-Faithfulness approach of OT constraints to capture opacity. If there are constraints that can measure well-formedness in relation to the input, then it's not gymnastics to propose constraints that measure well-formedness wrt some other input-like structure--an intermediate form. I was never a fan of it, but it doesn't seem like gymnastics.

As for more recent proposals, I'm out of the loop. But most of them seem to be about different ways of defining this intermediate form. And if, so completely within the mechanism of standard OT.

As to the embarrassment of riches argument, I don't know what to say. People disagree within OT. That doesn't necessarily appear to be a bad thing. Maybe it's worth contrasting phonology with syntax. McCarthy, Prince, and Smolensky seem happy to let the theory develop without too much top-down steering.

Eric said...

Eric posits a challenge in the blogosphere to some guy named Mr. Verb who is well known to be anonymous but anonymous comments on the discussion are not welcomed?

It wasn't my intention to imply that anonymous comments are not welcome -- this isn't my blog, after all. (Note that anyone is allowed to contribute full posts to phonoloblog, non-anonymously, and that anonymous comments are also more than welcome there.) My beef with your anonymity in this context concerned certain remarks aimed straight at me, which I feel less than inclined to respond to given that you're not willing to reveal who you are. That's my prerogative, just as commenting anonymously is yours.

Furthermore, does the lack of knowledge of the origin of a question make the question not worth answering? Would it be better if I asked questions just listed as anonymous? I wouldn't be able to post the links to cute otters holding hands then...

You're the only one I've been replying to at all so far, so I don't see your point. Neither one of us feels that that the other is answering the other's questions very well, so we seem to be even on that score.

Yes, let's play fair. Why should Mr. V or anyone who reads this blog or your blog rise to your challenge when you don't respond in kind to other reasonable questions?

Fair enough. I have been attempting to reply to the reasonable questions.

I think people would be interested in this list of 70's and 80's papers that are being ignored and the particular OT analyses that solve these problems.

Let's start with three works: Kisseberth 1970 on conspiracies, Kiparsky 1973 on the Elsewhere Condition and disjunctive application, and Kenstowicz & Kisseberth 1977 on the duplication problem. Conspiracies, disjunctive application, and duplication are but three theoretical issues that have generated a lot of discussion in those two decades (and beyond), and for which solutions that don't involve adding something that I would not simply refer to as "refinements" did not materialize. McCarthy's 2002 book summarizes the solutions that OT has offered to the duplication problem and to conspiracies, and cites a comprehensive set of the relevant literature (and not just that within OT). Disjunctive application is what OT constraint interaction does, and this point was made in P&S '93 (and amplified on in my 2006 paper, which I am in the admittedly slow process of scaling up into a book).

Why are you riled up about this reply and not my earlier reply? I asked a question there about how to understand Lexical Phonology with respect to SPE. I'd like to know what your answer to that question is because I don't think you have really worked out an answer to this.

I did reply to that afterwards, having read your comment on that post later than I had read your comment on this one. I got even more riled up about that one, to the point that Mr. Verb had to step in...

I'm not trying to be antagonistic on this point but you suggest that the Elsewhere Condition, Structure Preservation, the Alternation Condition and the Strict Cyclicity Condition are proposed in SPE. I think this is likely just shorthand because anyone who has read SPE knows that these devices are not proposed there.

I think I was clear enough that SPE was a shorthand for SPE-and-subsequent-developments (see my original phonoloblog post). I'm not just picking these examples out of a hat, you know: I have a well-worn copy of SPE, and I was well-trained up as an undergraduate at UCSC in graduate-level phonology before 1993, and I read, appreciate, and often defend the literature prior to 1993 (taking a look at almost anything I've written should convince anyone of this).

Most if not all of the phenomena that caused people to propose the EC, SP, AC and SCC were all discussed in SPE. Most of the work that these devices do now was originally accomplished via boundary symbols (+, #, $, ;-), =-[, etc.) and the transformational cycle.

This is certainly not true (as I said, my copy of SPE is well-worn). The AC was practically in direct response to SPE analyses, which were extraordinarily abstract and proudly so. The EC was an unformalized extension of a purely formal definition of disjunctive application, which was believed in SPE to apply only to stress-type rules. I'm struggling to see where Structure Preservation or the SCC were even remotely discussed in SPE. Some page references?

The Derived Environment Condition and other devices actually REFINE proposals from SPE and replace them. The purpose of the Derived Environment Condition is to remove the boundary '+' as a legitimate mental category in the theory. This is the essence of REFINEMENT. SPE accounted for these facts with boundary symbols, the field decided that boundary symbols were a bad idea, removed them and part of this exercise was the proposal of the DEC to account for the data. One thing in and one thing out...

What about the DEC as applied to phonologically-derived environments, which is the type of example I brought up? Certainly the '+' boundary has nothing to do with this. And I'm still waiting for some response to my assertion that "DEC effects" are opacity effects, and thus that there is no unified theory of opacity in serial, rule-based phonology, contrary to what you appear to be assuming.

The development of OT appears strikingly different to me though. Derived Environment Conditions CAN NOT be done at all in the original formulation (PS 93, MP 93).

Sure it could. If we limit ourselves to morphologically-derived environments as you have above, then we can simply allow '+' to be mentioned in the structural description of a constraint (as is sometimes found, in fact) just as SPE allowed it to be mentioned in the structural description of a rule. If your argument is that P&S/M&P should have explicitly acknowledged this in their original manuscripts, then I'm not sure we're discussing things under the same terms here.

This is where 'gymnastics' become involved because 'non-surface true generalizations' are not the core interest of OT. Whether this is formal necessity or convenience is for the particular researcher to decide but the most common implementation of OT make claims that constraints are evaluated at the 'surface'.

I can't say it better than Ed already has here: the entire purpose of constraint violability is that there are non-surface true generalizations to be accounted for. See the opening pages of my most recent article in Phonology oon this point.

SPE and descendants privilege 'underlying' generalizations and this is why 'opacity' does not require gymnastics. OT on the other hand privileges 'surface' generalizations and this is why 'opacity' requires gymnastics.

I could equally write: "OT privileges 'surface' generalizations and this is why 'conspiracies' do not require gymnastics. SPE and descendants on the other hand privilege 'underlying' generalizations and this is why 'conspiracies' require gymnastics." But have we gotten anywhere in our disagreement? Not really.

There does not appear to have been any pruning in OT formally or in practice. All the devices that have been proposed to do opacity are still acceptable as far as I can tell.

You're not reading the same literature that I am if you really believe this to be the case. Or, more realistically, I don't think there has ever been any less of a plurality of specific viewpoints at any given stage in the development of serial, rule-based phonology.

To be more specific, to do DEC effects in OT, which is the proposal to use, sympathy, conjoined constraints, strata, co-phonologies, positional faithfulness, alignment constraints, two level constraints? Which is it?

Why don't you do what everyone should do in a situation like this, regardless of their theoretical persuasion: read the relevant literature, sort out the various proposals, and decide on a rational basis? Or better yet, contribute to the literature with specific arguments for a particular way of doing things? What you appear to be suggesting is that there should be some final arbiter of "how to do X in OT" -- does such a person exist in the serial, rule-based world?

What are the major advances that have been achieved by OT since 93'?

Again, I would point you to McCarthy's 2002 book.

We may get lucky and see that there is a lot of common ground between everyone reading.

Wouldn't that be nice? I agree whole-heartedly.