Monday, May 19, 2008

Parallel vs serial phonology

Phonoloblog is one of my favorite linguistics blogs and a number of us around here read it. Over there, Eric has posted a pretty sharply worded "challenge" to Mr. Verb "to defend" claims from somebody else I paraphrased this way a while back in a post:
Opacity is ubiquitous in human language, and earlier theories of phonology could deal with it easily. It’s hard to see why those advantages have been abandoned for an approach that can’t handle opacity without lots of gymnastics, if at all, for benefits that don’t look all that great.
Here are the challenges:
  • OT is by definition monostratal.
  • OT requires “lots of gymnastics” to account for opacity, while SPE doesn’t.
  • SPE(-and-subsequent-developments) “could deal with [opacity] easily”.
  • The benefits of OT over SPE “don’t look all that great”.
I don't mind having a go at these challenges, even though my paraphrase truly was representing somebody else's positions as I recalled them. But first, a story (this version from here):
Farmer Jones got out of his car and while heading for his friend's door, noticed a pig with a wooden leg. His curiosity roused, he asked, "Fred, how'd that pig get him a wooden leg?"

"Well, Michael, that's a mighty special pig! A while back a wild boar attacked me while I was walking in the woods. That pig there came a runnin', went after that boar and chased him away. Saved my life!"

"And the boar tore up his leg?"

"No he was fine after that. But a bit later we had that fire. Started in the shed up against the barn. Well, that ole pig started squealin' like he was stuck, woke us up, and 'fore we got out here, the darn thing had herded the other animals out of the barn and saved 'em all!"

"So that's when he hurt his leg, huh, Fred?"

"No, Michael. He was a might winded, though. When my tractor hit a rock and rolled down the hill into the pond I was knocked clean out. When I came to, that pig had dove into the pond and dragged me out 'fore I drownded. Sure did save my life."

"And that was when he hurt his leg?"

"Oh no, he was fine. Cleaned him up, too."

"OK, Fred. So just tell me. How did he get the wooden leg?"

"Well", the farmer tells him, "A pig like that, you don't want to eat it all at once!"
This Phonoloblog post is kinda like that pig … so good you don't want to eat it all at once. Let's take the first of those four legs for now: parallelism versus serialism. Eric writes this below his list of challenges:
I’m pretty sure that it’s safe to assume that “earlier theories of phonology” refers to serial, rule-based generative phonology in the SPE-and-subsequent-developments sense, and that “any monostratal theory (one without stages of derivation)” and “an approach that can’t handle opacity without lots of gymnastics, if at all” refers to Optimality Theory. Correct me if I’m wrong.
My formulation refers not to OT but to "monostratal approaches". (I mention OT in a broader connection earlier in the post. Elsewhere on this blog I've used the term 'mainstream OT' specifically to mean monostratal OT.) Such approaches include 'classic OT', Declarative Phonology, and others; it's of course all monostratal approaches but not all OT approaches that have trouble with opacity. So, the first challenge seems to be a misunderstanding.

Eric continues with the details of Challenge 1:
1. OT is by definition monostratal.
In both published papers and (perpetually) forthcoming books, Paul Kiparsky and Ricardo Bermúdez-Otero — and many others, certainly — have developed and argued for a marriage between the basic assumption of Lexical Phonology and Morphology (that a grammar is multistratal) and the basic assumption of Optimality Theory (that an input-output mapping is defined by applying a ranked constraint hierarchy to a set of candidate outputs derived from the input). (The possibility of such a marriage was suggested, but not followed up on, in McCarthy & Prince (1993).)
It's true enough to say that stratal OT was mentioned as possible early on (including by Prince & Smolensky 1993), but it's not simply that it wasn't followed up on: A major theme in classic OT has been that serial derivation was a really bad idea, from P&S '93 up until almost the present day.* It's not hard to find places where Optimality Theorists appear to conflate parallel evaluation and OT, though, like in this quote from P&S 1993:137 [in the ROA version]:
When all of the relevant constraints are assessed in parallel, as in Optimality Theory, an entire completed parse is subject to evaluation. … [A] number of further cases of crucial parallelism are discussed in McCarthy & Prince 1993. The crux of the matter is that the grammar must determine which total analysis is wellformed— a task impeded by the use of serial algorithms to build structure step-by-step.
That's Part I, too much thinking for one day. But there's plenty more meat on this pig!

* I've always liked the index entry in McCarthy's Thematic Guide: "serial derivation, problematic", with a string of page references.

Image from here, continuing the linguistics/music connection thing.

7 comments:

Anonymous said...

Well, at the very least the *unmarked* version of OT obviously IS monostratal. The reason P&S and others slip into thinking that is that this is so. It's not logically this way, but in practice it certainly has been for the big players. And I remember hearing sharp criticism of Kiparsky for not really pursuing OT precisely because he introduced strata.

Anonymous said...

One nagging question is how monostratal is monostratal? In other words, if we're modeling phonology (the system of abstract units of sounds) are we also modeling phonetics or do we really have to care about phonetics as part of the single phonological stratum. OT proponents could win fans by staying off the phonetic ground (Halle 19anything) because we might suspect that the boundary conditions and functions of motor control and muscle movement fall under the purview of a different set of constraints (said in a general way).

The second part to the question how monostratal is monostratal has to do with inside (abstract) phonology. Assuming the newest rebranding of McCarthy's OT is more on the mark than McCarthy's OT of the 1990s, we might posit two levels splitting phonology benignly between cyclic and noncyclic word level, etc. But we knew this kind of division occurred without the help of OT. More importantly, though, if opacity can be shown to occur within levels (e.g., Hebrew, Idsardi Roca volume) then we need more than one level inside levels. Well, then, how many levels are there? OT is on the verge of discovering the answer to this question that was raised by anyone slogging through Halle & Mohanan's analysis of English in LI.

Reminds me of a cigarette ad of years gone by: "You've come a long way, baby."

Anonymous said...

Re: "It's true enough to say that stratal OT was mentioned as possible early on (including by Prince & Smolensky 1993), but it's not simply that it wasn't followed up on: A major theme in classic OT has been that serial derivation was a really bad idea, from P&S '93 up until almost the present day."

The relevant ideas mentioned in P&S '93 and M&P '93 are hugely different. The idea in P&S '93 was harmonic serialism, which has been followed up on by McCarthy in his most recent work. The idea in M&P '93 was to distinguish a "suffixed-stem level" from a "prefixed-stem level" in Axininca Campa (and then they focused with the suffixed-stem level phonology throughout the manuscript, which is what I meant when I said they didn't follow up on it).

The "really bad idea" arguments have virtually always been about harmonic (within-level) serialism, never really about multiple levels or strata. There was an early push to try to achieve level ordering effects with alignment constraints, and when that largely failed with output-output correspondence, and in the latter case one could hardly really support the radical view that it wasn't just level ordering in a very thin disguise.

Anonymous said...

Wow! Theoretical phonology content in the comments on Mr. V!

What is the world coming to?

I think the second anonymous post nails a very important question for this particular SPEOT cage match. How monostratal is monostratal? Or, let's put it another way, How derivational is derivational? This seems to be an important distinction for many.

Looking at Eric's post here, it appears to me that there is a suggestion that derivation is like goldilocks, not too much, not too little but just right!

No derivation at all, straight lexicon to phonetics with no intermediate levels of representation appears to be how some interpret 'monostratal'. Eric objects to OT necessarily having this characteristic. Eric is correct on this point about the formal system of OT. I'm not so sure that everyone understands this point though because I have heard multiple OT practitioners object to the assumption that there are separate morphology and phonology components. If a stratal division between morphology and phonology is not accepted then I don't really know how an intramorphology stratal division can be accepted. It is the common practice and rhetoric in many cases that cause this confusion between "fundamental principles... and convenient assumptions" (Odden 2008:62).

Eric also seems to suggest that too much derivation as in serial rule based phonology or SPE-and-following-developments is not good either. This is the only way that I can interpret the throw back of 'gymnastics' towards the non-OT enthusiast positions expressed by one Mr. V. As Eric asks Mr. V to defend his post against OT it would be nice to see these arguments against serial derivations. Why is a serial derivation so bad?

A very popular position appears to be 'just right' as in Lexical Phonology. Now, is the Lexical Phonology of the 1980's SPE or not? Halle had his hands in some of it but it was primarily driven by Kiparksy. There is also this whole Halle/Vergnaud cyclic/non-cyclic vs. lexical/non-lexical distinction that I don't think has ever been resolved. Is this still SPE or not? And how does Prosodic Morphology (1986) work into this mix?

The field of phonology coming to terms with how to view Lexical Phonology appears to me to be one of the real questions in this SPEOT cage match. It seemed to me that one of the 'bad theories' that OT was replacing in the early 90's was Lexical Phonology (Requiem anyone?).

We've now gone 15 years and it looks like Lexical Phonology had a lot of things right. Stem vs. word level (or maybe cyclic vs. non-cyclic), Structure Preservation and Derived Environment effects all appear to be hot topics again. But is the "new" Stratal OT actually OT? Is it actually Lexical Phonology? What is it?

I think the answer to this question tells more about linguistic clannishness than anything else. If you were an early adopter of OT or trained in OT as a grad student, you likely view Stratal OT as a novel development of OT as the theory evolves. Or possibly just a rehash of M&P 1993's stratal analysis of Axininca Campa. You also likely get annoyed when the full beauty of OT is not appreciated.

If you were not an early adopter of OT or were trained in something else in grad school (gasp! we might have to go back in time for this case) then you are likely snickering that OT is becoming to get its 'due'. You also likely have two or three hardback copies of SPE sitting on your desk right now (Halle and Vergnaud, Monhanan's dissertation and LSLT are acceptable substitutes).

If you really love phonology though, you're reading this and trying to think about what we have right from SPE, LP, PM, OT, Government Phonology, Declarative Phonology, Panini, Sapir, Bloomfield, the Structuralists, Postal, Joost, Bloch and also what we have wrong. Because, remember we're all wrong and we'll make a huge contribution to our understanding of phonology when we can make something less wrong... not necessarily right but less wrong.

Enough of this group hug... when do we get to eat the second leg of the pig Mr. V?

Final note: 'monostratal' in any theory is bad.

Really final note: Follow above link for Odden reference.

Anonymous said...

Looking at Eric's post here, it appears to me that there is a suggestion that derivation is like goldilocks, not too much, not too little but just right!

Is it too much to ask of a theory that all and only the empirically consequential assumptions it makes actually have empirical consequences?

Eric also seems to suggest that too much derivation as in serial rule based phonology or SPE-and-following-developments is not good either.

Nope, I have not suggested that.

It seemed to me that one of the 'bad theories' that OT was replacing in the early 90's was Lexical Phonology (Requiem anyone?).

No, it's always been serial rule ordering vs. parallel constraint evaluation within a lexical level. But, of course, once you replace rules with constraints, you may have to rethink the principles of Lexical Phonology (which is what e.g. Kiparsky has been doing for years).

But is the "new" Stratal OT actually OT? Is it actually Lexical Phonology? What is it?

What kind of question is this? Can a theory not be developed, and borrow ideas from another theory the domain over which it does not compete?

If you were an early adopter of OT or trained in OT as a grad student, you likely view Stratal OT as a novel development of OT as the theory evolves.

OK, what the hell? Do you even know what you're talking about? Point me to just one example of work in Stratal OT that does not recognize work in Lexical Phonology from the 1980s and 1990s. You've got to be kidding me with these pseudo ad hominem remarks (especially from behind a fake name).

Mr. Verb said...

OK, everybody, please calm down. I've been very reluctant to tackle this stuff anyway and I really don't match to be party to some kind of pissing match, let alone end up as the referee.

For the record, one more time, the original points that triggered this whole thread *really were* things I heard people say.

Anonymous said...

My apologies, Mr. V. I'm easily excitable when it comes to these issues, as anyone who knows me will tell you.

One more comment about this:

Looking at Eric's post here, it appears to me that there is a suggestion that derivation is like goldilocks, not too much, not too little but just right!

This reminds me of Morris Halle's comment after Andrea Calabrese's talk at the Phonology 2000 symposium at Harvard/MIT in 1999: in Morris's opinion, Andrea's proposal was "just the right mix" of rules and constraints...