Talk:Optimality theory

From Wikipedia, the free encyclopedia

This article is part of WikiProject Theoretical Linguistics, an attempt to build a comprehensive and detailed guide to theoretical linguistics and theories of language on Wikipedia. If you would like to participate, you can edit the article attached to this page, or visit the project page, where you can join the project and see a list of open tasks.

Contents 1 Making changes 2 There 3 Theory or not? 3.1 Response: OT has too many variants and options 3.1.1 Claim 1: phonological opacity 3.1.2 Claim 2: markedness 3.2 Response by Angr 3.3 Continuation by gpkh: How strictly should one define "theory"? 4 Optimality in biology? 5 Error in example analysis? 6 Tableau in example? 7 Optimality Theory versus Connectionism

[edit] Making changes

Hey Guys! I've decided to have a shot at expanding this article a bit and I'll probably have to change what is written quite a bit. This is just because it is written in a sort of concise way, not because it sounds bad at all. Please feel free to revert back if you don't like it. AnandaLima 01:24, 12 Jun 2004 (UTC)

[edit] There

Ok, I've added some more text to the stub (is it still a stub?). Changes very welcome though: I tried to keep it simple, but I am not too happy with my plural example, so feel free to replace for a better one. Is would also be nice to have and example for the emrgence of the unmarked maybe and maybe a tableaux. May get back to it shortly. —preceding unsigned comment by AnandaLima (talk • contribs) 02:20, 12 June 2004 (UTC)

[edit] Theory or not?

The word "theory" has a well-defined meaning in science. It means "something that makes a prediction that you can test". Einstein's Theory of Relativity does that; it makes quite a number of predictions. Likewise, Darwin's Theory of Evolution does. However, Optimality Theory doesn't. If anyone thinks that it does, please list the predictions that it makes instead of gratuitously un-doing the edits. I know of no reason to use any other definition of "theory". gpkh 18 December 2005.

In OT phonology, a case of purely phonological opacity (where the optimal candidate has worse faithfulness and no better markedness than a competitor candidate) that is uninfluenced by morphological boundaries or paradigm uniformity effects would (and has indeed been claimed to) falsify OT. Sympathy theory would allow for this, but sympathy theory is not widely accepted among OT linguists, precisely because it strengthens OT to the point of unfalsifiability. OT would also be falsified by a language that prefers cross-linguistically marked structures (e.g. syllables without onsets) to cross-linguistically unmarked ones (like syllables with onsets). Here again, there are people who have proposed constraints that say things like "syllables must not have onsets", but such suggestions are generally rejected by other linguists. In OT syntax, cases of ineffability (where a given input structure has no grammatical output) present a serious challenge to OT, since OT predicts that any input will have some grammatical output. There has been some discussion of ineffability in the OT syntax literature, but I don't know enough about it to say what the proposed explanation is or whether it's been accepted by the OT syntax community. --Angr (t·c) 05:56, 19 December 2005 (UTC)

[edit] Response: OT has too many variants and options

OK. Lets look at that piece by piece. For OT to be a falsifiable theory, OT must predict something like the following: No language shall allow sentence X under conditions Y. Then, if a language allows X, OT is invalid.

Note that some particular variant or extension of OT might be a falsifiable theory, but perhaps not OT as a whole. If you want to claim that OT as a whole is a theory, you need to restrict yourself to statements that have a broad consensus. One practical problem is that OT has been applied in many ways, with many different sets of constraints, so that finding a broad consensus is dificult. I think I can safely say that there is no universally accepted set of constraints (though some constraints are shared among many linguists who do OT). If you disagree, I'd like to see a citation pointing to a list of constraints that is (a) broadly accepted, and (b) extensive enough so that practioners rarely have to step outside the list.

So, one argument I would make is that OT is not definite enough to be a theory. In support of this, I would quote from http://www-csli.stanford.edu/~sells/votproc.pdf : "...they discuss difficulties in defining the INPUT in various MP-OT approaches..." and "...in turn, this involves a rather radical difference in the architecture of the OT grammar between phonology and syntax..." These are not quotes that support the idea of a single, well-defined, stable, generally accepted theory.

A related problem is that the constraints of OT are not defined in terms of observable quantities. For instance, there is no way to absolutely know if a particular vowel is reduced. In many cases acoustic measurements yield ambiguous answers; Phoneticians don't always agree with each other.

Further, a language is not a uniform entity. People speak dialects, and can change dialect in different circumstances. Nor are dialects uniform: different people can show mixtures of features from (e.g.) Estuary English and (e.g.) Cornish English. So, the data that any test of OT must depend upon is a bit wobbly.

[edit] Claim 1: phonological opacity

Now, let's take that first claim: ...a case of purely phonological opacity (where the optimal candidate has worse faithfulness and no better markedness than a competitor candidate) that is uninfluenced by morphological boundaries or paradigm uniformity effects would ... falsify OT.

You're referring to Moreton 1996. I note that the first paper I found via google ( http://privatewww.essex.ac.uk/~artstein/publications/underspec.pdf ) that points to Moreton describes it thus: Morton (1996) shows that, given some common assumptions about Optimality Theory.... That's hardly the wording that one would use in describing a well established, well-defined theory. Read on down the page to get a description of more options within OT (i.e. disagreements about what OT actually predicts).

But, leaving aside such concerns, one basic problem is that the term "marked" is not well-defined. One often hears about alternation between marked and unmarked forms. One can objectively tell that the two forms are different, but can one objectively tell which is marked? How? Suppose a violation of the first claim was proposed: what would keep people from saying "Oh well, the other form must be marked." Often, nothing! For an example, consider http://odur.let.rug.nl/~hendriks/otwab8 , which builds an argument that the pluperfect tense is marked compared to the past tense. OK, sure, I can believe that. The trouble is, I could believe it the other way around also.

(Admittedly, some forms of markedness are less easy to believe when reversed, but my point is that there's a lot of fuzziness here.) Much the same kind of argument could be made for "faithfulness" constraints. The key question is "faithful to what?" Again, there are enough loose ends so that a purported counterexample would not kill OT. Rather, it would likely be reinterpreted within the framework of OT.

[edit] Claim 2: markedness

Now, let's take that other example: OT would also be falsified by a language that prefers cross-linguistically marked structures (e.g. syllables without onsets) to cross-linguistically unmarked ones (like syllables with onsets).

First of all, that's darned weak. It's just a statement that "all languages we've seen so far have few syllables without onsets, therefore we predict the next language to have few syllables without onsets." One can make that prediction without all the mechanism of OT. So, if OT's falsifiability rests on that, it will fall to Occam's Razor.

Second, I'll bet that any good OT guy could come up with an explanation for any particular example of a syllable without an onset. One can always invent a new constraint if necessary.

Finally, I will argue as Hercule Poirot would say from the psychology of the individual. If people really believed that they could falsify OT, someone would be out there collecting data in New Guinea. If anyone could convincingly falsify OT, it would be the biggest result in linguistics in 50 years. If you couldn't get an endowed professorship out of that, well... Anyhow, given the lackluster attempts to disprove OT, I must conclude that people either believe it is 100% correct, or have a gut feeling that their data would just get absorbed.

So, overall, I don't buy your examples. Gpkh 00:21, 21 December 2005 (UTC)

[edit] Response by Angr

Yeah, I noticed. But nothing you've said is a unique problem of OT. If one follows your argumentation, then there has never been a single linguistic theory proposed at all. Perhaps linguistics uses the word "theory" in a different way from Karl Popper. Nevertheless, to me the fact that OT opponents like Idsardi have claimed to have falsified OT is a pretty good indication that it is falsifiable. There are certainly extensions of OT like sympathy theory or Alderete's "anti-faithfulness" that make the theory unfalsifiable, but these extensions are not widely accepted. If anything, OT's biggest weakness is not that it's unfalsifiable but that it's all too easily falsifiable: OT doesn't predict opacity or ineffability, and yet both seem to occur. OT predicts that CV is the unmarked syllable in all languages, yet the Australian language Arrernte has (if I remember correctly; I have no source to hand) been claimed to have VC as the unmarked syllable. Chomsky argued that OT predicts all words should be /ta/, which is also clearly not the case. (Pro-OT people say Chomsky apparently completely misunderstood OT when he said that, though.) This is why OT opponents still cling to Lexical Phonology and even SPE phonology, because those "theories" really do allow absolutely anything and make no falsifiable predictions. Perhaps you're right, perhaps OT doesn't meet all the criteria of a falsifiable theory. But it comes a damn slight closer to it than any other "theory" of linguistics. --Angr (t·c) 06:13, 21 December 2005 (UTC)

[edit] Continuation by gpkh: How strictly should one define "theory"?

I think that in practice, one shouldn't interpret "falsifiability" too strictly. In the social and life sciences, every theory is false (in the strict sense) because none are perfect*, but some are still better than others. So you can take the best theory(ies) and keep them around and consider them to be tentatively true. The rest, you consider to be false and throw out. Then, when a new theory comes along, you can compare it to your current champion.

(* My brother is a biologist, and he tells me that "In carefully controlled experimental conditions, genetically identical animals will do whatever they darn well please."

So, In practice, you can't afford to think true/false; instead one thinks better/worse. Then instead of "falsifiable" in the strict sense, you'll call it a theory as long as it is testable or rankable against other theories. One has to see if theory A works better or worse than theory B.

It's really just a labour-saving scheme. By ranking theories and throwing out the ones that don't work well, you free up smart people to work on other things.

Taking this weaker form of falsifiability, there are theories in linguistics. Not big, grand theories, but you can find testable predictions. For example:

• Probabilistic grammars of various kinds in computational linguistics can do a fairly good job of predicting human grammaticality judgements.
• There are models in intonational phonology that predict (for example) numerical values for a speaker's fundamental frequency based on combinations of lexical tones in Chinese. (I've done one.) Again, there's not really a "hard" pass/fail falsifiability, but you can say A is better than B, and C is pretty good, but D is terrible.
• There is work combining aerodynamics, modeling of the tongue and the like that promises to explain coarticulation. This is sort of a follow-on from Browman and Goldstein who asked the question "can coarticulation lead to phenomena that are misinterpreted as phonological changes?" They set out to test the hypothesis that (e.g.) vowel deletion is a phonetic process. Again, maybe not quite a theory in the strict sense, but that line of research is getting close.
• There is a lot of work in Experimental Psychology that addresses some linguistic questions, some of which involves models (theories?) of how the brain interprets sounds. It involves fairly careful hypothesis testing. I don't follow it quite closely enough to know if it's a "theory" but it has that flavour.
• And, for that matter, OT applied to second language learning might well breed a theory. Suppose you start the interlanguage as a copy of your first language, and then modify it by minimal exchanges of constraints: just swap one neighbouring pair at a time. Given some work, that kind of model might make testable predictions. It could be made into a theory of language learning.
• Speech synthesizers and speech recognition systems embody theories. Awful, messy, complicated theories, but they do make predictions about language.

So, it's not that linguistics cannot breed scientific theories. Consequently, I don't think it's appropriate to use the word too loosely.

One possibility is to look at OT as a cloud of variants. All related, but not identical. An individual researcher might have a version of OT that is a testable theory that is more-or-less true, another might have one that is false, while a third might have one that's not testable/falsifiable at all. In that case you can't say that OT is a theory: it's (plural) theories along with some things that aren't theories. It wouldn't be falsifiable because you probably cannot prove all the variants to be false at once.

In that view of the situation, it is strictly correct to say that OT is not a theory, but perhaps that's not the best way to phrase it. Do you think a cloud or cluster metaphor is appropriate?

Ta-Ta!

Very likely. I agree with what you've said above, but I hope you understand that I'm bound to react sensitively when someone comes along and says OT makes no falsifiable predictions, when I've spent several years writing a book (cited in the references of this article) arguing for a version of OT that is very constrained and testable. But of course, that's your point: it's a version of OT, and there are many other versions of OT that are unconstrained and untestable. --Angr (t·c) 05:59, 22 December 2005 (UTC)

[edit] Optimality in biology?

I'm very surprised that this article at a first glance does not seem to mention biology at all. Optimality criteria are widely used on studies of behaviour, and it has often been said that the accuracy of predictions made is some of the best evidence of evolution, with the quantitative agreement often coming close to that encountered in the physical sciences. I can think of Alex Kacelnik as one prominent author in this field, but there are many others. - Samsara 21:06, 13 January 2006 (UTC)

Perhaps a separate article on the biological subject would be called for. --Angr (tɔk) 21:16, 13 January 2006 (UTC)

[edit] Error in example analysis?

Correct me if I'm wrong but I don't think the constraint "one cannot have two /s/ sounds in succession in English" actually exists or it is misleading. Although most words spelled with double s's only pronounce one, some words like vicissitude and admissible, seem to be able to be pronounceable with two s's (one corresponding to the coda of the one syllable, the other to the onset of the next). It seems possible to have two successive s's in English if they are in different syllables in this manner. Should the rule be adjusted for this? Perhaps "a syllable in English cannot have two successive /s/ sounds". (I guess another possibility is that the rule is low ranking, and thus vicissitude and admissible can allow for its violation in favor of another constraint.)

Vicissitude and admissible are pronounced with only one /s/ sound in a row. English doesn't have phonological geminates except at the boundary of a compound, like bookkeeper. Angr (talk • contribs) 11:24, 1 May 2006 (UTC)

[edit] Tableau in example?

Could we show a little tableau for the example? This would exemplify tableaus as a means of Optimality Theoretic exposition, and give a couple very common constraints (*CC. Dep, Max). In addition, a short section on the candidate set and GEN could be added. I believe this would be appropriate for the entry, assuming that we can find the appropriate unicode finger symbol. mitcho/芳貴 18:23, 19 June 2006 (UTC)

Mr. Pointy Finger (as it was always called when I was in grad school) is at U+261E (☞ or ☞). Adding a tableau is a good idea! User:Angr 18:43, 19 June 2006 (UTC)

Here's an example tableau: for some reason I can't get the dotted or dashed lines to show up at 1px... In addition, an explanation of the thinking process, a definition of these constraints, and one more tableau (classes) is what I'm proposing we add. mitcho/芳貴 22:02, 20 June 2006 (UTC)

Tableau for 'cat+s'

/kæt + s/	Dep	Max	*ComplexCoda
kæt		*!
☞ kæts			*
kætəs	*!

Note also that the article says 'cats' 'passes all the markedness constraints' -- how incredibly wrong is that? We need tableau quickly. I'll keep playing with these. Does someone want to write neutral definitions for *[ss], Max, *Struc, and Dep, preferably with citations? mitcho/芳貴 02:22, 22 June 2006 (UTC)

I wouldn't use *Struc for the low-ranked constraint cats violates, it's too controversial and too hard to understand. I'd use something like ComplexCoda instead. User:Angr 07:26, 22 June 2006 (UTC)

Updated tableau and added tableau for classes. With some descriptive text for the process and constraints, we should be good to go. mitcho/芳貴 18:09, 2 July 2006 (UTC)

Tableau for 'class+s' given ranking from 'cat+s'

/klæs + s/	*[ss]	Max	Dep	*ComplexCoda
klæs		*!
klæss	*!			*
☞ klæsəz			*

Instead of *ss let's go for *SS, where S stands for any strident fricative or affricate. Thus

• SS - A complex coda consisting of two strident fricatives or an affricate and a strident fricative is prohibited.

Compare Staffs, wolves, heaths, and writhes. Here no insertion is present due to the codas consisting of a nonstrident fricative plus a strident one.

I'd also like to point out that the underlying form of the English plural, the possessive and the 3rd sing are all /z/. It's easy to get sidetracked by spelling. This sound change is due to fricatives becoming voiced in the coda position of unstressed syllables in Middle English.

Other than that, I must say that the fact that DEP and MAX are reversed going from the first tableau to the second must be resolved. Optimality theory is the theory of intuition and intuition doesn't change from one instance to another. --70.56.73.70 09:46, 10 September 2006 (UTC)

[edit] Optimality Theory versus Connectionism

The beginning of the article does not explain how OT and connectionism are different.

A key difference between OT and connectionism is the former's reliance upon the idea of strict domination. Strict domination essentially states that, for any pair of ranked constraints A >> B, a candidate which violates A is always - without exception - less optimal than a candidate that does not violate A, regardless of the number of times either candidate violates B. Thus, if some candidate C1 has five (or ten, or ten thousand) violations of B and some other candidate C2 has one violation of A, and A >> B, then C1 is the more optimal of the two. Even if candidate C1 violates some lower constraints C, D, E, etc., which are ranked A >> B >> C >> D >> E, and C2 has no violations other than of A, C2 is still less optimal because of its violation of the highest-ranking constraint. joo-yoon 19:06, 14 February 2007 (UTC)