I’d very much welcome comments or questions on my handout from the Manchester fringe session, held May 27th, 2015 [also: prepared introduction not on handout]. I’m not planning on turning it into a paper, though further discussion might change my mind! I also hope that this page can serve as a place for discussion of other issues raised at the workshop – links to handouts and other relevant materials would be very much appreciated.
Violable constraints in Classical Universal Phonology and beyond
Abstract: It appears to me that ‘Classical Universal Phonology’ (CUP) as a whole, rather than OT in particular, is receiving less attention from phonologists in the 21st century. I highlight some of the contributions of violable constraints to CUP, and provide an overview of some developments in generative phonology outside of CUP per se, again emphasizing the role of violable constraints, especially in formalizing grammatical learning. I conclude with some speculations about why CUP seems to be less popular these days, and about what we might expect for the future.
It seems to me that David Stampe’s name deserves to be mentioned in this context. I quote from an (OT) article of mine – Smith (1997):
“In the late 60’s Stampe proposed a set of unordered innate phonological constraints, which gradually became ordered or suppressed during the acquisition of their phonologies by children. This has signifcant parallels in OT, it seems to me, with Tesar & Smolensky’s (1994) initial state of a “degenerate stratified hierarchy of Constraints, which gradually becomes ranked during the course of the acquisition process”.
Stampe disagreed with me about any significant parallellism between his work and OT, a standpoint I disagree with. I referred in this article to Stampe (1969, 1972). The other reference is to a personal communication.
The article in which I refer to Stampe’s ideas is:
Smith, Norval. 1997. Shrinking and hopping vowels in Northern Cape York: Minimally different systems. In Frans Hinskens, Roeland van Hout & W. Leo Wetzels (eds.), Variation, change and phonological theory. Amsterdam: John Benjamins Publishing Company.
Thanks for your comment and the reference Norval!
Stampe gets a special call out in the preface to Prince and Smolensky 1993/2004, and there’s also Stampean Occultation later on in the book. On learning, I have a comparison between Natural Phonology and OT in my 1997 Language Acquisition paper (http://people.umass.edu/pater/pater1997.pdf).
This gives me a nice chance to share an anecdote. I gave my first talk to a big audience at the 1997 workshop held at Johns Hopkins, and I presented the material from that paper, including the NP/OT comparison. You can imagine my surprise when David Stampe himself asked me a question! I unfortunately have no idea what he asked me, because I wasn’t able to process his question due to being in shock. I think I probably just thanked him (at least I hope I did). Apparently he was there because he happened to be visiting Baltimore, and somebody told him about OT and the conference.
I disagree with your formulation of the goal of the CUP as that of building a system which generates all and only possible phonological systems. My issue with that formulation is that it makes reference to “possible’’ phonological systems, making any alleged solution unfalsifiable. Science cannot be about “possible’’ data.
I would rather state the goal of the CUP (or, for that matter, of generative linguistics) as that of building a system which generates a set of phonologies which is (1) large enough to include all of the attested phonologies and (2) small enough that you can prove that the resulting learning problem is solvable/tractable. In other words, the lower bound has a typological/empirical nature (it is based on the empirical typology of attested phonologies) while the upper bound has a computational/learnability nature (it is based on a learnability argument).
Once you restate the goal of the CUP this way, I think that the distinction you draw between the CUP stage and the beyond-CUP stage evaporates. The field might happen to focus on the empirical/typological lower bound during certain stages and on the computational/learnability upper bound during some other stages. But that does not say anything more than the focus oscillating between segmental and suprasegmental phonology. As a component of the generative enterprise, phonology has an intrinsically computational nature — as well as an obvious intrinsically typological/empirical nature. I agree with you that learnability/computation has become more prominent in recent times. But that is not evidence of any theoretical or paradigmatic shift, rather perhaps a sign of progress towards a goal which has remained constant over time.
Thanks for the comment Giorgio! Just to be clear – I am not trying to argue that what I defined as CUP should necessarily be the goal of generative phonology. I don’t think it’s controversial that much of the work in generative phonology, including OT, is pursuing the goal of generating all and only attested systems. The set of goals you mention, of having sufficient coverage and accounting for learning, but not worrying about restrictiveness, is certainly a possible research program, and a very interesting one, but arguments based on restrictiveness are clearly much of current practice in phonological theory, and shape many of the proposals within it.
What I see as the biggest change in recent years is that the idea of a self-contained phonological UG as an account of typology is increasingly being abandoned, and that typological generalizations are being seen as the interaction of properties of phonological grammar, the phonetics and learning. Elliott Moreton and I talk about this at some length in our Structurally Biased Learning paper (http://people.umass.edu/pater/pater-moreton-2012-offprint.pdf). Within the violable constraint + learning literature, it’s also discussed in Hayes and Wilson (2008), when they address the issue that their system can acquire unattested languages.
About attested vs. possible. I agree – I usually say attested in my own work, but I think possible is the usual formulation.
A word on violability, which is central for OT and in Joe’s Fringe talk, since it represents the alternative to Principle & Parameters. Undominated and/or inviolable constraints are commonplace in OT today. For various reasons such as the too many solutions problem or simply the recognition that some properties are universal (2 of the 4 cosntraints that the Prosodic Hierarchy was broken into by Selkirk in 1996).
The question is in which way OT with undominated/inviolable constraints is different from Principles & Parameters. And what violability means when constraints can be declared inviolable. That is, Principles are now called undominated/inviolable constraints, and parameters are violable constraints. The difference is the expression of cross-linguistic variation (violable constraints and parameters are not the same), but not violability as such.
Thanks for the comment Tobias. As you correctly point out, OT has a sort of analogue of principles, in terms of what Gen does and does not produce for a given candidate, as well as parameters, in its violable constraints. The first part of my Whither OT presentation was an overview of some of the differences between between violable constraints and parameters (inviolable constraints that can be turned off). Much of the OT literature, especially earlier work, discusses this point: you’ll find it referred to with “minimal violation”, “emergence of the unmarked”, and “nonuniformity”. McCarthy (2002: Thematic Guide) has a good extended discussion – I think under the heading of “why constraints must be violable”.