I’m clearly not above fighting on the internets, even with folks who choose not to reveal their true identities while making flippant remarks about the vices and virtues of competing theoretical frameworks. I’m referring, of course, to the discussion with Mr. Verb & friends that I initiated here, with the remainder of the discussion on Mr. Verb’s blog (follow the links in the comments section of my post).
In the fourth part of his response, Mr. Verb correctly points out that I started the nastiness with this remark (emphasis added to the quoted “quip”):
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.
And I’m the first to admit that I continued in the same nasty vein in the comments sections of Mr. Verb’s responses to my four challenges, with particular vitriol reserved for a certain “Cassaday Rassmussen” (who is doubtless an extraordinarily cute though cheeky little devil, much like the sea otters s/he loves so much). I was apparently inappropriately offended at Cassaday’s combination of willingness to be just as nasty as I was and unwillingness to be identified. (I am somehow less offended by Mr. Verb’s anonymity, given the clues he leaves here and there that help to narrow the field of possibilities down considerably.)
ANYWAY, now that Mr. Verb’s multi-part response to my multi-part challenge is over (save for an appendix that promises to tie up some loose ends), I figured it was an appropriate time to summarize some of my thoughts on the matter and the episode, beyond making rabid remarks in the comments area of an anonymous blog. Read on (and comment, anonymously or not!) if you’re interested, navigate away (to youtube, for example) if you’re not.
Part 1, or: the grass is always browner
I’d like to start at the end, more or less, with a quote from Mr. Verb’s final installment:
the complaint you can hear from people who’ve grown disillusioned with OT (or never accepted it) is that a huge amount of the work in the framework has tended to theory-internal tinkering, rather than research aimed at more general progress on how sounds work.
It’s certainly true that you can hear this. I hear it all the time, even from people who are OT proponents — hell, you’ll even hear me complaining like this a lot. There are lots of people who don’t like “theory-internal tinkering” without some sense of progress toward understanding something significant about the subject matter (or, at least, the damned theory). All I question here is whether this type of complaint is (justifiably) OT-specific. Here’s an experiment you can try if you’re near your books (remember those?): grab a copy of any theoretically-oriented conference proceedings before 1993, and browse through the phonology papers. Honestly now: are all of them (or even a significant majority of them) examples of “research aimed at more general progress on how sounds work” as opposed to “theory-internal tinkering”? Honestly?
I was an undergrad at the time, but with significant training in theoretical linguistics (thanks, UCSC), and I was reading this work. A whole lot of it was “theory-internal tinkering” of just the type that gets people’s goats about that “huge amount of work” in OT. (You cite your examples and I’ll cite mine.) Still, I was a budding phonologist and I gave (and in many cases still give) a lot of leeway to much of this work, as opposed to, say, work on syntax redefining the ECP for like the bajillionth time. (There were sometimes facts at stake, but why should I care? It was just not my cup of tea.) This is all by means of saying that we all have these kinds of gut feelings about the value of the work we read, and the strength of those reactions is quite often correlated with our existing predispositions to (or knowledge of) things like the theory, the subject matter, even the language, the author, the author’s institution, the journal it appears in, whether we received a personalized copy of the unpublished manuscript, and so on.
(That last example leads me down a small tangent, but I think a relevant one: the advent of OT coincided with the widespread use of the internet to disseminate our work, and I often hear OT detractors say things like “you can publish anything in OT because the Rutgers Optimality Archive has no peer-review process”. Aside from the fact that this sort of comment is a contradiction in terms (posting to ROA is emphatically not publication), there’s that (willful) ignorance again: before ROA and similar repositories like lingbuzz, you found out about the most current, as-yet-unpublished work only if you ran around in the right circles, scored a copy from a friend who did, and so forth. People complained all the time about how much citation of unpublished work there was in linguistics, how you had to be a member of the right clique to get your hands on it (usually the clique one heard about was “MIT-related”), and so on.)
Part 2, or: what can you expect from a theory?
Mr. Verb continues the paragraph quoted above:
One of the best-known phonologists around has apparently bemoaned the failure of OT to produce “deep descriptions of languages”.
This is a strange way to state something that, at its core, is not entirely false. What’s not false is this: there is very little OT work that aims to “deeply describe” a language. But a theory is not responsible for producing deep descriptions — theorists are! So what this well-known phonologist is really saying is that nobody doing OT is bothering to produce a deep description of a language.
The implication of the way Mr. Verb has stated this here, though, is quite different, especially on the heels of the previously-quoted sentence. One imagines an OT proponent somewhere attempting to use the theory to produce a “deep description”, failing miserably, and — having wasted all that time on a futile task — cranking out yet another example of that god-awful “theory-internal tinkering”.
The “deep description” side of work in OT has always been its focus not on individual languages but on language typology, which was never really a strong suit of rule-based serialism. Should we bemoan the failure of rule-based serialism to produce “deep descriptions of language typology”? (Perhaps more on this in a later post; I’m still working on an idea.)
Suffice it to say that I join the well-known phonologist that Mr. Verb quotes in the desire to see more “deep description” of languages — but I don’t need to see it done in any particular theory in order to be satisfied.
Part 3, or: if it walks like a duck…
It appears that I need to re-iterate a very crucial point that I made, apparently too obliquely, in the third of my four challenges to Mr. Verb & friends. (I have to add ‘& friends’ because, as Mr. Verb was careful to point out several times, he is mostly passing on things he’s heard others say.) This crucial point was a main theme of my recent Phonology article (pre-print here for the institutionally disadvantaged) — which Mr. Verb was gracious enough to read in the process of responding to the challenges — and so I quote here a relevant passage from that article (p. 219 of the published version, p. 3 of the pre-print). (Please note my use of ‘SPE‘ as a shorthand for “rule-based serialism, [in which] generalisations are expressed by ordered rewrite rules, and a given rule R‘s generalisation is obscured in some environment E when R is ordered before another rule that rewrites any of the information crucially referenced by R in environment E.” (p. 217 / p. 1).)
Recent work has focused almost exclusively on how opacity is a formal problem for OT, the prevailing view being that opaque generalisations […] are an established natural class of phenomena with a complete and unified analysis only within SPE.
In the article I even-handedly discuss examples of opaque generalizations (i.e., ones that conform to Kiparsky’s accepted definitions of opacity) that are relatively difficult (if not impossible) for SPE to account for, in addition to ones that are relatively difficult (if not impossible) for ‘classic’ OT to account for — with a special emphasis on the former, of course, not only because I work in OT but also because of my desire to expose the ‘prevailing view’ identified in the quotation above as misguided at best. Rule-based serialism does not in fact have a corner on the opaque generalization market (and certainly not on the somewhat broader ‘obscured generalization’ market, as my good friend Ed was level-headed enough to point out at the beginning of this comment.
Once completing the article, though, I had that nagging sense that there was something important that I had left out. And there was: phonologically-derived environment effects (as opposed to morphologically-derived ones, a very important distinction that escaped both Mr. Verb‘s and Cassaday‘s attentions). The classic example is the one I cited in my third challenge, the interaction between Assibilation (t → s / __ i) and Raising (e → i / __#) in Finnish. The facts to account for are:
- application of Assibilation when Raising has applied (e.g. /…te/ → |…ti| → […si]) and
- blocking of Assibilation when Raising is not applicable (e.g. /…ti/ → […ti], *[…si]).
Rule ordering alone is not enough:
- if Assibilation precedes Raising, Assibilation incorrectly fails to apply to /…te/ (→ *[…ti] by Raising) and incorrectly applies to /…ti/ (→ […si]);
- if Raising applies before Assibilation, Assibilation applies correctly to /…te/ (→ |…ti| → […si]) but incorrectly to /…ti/ (→ *[…si]).
The trick is to have the effect of the Raising-before-Assibilation in the case of /…te/ but to block application of Assibilation in the case of /…ti/, in which the structural description of the rule is not ‘derived’.
Rule ordering is not enough, but this interaction is an opaque one in one of the two senses originally defined by Kiparsky (and never contradicted by any subsequent work): the fact that […ti] surfaces in some cases means that Assibilation is an opaque (underapplying, non-surface-true) rule. That’s what the definition of opacity — anyone’s definition — tells you. And yet it cannot be handled by rule ordering alone. The only conclusion any reasonable person can reach (other than denying the validity of the relevant facts, of course) is that there is no unified account of opacity, not even in the relatively opacity-friendly world of rule-based serialsm.
As a result, extraneous devices like the Derived Environment Condition (eventually the Strict Cycle Condition, eventually Kiparsky’s underspecification-dependent 1993 account of non-derived environment blocking) have been found to be necessary add-ons to the theory of rule ordering. (I’ll refrain from continuing the so-far-unproductive debate about whether these are mere “refinements” to SPE or “gymnastics” of the kind that similar extraneous devices in OT are accused of being; in my view, it is incumbent on the accusers to define this distinction beyond the gut-level intuitions of clear proponents of one theory over the other.) I should mention that Ania Lubowicz is well-known for her lucid discussion of the “different and arguably superior” predictions made by her local conjunction OT analysis of phonologically- and morphologically-derived environment effects, as compared to predictions made by the Strict Cycle Condition, in her 2002 Lingua paper (pre-print here).
One thing we might do at this point is tally up the kinds of opacity that can be handled with rule ordering alone and those that can be handled with constraint ranking alone (all else being equal to the extent possible), then tally up the kinds of opacity that can’t be handled with rule ordering alone and those that can’t be handled by constraint ranking alone (again, all else being relatively equal), and choose to pursue the theory with the greater empirical coverage. But this brutish approach will quickly lead us down the same kind of “how/what do you count?” problems faced by the ‘Eskimo words for snow‘ canard (coincidentally the topic of my undergrad class lecture this past Wednesday, and also mentioned earlier today by Mr. Verb himself):
- Couldn’t it be that some types of opacity are more ’empirically weighty’ (typologically more common, more likely to develop spontaneously in acquisition, however you may want to assess it) than others?
- Why count just opaque generalizations? Why not ‘obscured generalizations’ more generally? Why not ‘phonologically significant generalizations’ (including conspiracies, which Mr. Verb has acknowledged are significant) even more generally?
Personally, I prefer to see what’s happening in a lot of current work: re-examinations of well-worn examples and discoveries of new ones in light of what appears to be (as well as what appears not to be) predicted by a diversity of existing theories/frameworks. (Note, for example, the recent announcement of IULC’s Phonological Opacity Effects in Optimality Theory.) Facts are illuminated by theories, even by ‘wrong’ theories — as must be the case, because, as Cassaday Rassmussen put it:
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.
Not that I want to participate in a ‘group hug’ with Cassaday; I still think s/he’s a sniveling pseudonymous stooge. I have a list of people who I’m guessing might be Cassaday, and I plan to perfunctorily reject all of their conference and journal submissions, paste their visages on my dartboard, and snub them at LSA parties. (Just kidding — I don’t own a dartboard.)
Part 4, or: territorial pissings
I’ll end here on a hunch that I’ve had for quite a while now about what’s going on here, and I think it nicely ties much of the above together. A lot of what working phonologists understand about phonology is filtered through the lens of theory — I think we can all agree on that. Perhaps more controversially, I think that the more temporally distant but familiar the origins of the theory-lens, the more likely we are to think it’s not much of a lens at all; similarly (though not necessarily conversely), the more unfamiliar or temporally proximal the theory, the more likely we are to think it’s nothing but a lens clouding our otherwise better judgment.
And so it is with SPE and OT. The origins of SPE are more temporally distant but somewhat familiar to most of us, and the facts as filtered by that theory are thus often taken to be bare facts with no lens attached. OT is newer and somewhat unfamiliar to some of us, and the facts as filtered by that theory are often taken to be simply clouded. (Of course there are those to whom SPE may be completely unfamiliar, those to whom OT may seem practically as old as SPE, and what have you — I’m speaking in generalities about certain groups of people here.)
The practical consequences of this are enormous. Opacity was originally defined in SPE terms, dammit, and so it must be that all cases of opacity are analyzable using the most basic building blocks of SPE. Forget phonologically-derived environment effects: OT can’t even handle a simple example like Canadian Raising! (Incidentally: there is only one example cited in the literature known to me that appears to require something other than stratal ordering to account for Canadian Raising; see the examples of the author’s own speech in (7) on p. 9 of this paper — the crucial case is (7b).)