[ Update, Sept. 21: I’ve decided to write the substance of this post up as a squib, which I’ve just posted to ROA. Comments welcome! ]
I just finished what I hope are final revisions for an article that has been accepted for publication in Phonology (tentatively for vol. 22, issue 3). One of my reasons for this post is to distribute the paper for comments, suggestions, etc. The other is to comment further on what I think is a significant result of the paper.
The result is sufficiently clarified in the paper itself, but I think there’s something to be gained from some perspective on how I got there and what I think it means beyond the immediate context of the paper. As we all (phonologists, linguists, academics …) know, it’s typically a bad idea for your paper to simply march the reader through your thought process as you arrived at your ultimate solution to a problem, and so my paper does not do any of that (though there were vestiges of that sort of thing even in recent drafts). But, for those who are interested, there is always phonoloblog.
The seed for my paper was planted early in the Fall of 2000. I had just arrived at UCSD and was teaching the first-quarter graduate intro to phonological theory course. For reasons that aren’t particularly important here, Sharon Rose and I had agreed that we would teach derivational phonology in this course and OT in the next course in the sequence, an arrangement we have basically stuck to since. I’m no stranger to derivational phonology, but had yet to teach it beyond the undergraduate level. I decided to start the quarter with a discussion of the distribution of the English plural and past tense suffixes, having had some success with it before in an undergraduate class. (Here are the two handouts I used; credit for the structure of the first is given in a footnote, and I believe the second is derived from a handout I got from John McCarthy.)
By the end of the first class period we got to the following two familiar-looking rules, which are ordered as shown, assuming /z/ and /d/ as the underlying forms of the suffixes and ignoring the precise quality of the epenthetic vowel.
- Epenthesis: ø → V / C1 __ C2 #,
where C1 and C2 differ at most in their value of [±voice].
- Devoicing: [-son] → [-voice] / [-voice] __ #
It was of course not the first time I had looked at these rules, but it suddenly struck me that they both mention the feature [±voice]. Some of my students were struck by this coincidence, too, and we had a long discussion about it. After class, some of the students followed me back to my office to discuss it some more and, for some reason, I ended up showing them how the analysis might be translated into OT. I drew something like the following tableau up on my whiteboard:
|Input: /…t+d/||Don’t be identical, ignoring [±voice]||Don’t disagree in [±voice]||Don’t epenthesize||Don’t change [±voice]|
|1. […td]||* !||* !|
|2. […tt]||* !||*|
|3. → […tVd]||*|
It was suddenly obvious: that first constraint doesn’t need to ignore [±voice]. The crucial candidate that it rules out, #1, is already ruled out with the independently-necessary second constraint; so long as these two constraints both crucially dominate the third (and the third dominates the fourth, to explain why there’s assimilation in other cases of [±voice] disagreement), the right result is obtained:
|Input: /…t+d/||Don’t be identical, period||Don’t disagree in [±voice]||Don’t epenthesize||Don’t change [±voice]|
|1. […td]||* !|
|2. […tt]||* !||*|
|3. → […tVd]||*|
(Further details on the formulation and ranking of these constraints can be found in section 3 of my paper. Extension of the analysis to the alternation of the plural (and possessive, third person singular present tense, etc.) suffix is given in my WECOL 2004 paper. There are some interesting articulatory predictions made by that extension, a preliminary investigation of which I discussed here on phonoloblog last year; the final results of a complete palatographic study were presented by my co-author Cynthia Kilpatrick at the First ICLCE in Edinburgh about three months ago.)
At the very least, this new analysis captures a significant generalization that the derivational analysis further above (and the first-pass translation of that analysis into OT) does not: [±voice] is ignored by epenthesis because of the particular way in which all of these constraints interact, not because a feature ignored by the first constraint just so happens to be the same as the feature that also assimilates in this context. As significant as I thought this was at the time, it somehow got filed away for almost 3 years. Then I started thinking about it again, and realized how complicated — practically impossible, in fact — it is to get this result in derivational terms.