Here’s just some thoughts I’ve been mulling over on representing segment length. I’d love to get feedback. I’m a bit rusty since I haven’t really thought hard about phonology for a couple of years.
It seems that representational theories of segment length (two-root theory or moraic theory) are pretty good at addressing some basic properties of long segments. For example, in some cases length is preserved (compensatory lengthening) when the segment degeminates. That fact makes sense if the length is represented as double linking to a timing slot and degemination is simply unlinking to the extra slot, leaving it free to relink somewhere more hospitable.
But, an interesting issue with both two-root theory and moraic theory is that length really seems to be a binary distinction. A segment is either long or short and that’s that. But the representational theories don’t really capture this. And they don’t provide any reasons why there isn’t three-way or four-way contrasts in segment length (or more). This is pretty obvious in the two-root theory since there really is no reason not to have a single segment linked to more than two timing units. I think it’s also a problem in the moraic theory since vowels at least allow linking to multiple morae. And there is no explicit reason why consonants can’t be doubly linked. At best there seems to be a general rule against having more than one long segment in a syllable (except maybe word-finally).
Maybe this isn’t such a big problem for moraic theory. In fact, if there is a very general syllable markedness constraint (or set of constraints) that cap things off at 3 moras per syllable, then you’ve subsumed this to general principles and not just a stipulation about [±long].
But then, how do you enforce these restrictions in OT? Inviolable output constraints? Restrictions on GEN? Isn’t that just trading stipulations up? For some reason I can’t help but feel a little let down by these representational theories.
There is a three-way length contrast in Estonian (short, long, and overlong).
North Saami too I believe (at least consonants, I’m not sure about the vowels).
Although merely allophonic in contrast even English has at least three vowel lengths. e.g. beat, bead, bee. There must be reason to determine if one of these lengths is the UR.
The question of whether universally absent structures should be ruled out in Eval (as you suggest, Ed, with the question about inviolable output constraints) or in Gen was one of the topics discussed at last year’s “Freedom of Analysis?” workshop, and it would be interesting to discuss it further here. From my point of view, one of the themes in the OT literature is the attempt to analyze such situations with constraints, although not with universally inviolable ones, but rather showing that the universally absent structure is harmonically bounded. In this way, the work is done in Con and Eval, but no stipulation of inviolability is necessary. And, of course, it’s a pleasant result to actually explain the absence of some structure because of its harmonic boundedness. But, this isn’t always going to be possible — perhaps Ed’s case regarding limits on the moraic size of syllables is such an example — and I personally am an advocate of opening the door to more careful study of Gen as a domain in which structural generalizations might be modeled.
I know it’s really boring to promote one’s own work and I apologize in advance for this, but I’ve just written something on exactly the trade-off bt Gen and Con quite along the lines of how Ed describes it, so I’ll let myself mention that it’s available here.
Curt
Curt,
Thanks for the pointer to your paper–I don’t find it boring at all. My dissertation, several years back, was about ruling out universally absent structures in Eval, but it requires some redefining of faithfulness constraints. To me that seems more satisfactory for the reasons you mention. But I’m not sure you can get away with a completely free Gen. And I think the discussion of Gen has to include a discussion of the representations that Gen acts on, especially since a lot of the motivation for phonological representations pre-OT had a kind of OT motivation to it–explaining exceptions to output constraints.
Thanks to Curt for further discussion on Ed’s last questions, and links to current papers are always welcome. As for Ed’s original post, I would like to comment on this part: “A segment is either long or short and that’s that. But the representational theories don’t really capture this. And they don’t provide any reasons why there isn’t three-way or four-way contrasts in segment length (or more).”
Moraic theory can represent distinctions in length other than just long or short through means of mora-sharing, as demonstrated by Broselow, Chen, and Huffman (1997). For example, in their article they show that in Malayalam a heavy syllable can be of the type VV or VVC. The long vowel in the closed syllable is phonetically shorter than in the open syllable, a fact that can be explained if both syllables are bimoraic: the open syllable has a vowel linked to two moras and the closed syllable has a vowel linked to two moras and sharing one mora with the consonant. If the input consists of a vowel attached to two moras (and a consonant attached to none, in the case of the closed syllable), a constraint against consonants heading a mora must dominate a constraint against a mora being attached to more than one segment to predict the described structures.
I have found the phenomenon of mora-sharing to be effective at explaining subtle but consistent differences in length in the nuclei of monosyllables in English ( Frazier 2005).
Both of these papers use mora-sharing to explain differences in length that are predictable and not phonemic. However, the existence of mora-sharing in the structures of outputs leads to the question of whether or not such structures (are predicted to) exist as inputs. Is it the case that, underlyingly, a segment is either long or short (i.e. attached to zero, one, or two moras)? Is there any evidence for phonemic differences in length that require mora-sharing structures?
references:
Broselow, Ellen, Su-I Chen, and Marie Huffman. 1997. “Syllable weight: convergence of phonology and phonetics.” Phonology 14: 47-82.
Frazier, Melissa. 2006. “Output-Output Faithfulness to Moraic Structure: Evidence from American English.” To appear in NELS 36.
Melissa,
Thanks for the input. My contention is that there are no phonemic differences
in length beyond “short” and “long”. And since mora sharing is a possible
representation, it makes me uneasy about moraic theory. Now it seems that
North Saami needs to be looked at as a counterexample to my contention. As
far as I know, the Estonian case is not phonemic.
Has anyone really tried to explicate what moraic OT looks like?
Hei,
There’s a recent UW dissertation (2003) on this topic, by Emily Curtis (which I don’t think I see on ROA).
For the Saami facts, one available source is Patrik Bye’s dissertation, which is on ROA. His MPhil thesis also explicitly considers various mora sharing issues.
Curt