Sunday, June 22, 2008

the NoCoda constraint

As I mentioned last week, there is a cross-linguistic tendency to assign as many consonants as possible to an onset when dealing with a consonant cluster (review: syllables have three parts - the onset, which consists of the initial consonant or consonant cluster, the nucleus, which consists of the vowel or diphthong in the syllable, and the coda, which is the final consonant or consonant cluster; NB: this is a simplified description). The reason for this is two complementary cross-linguistic trends: syllables like onsets, and they don't like codas. These trends are formalized in Optimality Theory as the constraints Onset (syllables should have an onset) and NoCoda (syllables should be open).

If these were the only two relevant constraints, there wouldn't be any syllables with codas, because all of the consonants would get piled on the onset of the following syllable. However, other preferences override these constraints in certain situations. Take the word "constraint," for example. The word has the massive "nstr" cluster in the middle. If all we cared about was getting rid of codas, we'd shove all those consonants onto the second syllable, giving a syllabification of co.nstraint. Go ahead, try pronouncing that. It's not very fun. The reason is that there is another constraint at work here, called sonority sequencing. This refers to the general tendency of onsets to increase in sonority and for codas to decrease in sonority (I'll leave an explanation of sonority for another post, but generally speaking, in order from least to most sonorous, sounds are classified this way: stops < fricatives < resonants < vowels, with voiced sounds being more sonorous than voiceless sounds within each category). In OT this constraint is Son-Seq (sounds increase in sonority moving from syllable margin to syllable nucleus). In the "nstr" sequence, the n is higher in sonority than the s, which means putting them together in an onset violates sonority sequencing. Since in English Son-Seq is ranked higher than NoCoda, we would rather syllabify the word as con.straint than co.nstraint.

The reason I'm talking about this at all is from having set our DVR to record the show Good Eats on the food network. The host of the show, Alton Brown (who is also the announcer/narrator for Iron Chef America) has a formidable NoCoda ranking, and almost always assigns many more consonants to onsets than most of us would ever want to. An example of this is his syllabification of the word "fifteen." I say fɪf.tʰin, with aspiration on the t because it begins a stressed syllable. However, Mr. Brown says fɪ.ftin, a syllabification distinguishable from my own pronunciation by the last of aspiration on the t, which signifies that it is not syllable-initial (compare pʰɪt with spɪt). This unusual syllabification violates sonority sequencing (remember that stops are less sonorous than fricatives), but he prefers it because it results in only one NoCoda violation instead of two.

Another example of an extremely high-ranking NoCoda constraint is the syllabification of a word-final coda with the following word, even in careful speech. I wish I could remember the token Alton produced to make me think of this, but any random example will do. Usually in rapid speech, with a phrase like "farm aid," we say far.maid, because of the constraint *Complex (syllable margins should be simple). Our Align-Morph-R constraints are being violated here, since the first morpheme is being split between two syllables, but in rapid speech we generally prefer violating that to violate *Complex, which is an articulatory constraint as opposed to a semantic one. However, I noticed that even in careful speech, Alton Brown continually violates Align-Morph-R, because he would always rather fulfill phonetic and articulatory constraints over more abstract semantic ones.


N said...

I'm pretty familiar with the workings of OT, but for the constraint "*Complex (syllable margins should be simple)," what does 'simple' mean? Maybe needing a certain amount of difference between adjacent sounds?

linguistlessons said...

By simple I (and Kager (2001), whose wording I'm using) mean only one consonant, e.g., "cat" has a simple onset and coda, whereas "strides" has a complex onset and coda.

Anonymous said...

I have a question,but not about this article,
What does s-obstruent cluster refer to?
1./s/ phoneme?
please, could u help me understanding.

MA student of applied linguistics.

Thnak you

linguistlessons said...

I would assume "s-obstruent cluster" refers to a sequence of an /s/ followed by an obstruent. An obstruent is a fricative or a stop, i.e., a non-resonant.

Joe.L said...

Why is it that a s+plosive cluster in English is never considered a violation of the Sonority Sequencing Principle?

Coz according to the SSP:

"stops < fricatives < resonants < vowels"
an s+plosive sequence should be considered a violation of SSP.

Thus, why is the word constraint syllabify as "cons.traint" instead of "con.straint"?

Great thanks!

Ryan Denzer-King said...

Joe.L: The usual analysis is that /s/ in these cases is extrasyllabic, i.e., it is not attached directly to the syllable, but instead to a higher prosodic unit like the foot or prosodic word. Essentially this means that look at the syllable itself /s/ does not exist in the syllable onset, and thus doesn't technically violate sonority sequencing at that level. To me this isn't a very satisfactory explanatory answer, however theoretically useful it may be, and I'm very interested in special properties of /s/, since they seem to occur almost universally. Many languages other than English allow exactly those types of sonority "violations" only with /s/.

Joe.L said...

Thanks for your kind reply!
Actually I'm an English undergrad student in Hong Kong who is very interested in OT and I find your blog very informative in terms of elementary linguistic knowledge.

Actually I am trying to use OT to explain the syllabification of English words.
So far I've come up with this ranking:

however, I have discovered that under this ranking, words like Islam /ɪzləm/ is syllabified as [ɪ.zlem] instead of [ɪz.lem], and Atlanta /ətlæntə/ is syllabified as [ə.tlæn.tə] instead of [ət.læn.tə].

I reckon that clusters which abide the SSP and yet forbidden in English such as /tl/ and /zl/ above should be banned.

Do you think that this is caused by errors in my ranking or have I missed a crucial constraint in my analysis?

Great thanks!


Ryan Denzer-King said...

Without taking a deeper looking at the constraint set, I wonder if you really need MOP and NoCoda; these are essentially doing the same work, or am I missing something? Note that /zl/ and /tl/ are banned for different reasons. I'm not sure how you would ban /zl/. If you decompose the sonority scale into voiced versus voiceless at each category you could invoke minimal sonority distance, where /tr/, /dr/, and /sl/ are okay, but not /zl/. /tl/ is bad almost universally; languages generally disallow this sequence entirely or turn it into a lateral affricate. This is likely a perceptual reason related to the difficulty in perceiving alveolar stop bursts that overlap lateral gestures (note that no language contrasts affricates /tl/ and /kl/). So that would almost certainly deserve its own constraint.

vp said...

I'm not familiar with the sonority hierarchy, but, based on your description of it, it seems that "sonority tracks two independent properties of phones:

* degree of voicing (i.e. for what percentage of the time the vocal cords are activated)

* constriction of the vocal tract (less constriction > greater constriction).

In the case of a /s/ + plosive sequence in English (e.g. /st/), it is worth noticing that these two properties are in competition with each other. /s/ constricts the vocal tract less than /t/, so on that count /s/ is more "sonorous" than /t/.

However, in English (in every accent that I am aware of), /t/ following /s/ is unaspirated. Because of this lack of aspiration, the Voice Onset Time for /t/ will typically be less than for /s/. Therefore, in this respect, /t/ is more sonorous than /s/.

This may explain /s/ + plosive sequences in English. (However, there are other languages that allow initial /s/ + aspirated stop -- e.g. Sanskrit "sthān", still found in the name of the Indian state of Rajasthan).