Two (informal) constraints on Okinawan phonology

In this post, I’d like to informally detail two phonotactic constrains in Okinawan: the bimoraic constraint and the word-initial non-syllabic onset constraint.

Bilabial Alveolar Alveolo-Palatal Palatal Velar Glottal
Tenuis Stop (p) b t d k g  ʔ
Palatalized Stop (bʲ)  ʔʲ
Labialized Stop kʷ gʷ ʔʷ
Flap ɾ
Affricate (ts dz) tɕ dʑ
Fricative s (z) ɕ h
Nasal m n
Approximant j w

Table 1: Okinawan consonants.

Table 1 details Okinawan consonants. Consonants in parentheses are marginal, occurring only on onomatopoeia (as is the case for /p/ and /bʲ/) or in the speech of people historically from the gentry (or the Shuri sociolect) (as is the case for /ts/, /dz/, and /z/, which have merged with /tɕ/, /dʑ/, /dʑ/ in the speech of people historically who were commoners (or the Naha sociolect).

The first constraint I’m going to discuss is the bimoraic constraint.

(1) The Bimoraic Constraint
If a phonological word is one syllable, it must be two mora in length.

So what is a mora in Okinawan? Vowels count as one mora each, syllabic nasals count as one mora, and geminate consonants count as one mora. For one syllable words, this means the following shapes are possible:

(2) Monosyllabic words in Okinawan
CVV – [jaa] ‘house’, [ʔʲaa] ‘you (nonpolite)’, [ʔaa] ‘bubble’
CVN – [ʔin̩] ‘dog’, [sun̩] ‘to do’, [mun̩] ‘thing’
QCV – [kkʷa] ‘child’, [ttɕu] ‘person’ (these may be the only two words of this shape)

The potential shapes *V , *VV, *VVC and *VCC do not occur. *V, and *VVC do not occur due to the fact that they inherently violate the bimoraic constraint. *VCC does not occur as it has no potential historical antecedents. Finally, the CCV shape is limited to QCV (where QC is a geminate version of C). As we have seen, the bimoraic constraint cannot explain all of these non-occurring forms, specifically *VV. For that, we need at least one more constraint that holds for all phonological words in Okinawan: the word-initial onset constraint.

(4) The Word-Initial Onset Constraint
A phonological word must begin with an onset.

Note that there are potential counter-examples of this constraint. For instance, [ʔeema] ‘interval’ contrasts with [eema] ‘Yaeyama’, with the latter apparently not having an onset, though it did historically (< *yaema < PR *ya-pe+yama |eight-layer.CLF+mountain|; I have no explanation for the missing syllable). Speakers, however, typically add a consonant (in this case the glide /j/) to repair the supposed violation of this constraint. Recordings of this word in particular from the Shuri-Naha Hōgen Onsei Dētabēsu pretty clearly reflects this, as does an acoustic look at the word:


Figure 1 – Waveform of [jeema]
Figure 2 – Spectrogram of [jeema]

It seems in most cases that speakers repair violations of this constraint, but there may be a few onomatopoeia and para-linguistic sound patterns that violate this constraint.

Both of these constraints and how I define a mora in Okinawan have significant influence on how I analyze Okinawan phonology. For instance, counting geminate consonants as a mora means that I must propose palatalized and labialized glottal stops—typologically quite rare—because otherwise a word like *[ʔjaa] (which I analyze as [ʔʲaa] ‘you (nonpolite)’) would violate the bimoraic constraint (as it would have three morae).

We also have two contrasting sets of words which have a syllabic nasal as the nucleus for their first syllable. For instance, we must analyze what is written as <ンニ> or <Nni> (with the capital <N> stands for a syllabic nasal) ‘breast, chest’ as [nn̩ni], rather than *[n̩ni], as the latter violates the word-initial onset constraint. Note that this contrasts with a glottal-stop initial [ʔn̩ni] ‘rice (plant)’. Historically, these seem to make sense to me, as the former originally had a consonantal onset (Proto-Ryūkyūan *mune ‘breast, chest’), while the latter did not (Proto-Ryūkyūan *ine ‘rice (plant)’). This constraint seems to be a Northern Ryūkyūan innovation, as southern Ryūkyūan varieties do not face this restriction; compare Ōgami Miyako [ɑmi] ‘rain’ (a Southern Ryūkyūan variety), Yamato Amami [ʔamï] (a Northern Ryūkyūan variety), and Shuri-Naha Okinawan [ʔami] ‘rain’ (another Northern Ryūkyūan variety), all from Proto-Ryūkyūan *ame ‘rain’.

4 thoughts on “Two (informal) constraints on Okinawan phonology

  1. Nice! I especially like the discussion of the implications – the edge cases are always where the fun is. A couple of questions that occurred to me:
    – If there are no VV words because they are all repaired a la [eema], why are there words like [eema]? That is, even if the end result is that all words actually start with a consonant, why is that rule reflected in the lexicon for monosyllabic words but not polysyllabic ones?
    – What makes palatalized/labialized glottal stops preferable to exceptions to the bimoraic constraint, in terms of explanatory power?
    – Why not analyze “breast” as /Qni/, a third member of the QCV family?

  2. (Whoops, looks like WP interpreted “Nni” in angled brackets as an HTML tag and didn’t escape them for me. Should have gone just before “breast”.)

  3. I think the issue with words like /eema/ ‘Yaeyama Is.’ is the same as why we don’t reconstruct a word like [ʔaa] ‘bubble’ with a glottal stop: the rule that required onsets was a relatively late occurrence in the history of Okinawan, or at least one that occurred after the various mergers that led to things like /eema/.

    For me, what makes palatalized/labialized glottals preferable is twofold. First, if we can make a rule that we don’t have to violate, we’d prefer that. But there’s also the parallelism with the velars (or at least there was parallelism with the velars–the palatalized velars *kʲ and *gʲ merged into the modern affricates /tɕ/ and /dʑ/, respectively).

    I hadn’t thought of ‘breast’ like that. It could be, but I think we’re dealing with two syllables rather than one, though I don’t at this time have any formal tests to distinguish mora from syllables in Okinawan. Something to do some fieldwork on, I guess.

  4. Thanks for this reply! The parallelism with the velars is interesting.

    Let me get at the /eema/ issue from another angle. Basically I am trying to understand why you transcribe the word as [eema] rather than [jeema], even though (as you explain) speakers usually add a glide so that it does not violate your second constraint. If this is the case, why not go big and argue that the word actually is [jeema], at least at this point in history?

    Is it just that the glide-adding process is not quite consistent enough to justify an argument that goes that far? (In which case perhaps we are witnessing a phonotactic change in progress? — that would be cool.)

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.