Is accent position predictable for -i verbs in Osaka/Kansai?

Question

Background

While I'm not especially confident about the reliability of my sources (Wikipedia (0) (1) (2) and some guys' personal websites), and it's possible that I transcribed some of these wrong, the sources agree on the following accent rules for -i verbs:

• The few bi-moraic ones are all unaccented with low initial tone (nà-i 'not', yò-i ~ èe 'good', kò-i 'dark, thick, strong', sù-i 'sour')

• All tri-moraic -i verbs I've seen specified give the accent on the 1st (penultimate) syllable; for example

  • âtu-i [á.tsʉ̀.ì] 'hot, warm'
  • îta-i [í.tà.ì] 'hurt'
  • kûro-i [kú.ɽò.ì] 'black'
  • kûsa-i [kú.sà.ì] 'smell like'
  • sîro-i [ɕí.ɽò.ì] 'white'
  • etc.

• For longer words, the accent seems to usually fall on the penultimate syllable. Initial pitch is usually high:

  • tánôsi-i [tá.nó.ɕì.ì] 'fun'
  • ómôro-i 'interesting, amusing'
  • káwâi-i 'adorable'
  • síkâku-i 'square-shaped'
  • yáyakôsi-i 'complicated'

  and sometimes low:

  • syòomôna-i [ɕò.ò.mó.nà.ì] 'boring'
  • àhokûsa-i 'idiotic'

  But there seem to be a few exceptions:

  • òisî-i [ò.ì.ɕí.ì] 'tasty'
  • sìɴdô-i [ɕì.ɴ̀.dó.ì] 'tired'

Question

Now this looks like a straightforward rule: accent shifts to the right when it would otherwise fall on the second mora of a syllable. If valid, this rule entails that accent position is predictable for -i verbs. However, I'm suspicious of that because it has been emphasized (e.g. here) that unlike in Standard Japanese, the Kansai dialect's accent system is purely moraic and does not constrain accent to the first mora in a syllable.

Also, there may be exceptions to the proposed rule, possibly mímiʔ̂ti-i 'stingy' or bábâʔti-i 'dirty, messy' or déʔ̂kai 'huge' or gôʔtu-i 'huge'. (But it's hard to tell the underlying accent pattern, since /ʔ̩/ is inherently voiceless, so these aren't clear exceptions.)

So: is this rule valid? Or are òisî-i / sìɴdô-i / bábâʔti-i / gôʔtu-i irregularly stressed?

---

Notation: Here I'm using the circumflex ⟨ ̂⟩ to mark accented syllable, and acute ⟨ ́⟩ and grave ⟨ ̀⟩ to indicate initial high and low pitch respectively (following the IPA).

Answer 1

I believe so. I can't find an explicit affirmation (I provided sources which I've read before, but I could have forgotten or missed such a statement), but for present tense adjectives in the Kyoto-Osaka dialect, it seems the accent falls on the antepenultimate mora (third to last) for trimoraic words or longer, otherwise it falls on the penultimate mora for bimoraic adjectives. I'm assuming there are no monomoraic adjectives. Although that's a reasonable proposition considering that non-accentless いadjectives in the Tokyo dialect are completely predictable with the accent kernel on the penultimate mora (e.g. あつい,うれしい). I'll denote an accent kernel carrying mora with boldface.

The predictability of the location of the accent kernel (アクセント核) does depend on the dialect in question so let me just make a note of that to be clear. The most immediate dialect division is between Eastern and Western Japanese isoglosses. Within Western Japanese the most prominent isogloss is the Kyoto-Osaka dialect. Then finer isoglosses within the Kyoto-Osaka dialect can be made, which are the Kyoto dialect and the Osaka dialect. The term "Kansaiben" is mostly an informal term and is generally eschewed in technical materials. The usual convention for naming dialects is to name them after the metropolis that is the dialect's geopolitical progenitor.

In some dialects the accent kernel's location is totally predictable. In in the Miyakonojo dialect it's always on the final mora of every word in the utterance and in the Ozu dialect it's always on the initial mora of every word. Then in the Kagoshima dialect not all words have an accent, but if there is one present then its location is predictable; it is always on the penultimate mora. At the other end of the spectrum, some dialects do not have accents; they are lexically isotonic (but there is of course still intonation). In case of the Kyoto-Osaka dialect, which is probably the dialect you have in mind, the accent location is generally unpredictable, but for the subset of adjectives, I can't find an explicit statement, but it seems predictable.

Your rule seems pretty accurate but it might be rephrased in terms of mora. I would change it to: The accent kernel associates to the antepenultimate mora if the word is trimoraic or longer, otherwise it associates to the penultimate mora (in the case of bimoraic adjectives). As, Labrune (2012) and Kubozono (2012) mention, if the antepenultimate mora is a geminate obstruent っ, then keep shifting the accent leftwards onto the first available mora (I'm pretty sure at most one leftward shit will ever be necessary due to the phonotactic constraints of Japanese). Labrune mentions that the accent kernel can fall on ん and the long part of long vowels.

It might be worth it to start afresh with some definitive raw sources:

1. PDF The Languages of Japan (Shibatani, 1990) pages 177-184 and chapter 9 (Shibatani's accent typology is not used)
2. PDF The Old Kyoto dialect and the development of the Japanaese accent (Ramsey, 1979)
3. PDF Varieties of pitch accent systems in Japanese (Kubozono, 2012) (Not too advanced, you should be able to understand it)
4. PDF Accent in The Handbook of Japanese Linguistics (Hiraguchi, 2001)
5. The Phonology of Japanese (Labrune, 2012 pages 251-258)

Kubozono (2012) works with the Kyoto dialect instead of the Kyoto-Osaka dialect, but I believe the accent systems of the two are identical. To be consistent I'm just gunna call it the Kyoto-Osaka dialect. Kubozono's typology is very detailed so I'm gunna use his system.

6 Criteria for classification by accent typology:

1. Number of total combinatory accent realizations possible for a phonemic string of length (in mora) n. This number is called N. So, what you do to find N is you take all the words of moraic length n, and take a look at where the accent kernel is among all these words. The number of distinct patterns for which accents are located is this N.
2. Obligatoriness (whether every lexeme has at least one accent kernel)
3. Culminativity (whether every lexeme has at most one accent kernel)
4. Unit of prosody (whether the mora or the syllable is the imputed unit of prosody) This prescribes how phonemic strings are indexed in order to position the accent kernel. If you index with the mora, it is said to be mora counted. Likewise with the syllable. Mora based means that given an index, the tone associates with respect to the mora. Syllable based means association occurs with respect to the syllable. Even though a dialect may be mora counted it can still be syllable based. For example, the rule for loanword accentualization in the Tokyo dialect: Accent falls on the syllable (syllable based) containing the antepenultimate mora (mora counted).
5. Domain of accent application (rules for computing the accent of X+P given the accents of X and P where P is a concatenated array of particles) Three possible domains: word, phrase, and syllable but this won't concern us here because were talking about just isolated present tense adjectives.
6. Prosodic compound rule (rules for computing the accent of X+Y given the accents of X and Y) This won't concern us here for plain present tense adjectives.

As far as the accent in the Kyoto-Osaka dialect. Remember that an accent kernel is an HL transition (LH transitions do not signify an accent):

1. If a lexeme has moraic length n, there are 2n+1 possible accent realizations (in general, but not for いadjectives). Here's how the 2n+1 number is counted. First fix n. Then, the +1 is for when the word doesn't have an accent. Then if there is an accent the accent can fall on any of the n mora, so +n. However even if there is an accent on any of the n mora, there is always a possible contrast between an initial H tone and an initial L tone, hence the +n is doubled to +2n. Here is an example. Consider just three trimoraic words, n=3.
   
   
   1. さくらが atonic
   2. かぶとが LHL-L
   3. にわしが HHL-L
   This isn't all the possible permutations of trimoraic words, but it illustrates only the point that needs to be made. Take the last two tonal patterns LHL and HHL. Notice the accent kernel is in the same exact location on those last two words, but yet the tonal melody is not identical. This is because after the accent kernel is identified, there can be two possible tones on the initial mora, either H or L. This is where the 2 comes in for 2n+1. When H is the first tone of the word, the tonal type is said to be an HL type. When L is the first tone of the word the tonal type is called the LHL type. For the Kyoto-Osaka dialect, all plain present tense いadjectives are of the HL type, i.e. all of them start with H. However there is one conspicuous adjective, ええ, for which tonal type is LH eg L-initial.
2. In general not every lexeme has an accent, but luckily in the case of present tense adjectives there is always an accent present.
3. Every present tense adjective has no more than one accent kernel. Then together with the observation above we can say: every present tense いadjective has one and only one accent kernel.
4. It is mora based and mora counted. So positions in the phoneme string are indexed by mora, and the accent kernel associates to that mora. This I know for sure, so you must switch your metric to the mora.
5. Not relevant for isolated finite adjectives
6. Not relevant for isolated finite adjectives

So, いadjectives always have an accent (and never more than just one), so our predicting is not in vain. Including your examples (assuming they are reliable) here's the data so far:

1. HL: すい
2. HLL: たかい, あかい, あつい, いたい, くろい, くろい, くさい, しかろい
3. HHLL: うれしい, たのしい, かなしい, おもろい, しかくい
4. HHHLL: おもしろい, ややこしい, しょうもない, あほくさい
5. Exception: ええ LH

The entire tonal melody of an いadjective is predictable then. The accent kernel associates to the antepenultimate mora (penultimate if the いadjective is only bimoraic) with all preceding mora having an H tone and all subsequent mora having an L tone, except for the possibility of an initial contrast. Moving from left to right, if there is an H to L transition (an accent kernel) there will not be an L to H transition (the tone melody will not jump back up).

Also, don't forget that superimposed atop these pitch fluctuations is intonation.