Kawahara (2015) @cite{kawahara-2015} #
The phonology of Japanese accent. In Haruo Kubozono (ed.), The handbook of Japanese phonetics and phonology, 445–492. Berlin: De Gruyter Mouton.
This survey chapter provides a comprehensive overview of Tokyo Japanese pitch accent, covering default accent assignment, compound accent, verb/adjective paradigms, affix accent typology, and interactions with epenthesis, rendaku, and vowel devoicing.
Formalized contributions #
- AAR vs LSR (§2, Table 1): the two accent assignment rules agree on 6 of 8 trisyllabic weight conditions and diverge on HLH and LLH.
- Accent-to-tone derivation (§1.4): surface tones are fully determined by accent location via accentual HL, initial rise, and tonal spreading.
- Eight affix accent types (§6): the traditional dominant/recessive distinction is refined into an 8-way typology.
- Compound accent (§4): short N2 (≤2μ) triggers pre-accenting or retention; long N2 (≥3μ) triggers N2-initial accent or retention.
Table 1 completeness: exactly 2 of 8 conditions produce different predictions between AAR and LSR. The mismatches are HLH and LLH.
In both mismatch cases, the AAR predicts accent on σ₂ (the syllable containing the antepenultimate mora) while the LSR predicts accent on σ₁ (the antepenultimate syllable, because the penult is light).
@cite{kubozono-2008} argues that LSR-conforming pronunciations appear as variants even in the mismatch cases, suggesting the LSR may be the better characterization.
For n-mora words, there are n+1 distinct accent patterns: accent can fall on any of the n moras, or the word can be unaccented.
However, word-internally, final accent and unaccented produce the same tonal contour (both LHH for 3 morae), because the post-accent L falls off the word edge. They are distinguished only in phrasal context (e.g., followed by a particle).
We verify that the first 3 patterns (unaccented, initial, medial) are pairwise distinct, and that final-accented = unaccented word-internally.
Culminativity: Japanese allows at most one accent per prosodic word.
We verify that JProsodicEntry structurally enforces this — accent
is a single Option Nat, so there can be at most one accented mora
by construction.
The 8 affix types collapse to exactly two prosodic dominance classes: recessive (3 types) and dominant (5 types).
Short N2 pre-accenting moves accent off the final syllable of the compound, consistent with NonFinality(σ).
Long N2 with unaccented N2 assigns accent to N2-initial syllable, which is never word-final (since N2 is ≥ 3μ).
Japanese moraic structure: WBP is active (coda consonants are moraic). This means coda nasals (/N/), geminate first halves (/Q/), and long vowels all contribute to syllable weight. The weight sensitivity of accent assignment (AAR/LSR) operates on these moraic weights.
@cite{kawahara-2015} §2.3: syllables with coda nasal, first part of geminate, long vowel, or diphthong are bimoraic (heavy). Open syllables with short vowels are monomoraic (light).