Chapter 159Hokkaido Dialect Classification and Dialectometry

159.5.1 Ono 2020: spectral clustering

Ono (2020) reanalyses Asai's cognacy matrix (110 items over 21 lects) with Spectral Clustering (MinmaxCut / Mcut algorithm). The methodological argument is that Asai's matrix is a commutative monoid: only addition and comparison of cognacy percentages are linguistically meaningful; converting them to "dissimilarity" by subtraction destroys information. Ono further shows that both of Asai's clustering methods (large-linkage, small-linkage) lack identifiability — they produce identical dendrograms for very different matrices because they ignore mid-range values. Spectral Clustering does not share this limitation.

The result is two groups, not three: Hokkaido (including North Kuril) versus Sakhalin. The three-way Major Division is not statistically supported ‹contested›. Within the Hokkaido cluster, North Kuril patterns most strongly with the northeast Hokkaido group — Samani, Obihiro, Kushiro, Bihoro, Sōya — while southwest Hokkaido forms a second tier. Ono notes this converges with earlier work placing North Kuril close to northeast Hokkaido, while a separate observation by Satō and Bugaeva (via Ono (2020: §4)) clusters 19th-century Kuril items with the Saru/Chitose zone — a tension the 2020 study does not resolve ‹contested›. Ono also cautions that Asai's cluster analysis may never have been intended to establish a canonical classification; its standard status in subsequent literature may exceed its original scope.

159.5.2 Fukazawa and Ono 2024: boundary dialects

Fukazawa & Ono (2024) takes a further methodological step: rather than assigning dialects to groups — which forces a decision about where "boundary dialects" belong — the framework selects a reference dialect, computes similarity from all other lects to that centre, and visualises each variety's "central-dialect-likeness" via correspondence analysis (対応分析) and projection pursuit (射影追跡法). Applied to Ainu with Sōya (the Sakhalin-facing margin) and Asahikawa (the NE–SW junction) as reference centres, the approach reframes these "boundary dialects" as analytically primary objects rather than residue from a grouping procedure. The published abstract reports that the isogloss topology around Sōya and Asahikawa is more complex than any single bipartition captures; the full empirical dataset has not been assessed here ‹speculative›.

159.5.3 Kasuga 2026: 75-locality quantitative classification

Kasuga (2026) extends the quantitative tradition to 75 localities, substantially broadening the geographic coverage of the Hattori–Chiri / Asai matrix. The study — published in Japanese — has not been assessed in full for this chapter; its specific sub-groupings and methodological choices remain to be synthesised ‹speculative›.

159.7.1 Pitch-accent types

Three accent types are attested across Hokkaido Nakagawa & Fukazawa (2022: §3.1):

• Standard (Saru, Chitose, Horobetsu, Asahikawa, Nayoro, Obihiro, Sōya): two placement rules — a closed first syllable takes high pitch on σ1; an open first syllable takes high pitch on σ2.
• Irregular (Yakumo, Oshamambe): when both σ1 and σ2 are open, high pitch falls on σ3. Saru ipérekut corresponds to Yakumo iperékut; Saru hopúni to Yakumo hopuní.
• Unaccented (Shizunai, Urakawa, Samani, Bihoro, Kushiro, Shiranuka): pitch is non-distinctive. Samani tends toward final-syllable high; Shizunai shows no syllable-position relation to pitch at all. The origin of the unaccented type in Hokkaido differs from the Sakhalin unaccented system, where the driver is loss of vowel length rather than loss of pitch (see Chapter 157 (The Final-h Problem and Its Sakhalin Vowel-Length Reflex)).

159.7.2 Word-initial h-elision

In the eastern standard-accent dialects, initial h is elided when the pitch accent falls on the second syllable Nakagawa & Fukazawa (2022: §3.2, Table 4): Saru/Horobetsu huráye 'wash' → uráye in Asahikawa/Nayoro; Saru hokú 'husband' → okú; hekáci 'child' → ekáci. The conditioning is accentual and areal: Horobetsu, Biratori, Nukibetsu, Niikappu, Samani, and Sōya retain initial h, while Obihiro, Kushiro, Bihoro, Asahikawa, and Nayoro do not.

159.7.3 Regressive consonant assimilation

The eastern dialects, most extensively Bihoro, show regressive consonant assimilation at morpheme and compound boundaries Nakagawa & Fukazawa (2022: §3.3, Table 5): hotke 'lie' → hokke (Obihiro, Kushiro, Bihoro); sirsesek 'hot' → sissesek (Bihoro, Asahikawa, Nayoro); tupte → tutte (Bihoro); upsor → ussor 'bosom'. The active patterns include tk → kk, pt → tt, rs → ss, ps → ss, kp → pp, and kt → tt. Nakagawa & Fukazawa (2022) treats these as historical word-form changes rather than productive synchronic sandhi rules. A community reading, noting that Bihoro's patterns are non-conditional and lexeme-specific, proposes them as evidence for a distinct underlying Proto-Ainu consonant series aynu-corpora Discord (2023–2026) (antitwilight, November 2024); the proposal goes beyond the sources consulted and awaits verification ‹speculative›.

159.7.4 The par/car pseudo-correspondence

A small set of words in the mouth-region semantic field shows a geographic alternation: par (southwest) : car (northeast and Kuril) for 'mouth'; papus / patoy : capus / catoy for 'lip'. Nayoro and Asahikawa attest both forms. The alternation affects the labial–alveolar onset position within this semantic field only; there is no comparable p-/c- alternation for e-, i-, o-, u-initial words, which is why Nakagawa & Fukazawa (2022: §3.4, §5.1) call it a pseudo-correspondence rather than a sound correspondence in the comparative linguistic sense.

Place-name evidence from Kirikae (1994) supplies a geographic profile: in the Ishikari watershed, par-type toponyms predominate (11 : 0 against car-type), while Nemuro–Kushiro coast place names are exclusively car-type (13 occurrences). Community observation independently noted that place names favour car broadly across the northeast and the Kuril data aynu-corpora Discord (2023–2026) (nishiyasu, May 2026), consistent with Kirikae (1994) ‹corpus-suggested›.

Four analyses of the alternation appear in the literature, and they are genuinely incompatible ‹contested›:

• Diachronic merger (Kindaichi, Chiri, Naert 1958, via Nakagawa & Fukazawa (2022: §3.4)): either pa- > ca- historically, or both trace to a single proto-form *Xa-.
• Two lexical items from the start (Kirikae 1994, via Nakagawa & Fukazawa (2022: §3.4)): Proto-Ainu already had distinct p- and c- 'mouth' words, later sorted geographically.
• ca- as older (Nakagawa 1996, via Nakagawa & Fukazawa (2022: §3.4)): ca- is attested inside par-region place names (e.g. sanca or ta), suggesting it was the earlier form in that area too.
• Separate morphemes ab initio (Fukazawa 2014/2017, via Nakagawa & Fukazawa (2022: §3.4)): ca- and pa- are independent morphemes throughout their full distribution — pa- also means 'head'; irreducible lexical pairs exist (can : pan 'thin', cas : pas 'run').

Alonso de la Fuente (2022: §2.2.6) adds a reconstruction perspective: the split reflects a palatalized series in Proto-Ainu, with *pʸaɾ as the reconstructed form and *pʸ > ca- in one regional stream, *p > pa- in another. This account integrates the two-morpheme view with the palatalized-series hypothesis (see Chapter 156 (Proto-Ainu Reconstruction: Segments and Accent Classes)), but Alonso himself flags the full palatalized series as typologically motivated rather than philologically forced ‹speculative›.

159.7.5 Further phonological alternations

Additional phonological isoglosses documented by Nakagawa & Fukazawa (2022: §3.5–3.8, Tables 7–17) include:

• -iw : -uy metathesis — Bihoro and Kushiro: siwnin → suynin 'green'; ciw → cuy 'stab'; nociw → nocuy 'star' (Kushiro); riwka → ruyka 'bridge'.
• Vowel alternations — u:o (Sōya opas : standard upas 'snow'); o:e (Yakumo etop : otop 'hair'); V(e,u):i (Samani / eastern sitá : setá 'dog').
• Consonant alternations — y:w (Yakumo puwar, nuwe vs standard puyar, nuye); n:m (nimak : mimak 'tooth' — Yakumo, Nukibetsu, and Niikappu use mimak; Sōya ónke : ómke 'cough').
• -y elision and extension — Saru/Chitose: plural existential okay → oka, extended to the pronoun ciokay → coka. Eastern dialects (Obihiro, Kushiro, Bihoro) show the inverse: particles ka and korka acquire -kay extensions Nakagawa & Fukazawa (2022: §3.7, Table 17). Community observation confirms ka → kay as a Kushiro-diagnostic marker aynu-corpora Discord (2023–2026) (thegodofneet, February 2026) ‹corpus-suggested›.

No phonological isogloss replicates the NE/SW bipartition exactly; each cuts the map at a slightly different line, which is the fundamental reason that "bundle of isoglosses" descriptions are more accurate than a single dividing line.

159.8.1 Person-marking type

The sharpest morphological isogloss separates Saru and Chitose from every other Hokkaido variety. In Saru and Chitose, transitive-verb person marking is carried entirely by prefixes; all other Hokkaido dialects use circumfixes — an object prefix before the stem and a subject suffix after it Nakagawa & Fukazawa (2022: §4.1). The 'I give to you' cell illustrates the contrast: Saru/Chitose eci=kore (prefix only) versus Horobetsu e=kore=as and Ishikari e=kore=an (circumfix in both).

The implication is that Horobetsu patterns morphologically with the geographically distant Ishikari and Tokachi varieties, not with the adjacent Saru region Nakagawa & Fukazawa (2022: §4.1) ‹contested›. The basic-vocabulary lexicostatistic matrix does not capture this: Hattori–Chiri's 200-item list excluded grammatical words and inflectional paradigms, so the morphological and lexical isogloss pictures diverge at this point. Horobetsu is lexically Central-South but morphologically northeast. The full person-marking paradigm and the architectural implications are treated in Chapter 62 (Architecture of the Personal-Affix System: The Four Persons and the S/A/O Paradigms).

159.8.2 Plural marking

The verbal plural morpheme shows a three-way geographic split Nakagawa & Fukazawa (2022: §4.2):

• Sakhalin: (a)hci / hcin (the ‹SA› form; see Chapter 162 (Sakhalin and Kuril Ainu: The External Comparison));
• Shizunai, Horobetsu, Shiranuka: ci;
• Saru, Chitose, Ishikari: pa.

The historical direction is recoverable: ci is the older Hokkaido form, cognate with Sakhalin (a)hci, and pa is an innovation that spread within the Saru–Ishikari corridor, displacing ci there Nakagawa & Fukazawa (2022: §4.2). The verbal number system that houses these morphemes and their interaction with the possessed form paradigm are treated in Chapter 62 (Architecture of the Personal-Affix System: The Four Persons and the S/A/O Paradigms).

159.8.3 Particles and grammatical words

Several grammatical morphemes distribute along isoglosses that match neither the NE/SW bipartition nor the lexicostatistic clusters, illustrating how the different layers of dialectal organisation fail to converge Nakagawa & Fukazawa (2022: §4.3, Tables 21–24).

The allative particle ('toward') takes four attested forms:

• un — Yakumo, Horobetsu, Chitose, Biratori, Shizunai;
• en — Urakawa, Tokachi, Eastern Hidaka;
• (e)ne — Shiranuka, Kushiro, Nemuro, Bihoro, Asahikawa, Sōya, and also Sakhalin and Kuril;
• ekota — an Asahikawa alternative form.

The examples below illustrate two of the attested allative forms in running discourse:

The juxtapositional particle newa (and/with) is absent in Saru and Chitose, where bare juxtaposition does the same work. The successive particle tek functions as a conjunction ('and then') in Shizunai, Samani, Tokachi, Kushiro, Shiranuka, and Ishikari, but as an auxiliary meaning 'do briefly' in Saru, Chitose, and Horobetsu; the Sakhalin cognate teh is the older form (see Chapter 157 (The Final-h Problem and Its Sakhalin Vowel-Length Reflex)). The conditional particle shows a three-way split: ko (Abuta, Horobetsu, Samani, Kushiro, Shiranuka) vs kor (Tokachi, Ishikari, Saru, Chitose) vs Shizunai konno (< kor-no); ko is reconstructed as the older form Nakagawa & Fukazawa (2022: §4.3). The cohortative particle is ro in Saru, Chitose, Horobetsu, Yakumo, Asahikawa, Nayoro, and Sōya; rok in Asahikawa, Nayoro, and Obihiro; and no in Urakawa and Samani.

The imperative particle ya (singular) is specific to Ishikari — another morphological feature that cross-cuts the lexical clusters Nakagawa & Fukazawa (2022: §4.3).