Chapter 161Lexical Dialectology and the Dialect Atlas

161.1 The survey apparatus and its primary data

The systematic comparison of lexical variation across Hokkaido Ainu rests on a survey carried out by Hattori Shirō and Chiri Mashiho in 1960. Working from a 200-item basic-vocabulary list elicited at 19 localities — 13 across Hokkaido and 6 in Sakhalin ‹SA› — together with Torii's 1903 Kuril ‹KU› materials, they scored each item for cross-dialectal cognacy as positive, negative, or partial (Nakagawa & Fukazawa (2022: §2.1)). The 13 Hokkaido localities were Yakumo, Oshamambe, Horobetsu, Biratori, Nukibetsu, Niikappu, Samani, Obihiro, Kushiro, Bihoro, Asahikawa, Nayoro, and Sōya. The resulting cognacy percentages supplied the first quantified measure of lexical distance between Hokkaido varieties and remain the empirical layer beneath every subsequent dialectometric classification.

The survey's reach is narrower than the item count suggests. Of the 200 elicited items, only 91 actually contributed to the cognacy matrix — the remaining 109 showed no cross-dialectal stem difference and therefore carried no discriminating information (Nakagawa & Fukazawa (2022: §2.3)). Those 91 variable items are the effective data for Hokkaido lexical dialectology: a diagnostically rich but typologically restricted window, concentrated on body parts, basic environmental vocabulary, and core verbal concepts.

Two further resources expand and document the survey results. Hattori's 1964 dialect dictionary (Hattori (1964)) organized the comparative material across the 19 localities into a single reference work, giving paradigmatic forms and attestations for each item. Fukazawa (2025) has since brought to light a previously unpublished document: Chiri Mashiho's own handwritten and annotated copy of the basic-vocabulary survey tables, preserved at the National Ainu Museum (Fukazawa (2025)). These tables, covering at minimum the Yakumo, Oshamambe, and Horobetsu localities, carry Chiri's marginal variant readings and hesitation marks, providing a check on the published cognacy scores and revealing forms that may have been suppressed in the final scoring.

161.2 Coverage and documentation gaps

The geographic distribution of the 13 Hokkaido localities is uneven. The Pacific coast and the Ishikari–Hidaka corridor are represented by multiple sites; the Sea-of-Japan coast contributes only Yakumo, and the Sea-of-Okhotsk coast only Bihoro. The northernmost point, Sōya, was effectively a single speaker — Ben Kashiwagi — so its data reflect an idiolect rather than a community sample (Nakagawa & Fukazawa (2022: §2.4)). Many areas that certainly held Ainu-speaking populations into the nineteenth century, particularly along the Sea-of-Japan coast, left no post-Meiji lexical record at all ‹corpus-suggested›.

Kasuga's 2026 expansion to 75 localities, incorporating place-name evidence and historical documents, begins to fill some of these gaps (Kasuga (2026)), but systematic multi-speaker lexical data from the poorly attested regions remain largely absent. The place-name record is geographically broader than the spoken-language survey, though it introduces its own complications: toponyms reflect an earlier language state, their orthographic conventions vary across sources, and phonological changes may have operated differently on frozen names than on the living lexicon ‹speculative›.

161.3 Isogloss types and geographic distribution

Mapping the 91 variable items across the 13 survey localities produces several recurrent distribution patterns. Nakagawa and Fukazawa, building on Nakagawa's earlier geographic analysis, distinguish five types (Nakagawa & Fukazawa (2022: §5)):

These five types do not bundle neatly into coinciding isoglosses. Each lexical item can produce a different boundary, with particular localities appearing in opposite groups depending on which item is considered. This non-congruence is consistent with a dialect continuum undergoing diffusion from multiple centers rather than a single ancestral branching event (Nakagawa & Fukazawa (2022: §5); Ono (2020: §3)) ‹contested›.

161.4 The eastern–western lexical divide

The most robust lexical signal dividing northeastern from southwestern Hokkaido is a set of interrogative and temporal stems. The southwestern dialects (including Saru, Chitose, Horobetsu, and Yakumo) use interrogative stems beginning in hem- / hen-: 'when' is hempara and 'how many' is hempak. The northeastern dialects (Asahikawa, Nayoro, Sōya, Obihiro, Kushiro, and Bihoro) use ne--initial forms: nempara and nempak (Nakagawa & Fukazawa (2022: §5.1)). The Saru–Chitose group adds its own specific interrogatives: hemanta 'what', hunna 'who'.

Nakagawa and Fukazawa read the interrogative distribution as a diachronic contrast. The Saru–Chitose group preserves two etymologically distinct stems — an interrogative hVm- / hVn- base and a separate indefinite ne- base — while most northeastern dialects generalized the indefinite stem to interrogative function, yielding ne-mpara by analogy with other ne- indefinites (Nakagawa & Fukazawa (2022: §5.2)). On that analysis the southwestern and Saru–Chitose patterns are retentions and the northeastern forms are innovations ‹contested›. Community discussion at the aynu-corpora Discord independently arrived at the same two-stem analysis — interrogative hem / hun versus indefinite ne — and further segmented hunna as hun + na and hempak as hem + pak (aynu-corpora Discord (2023–2026), nukopoli 2023-12-08, proposed). The segmentation is corroborated by the comparative data in Nakagawa and Fukazawa's analysis ‹corpus-confirmed›.

161.5 The allative particle as a morphological isogloss

A morphological item that closely parallels the lexical NE/SW divide is the allative postposition. Southwestern dialects use un; eastern Hidaka and Tokachi use en; and the northern and far-eastern dialects (Sōya, Asahikawa, Bihoro, Kushiro, Nemuro) use ne or its extended form, with Asahikawa additionally using ekota (Nakagawa & Fukazawa (2022: §4.3)). Because this particle distribution matches no other lexical or morphological isogloss exactly, it functions as one of the single clearest indicators for locating an undescribed text within the Hokkaido dialect area.

The allative alternation is treated as a morphological isogloss here rather than a purely lexical one because the variation involves bound grammatical morphemes, not independent lexical stems. Its geographic distribution nevertheless contributes to the same east–west picture that the basic-vocabulary survey drew, and the two types of variation align without being identical — a further demonstration that no single isogloss bundle defines the Hokkaido dialect area as a whole.

161.6 The par / car 'mouth' isogloss

Among the lexical differences documented by the survey, the word for 'mouth' has attracted the most analytical attention. Southwestern Hokkaido dialects use par ‘mouth’ and its affiliative form paro; northeastern dialects, Sakhalin ‹SA›, and Kuril ‹KU› varieties use car ‘mouth’ and caro. The related body-part terms follow the same division: 'lip' is papus / capus or patoy / catoy depending on region (Nakagawa & Fukazawa (2022: §3.4)).

The contrast superficially resembles a regular sound correspondence between a southwestern /p/-initial and a northeastern /c/-initial dialect group, but it does not behave like one. No other set of unrelated Hokkaido words shows the same /p/ ∼ /c/ alternation in any systematic pattern; the variation is restricted to the 'mouth' semantic cluster and a small number of body-part and motion terms (can / pan 'thin'; cas / pas 'run'). Nakagawa and Fukazawa therefore call it a "pseudo-correspondence": it is geographically distributed like a sound change but is not phonologically general (Nakagawa & Fukazawa (2022: §3.4)).

Four analytical positions have been advanced for the origin of the split, all reported here via Nakagawa and Fukazawa (Nakagawa & Fukazawa (2022: §3.4)):

• Single-change account (Kindaichi, Chiri, Naert 1958): a historical change *pa > ca propagated through the northeastern region, or both derive from a single ancestral form *Xa.
• Two-'mouth'-words account (Kirikae 1994): Proto-Ainu already had two distinct words for 'mouth'; their regional distribution is a fossilized retention of this original synonymy.
• Age-area / ABA account (Nakagawa 1996): car is the older form, preserved at the northeastern periphery and in ‹SA›/‹KU›, and retained in Saru-region place names (e.g. sancaorta); par is a Saru-Hidaka innovation that spread through the southwest, eventually producing the compound par-unpe 'tongue' (lit. 'that-which-lies-in-the-par-region').
• Two-morpheme account (Fukazawa 2014/2017): ca- and pa- were separate morphemes from the outset, each appearing across a range of body-part and spatial vocabulary (pa- also meaning 'head'), so the alternation reflects a productive morphemic distinction rather than a lexical split or sound change.

All four accounts rest on the same survey dataset and on place-name evidence with its own chronological uncertainties; none can be ruled out on the available materials ‹contested›. Place-name counts — 11:0 in favor of par- in the Ishikari basin, 13 instances of car- in Nemuro and Kushiro combined (Kirikae 1994, reported via Nakagawa & Fukazawa (2022: §3.4)) — are consistent with a sharp geographic boundary but do not settle the mechanism. A community discussion at the aynu-corpora Discord (nishiyasu 2026-05-07, proposed) independently observed that place-name distributions favor car in the northeast and east, strengthening the place-name evidence without resolving the analytical question (aynu-corpora Discord (2023–2026)) ‹corpus-suggested›.

A further phonological account, not among the four original proposals, reconstructs a Proto-Ainu palatalized initial *pʸaɾ, whose northeastern reflex yielded /c/ while the southwestern region generalized plain /p/ (reported via Nakagawa & Fukazawa (2022: §3.4)) ‹speculative›. A community member (nukopoli 2023-12-08) independently proposed a related palatalization pathway via a reconstructed intermediate *ti (aynu-corpora Discord (2023–2026), proposed) ‹speculative›; the two mechanisms converge in positing a palatal intermediate but differ in the structural details.

161.7 ABA distributions and retention at the periphery

Several variable items show an ABA distribution: the same form occurs at the geographic margins of Hokkaido — both east and west — while a different form occupies the center. Under the wave-model reading standard in Japanese dialectology and adopted here by Nakagawa (reported via Nakagawa & Fukazawa (2022: §5)), the peripheral form (A) is typically the older retention and the central form (B) is an innovation that has not yet completed its spread to the edges.

The word for 'tongue' is the primary example. The peripheral dialects — Yakumo in the southwest, Kushiro and Bihoro in the east — retain aw ‘tongue’. The central dialects, including Saru, Chitose, Horobetsu, and Asahikawa, use parunpe, a transparent compound: literally 'that which lies inside the par region / the mouth-interior'. The ABA reading takes aw as the form reconstructable for pre-innovation Hokkaido and parunpe as an innovation coined in the par-using southwestern area and spreading outward without displacing aw at the farthest edges (Nakagawa & Fukazawa (2022: §5)) ‹speculative›. That interpretation is also consistent with the Nakagawa 1996 age-area account of the par / car alternation: if par 'mouth' is itself an innovation displacing car, then parunpe 'tongue' (par + locative/relational element) is a secondary compound built on that innovation and will share the same geographic profile — which is exactly what the survey data show ‹speculative›.

The 'father' terms offer a second, less clear-cut case. The form mici appears along the Pacific Coast arc from the Kuril Islands through the eastern coast to Horobetsu, a distribution that cuts across the NE/SW divide. The form iyapo is recorded at an isolated inland point near the Saru river mouth. Yamada's proposed "Pacific Tribe" — a hypothetical coastal population whose lexical traces are visible in forms like mici and in place-name elements (koyka / koypok) — is a speculative interpretation of this coastal continuity (reported via Nakagawa & Fukazawa (2022: §5)) ‹speculative›.

161.8 Semantic field variation beyond the basic list

The 91-item variation set is limited to high-frequency basic vocabulary. Systematic cross-dialectal comparison in other semantic fields — animal and plant names, technological vocabulary, ceremony-related terms — is absent from the survey, and the dialect dictionary (Hattori (1964)) covers the same 19 localities but does not extend systematically beyond the survey items. The domain of place names has been used as a partial supplement: because toponyms encode older vocabulary and are geographically pinnable, they allow isoglosses to be reconstructed for areas with no living-speaker data.

Fukazawa's 2025 edition of Chiri Mashiho's annotated survey tables (Fukazawa (2025)) adds a qualitative dimension to the Yakumo, Oshamambe, and Horobetsu data. Chiri's marginal notes reveal the range of variant forms that a skilled lexicographer heard during elicitation but collapsed in the final binary cognacy score — information that the published 1960 dataset conceals. This document is a significant resource for reassessing the resolution of the survey data, though it covers only three southwestern localities and cannot compensate for the absent data from underattested regions.

One kinship term deserves note as a clear inter-dialectal lexical isogloss outside the basic-vocabulary list: the terms for 'father' (see above, §aba-patterns) and the broader kinship vocabulary in general have not been systematically mapped across the survey localities. Regional variation in kinship terminology is attested in scattered sources but lacks a comprehensive comparative treatment comparable to the basic-vocabulary survey ‹corpus-suggested›.

161.9 Saru–Chitose as a lexically distinct group

Within the southwestern zone, the Saru–Chitose group shows a cluster of lexical and morphosyntactic innovations that set it apart from neighboring Horobetsu and Biratori, even though basic-vocabulary cognacy alone would place Saru, Chitose, and Horobetsu in a common southwestern group. The distinctively Saru–Chitose items include the interrogative forms hemanta 'what' and hunna 'who', the existential plural oka, the indefinites nep 'something' and nen 'someone', and the prefix eci- 'I to you (pl)' (Nakagawa & Fukazawa (2022: §5)).

Several of these Saru–Chitose forms recur in ‹SA› Sakhalin: the form hemata (the Sakhalin cognate of hemanta) 'what' and the prefix eci- appear in both Saru–Chitose and ‹SA›, yet are absent from much of northeast Hokkaido. Nakagawa and Fukazawa find this distribution "hard to believe [is] independent" and leave it as an unresolved puzzle about the internal phylogeny of Ainu (Nakagawa & Fukazawa (2022: §5)) ‹contested›. One reading is that these forms are innovations shared by an ancestral coastal or southern group before the Sakhalin branch diverged; another is that they are archaisms preserved separately in Saru–Chitose and ‹SA› while the northeast innovated. Neither account has been demonstrated from the available data.

161.10 Lexical data and dialectometric classification

The cognacy matrix computed from the 91 variable items has been the primary input for the main proposals about how Hokkaido dialects group. Asai's 1974 cluster analysis of this matrix proposed a partition into northern (Sōya), eastern (Obihiro / Kushiro / Bihoro / Asahikawa / Nayoro), central-southern (Yakumo / Oshamambe / Horobetsu / Biratori / Nukibetsu / Niikappu), and eastern-Hidaka (Samani) zones for Hokkaido, together with a major split between the Hokkaido+Kuril group and Sakhalin (reported via Nakagawa & Fukazawa (2022: §2.2); Ono (2020: §2)).

Ono's 2020 statistical reanalysis demonstrates that the claimed three-way "Major Division" of Hokkaido / North Kuril / Sakhalin is not supported when methodologically appropriate algorithms are applied to Asai's original data. Asai's linkage methods (single-linkage and complete-linkage agglomerative clustering) ignore mid-range cognacy values that are precisely where the evidence for sub-groupings lies; spectral clustering over the same matrix yields two groups — Hokkaido plus North Kuril vs Sakhalin — without any robust internal three-way split (Ono (2020: §3–4)) ‹contested›. Within the Hokkaido+Kuril group, North Kuril clusters with the northeastern Hokkaido localities (Samani, Obihiro, Kushiro, Bihoro, Sōya) rather than with the southwestern group, while the southwestern and Saru–Chitose varieties form the second sub-cluster. Ono also notes that Asai (1974) may never have intended to canonize a tripartite dialect classification, yet his result became the standard reference beyond its original purpose.

Fukazawa and Ono's 2024 paper takes a complementary methodological direction (Fukazawa & Ono (2024)). Rather than assigning each locality to a group, it maps the similarity of every Hokkaido point to two "boundary" reference dialects — Sōya (the Hokkaido/Sakhalin boundary variety) and Asahikawa (the NE/SW boundary variety) — through correspondence analysis combined with projection pursuit. This dissolves the unresolvable question of which group a boundary dialect belongs to and instead treats geographic gradience as the object of description. The resulting picture is one of overlapping similarity fields converging on the boundary localities rather than a set of discrete blocs.

A persistent tension runs through the classification literature: lexical and morphosyntactic isoglosses do not always converge on the same groupings. The clearest case is Horobetsu. Basic-vocabulary cognacy places Saru, Chitose, and Horobetsu together in a southwestern cluster; person-marking morphology (Horobetsu uses prefix–suffix circumfixes while Saru–Chitose use prefix-only marking) places Horobetsu with the geographically distant Ishikari and Tokachi varieties (Nakagawa & Fukazawa (2022: §4.1)) ‹contested›. Lexical distance is the broadest signal; finer grammatical features track different, partly independent histories. The morphosyntactic dimension is developed in Chapter 160 (Hokkaido Dialect Microvariation: Phonology and Morphosyntax), and the full statistical classification debate in Chapter 159 (Hokkaido Dialect Classification and Dialectometry).

161.11 The Sakhalin and Kuril data as comparative evidence

The 1960 survey included six Sakhalin localities and the Torii 1903 Kuril materials alongside the 13 Hokkaido points, giving the lexical comparisons a triangulable external dimension. For the present chapter, two findings are directly relevant.

First, the form car ‘mouth’ — northeastern in Hokkaido — is the standard word for 'mouth' in ‹KU› and ‹SA›, attested in early Kuril sources including Krasheninnikov (1738) and Golovnin (1811) and throughout the Sakhalin records (reported via Nakagawa & Fukazawa (2022: §3.4)). If the age-area account of the par / car split is correct (§mouth-isogloss), the agreement between northeastern Hokkaido, ‹SA›, and ‹KU› in using car supports reading that form as older, and par as an innovation limited to southwestern Hokkaido ‹speculative›.

Second, Ono's reanalysis of the full dataset, including the Sakhalin and Kuril points, places North Kuril lexically closer to northeastern Hokkaido than to southwestern Hokkaido, with Sakhalin forming a clearly separated external group (Ono (2020: §3–4)). This topological finding — NE Hokkaido+Kuril as one cluster, SW Hokkaido as a second, Sakhalin as the external comparator — is the most statistically defensible summary of the lexical evidence as the dataset currently stands ‹contested›. The Sakhalin and Kuril varieties in their full phonological and morphosyntactic dimensions are treated in Chapter 162 (Sakhalin and Kuril Ainu: The External Comparison).