Chapter 28Assimilation and Cluster Simplification: Coda /r/, Nasals, and Clusters

28.1 Overview

Hokkaido Ainu has two intersecting regressive assimilation systems operating across morpheme, compound, and word junctures: coda /r/-sandhi and coda /n/-sandhi. In /r/-sandhi, the coda /r/ of one morpheme is deleted and the following consonant geminates; /r/ is thereby the most morphophonologically productive segment in the language Nakagawa (2024: 35–38); Shiraishi (2022: §6.1); Kindaichi & Chiri (1936). Independently, the coda nasal /n/ undergoes two distinct processes: it assimilates in place of articulation to a following labial or velar (an allophonic rule), and before following continuants /s/, /y/, and /w/ it triggers a morphophonological change — /n/-sandhi proper — which is more restricted in domain than /r/-sandhi Nakagawa (2024: 30, 36–37); Shiraishi (2022: §3, §6.1). A third, dialect-specific set of cluster simplifications is attested in eastern varieties, most prominently Bihoro, but Nakagawa (2024: 36) treats these as historically fixed rather than productive Nakagawa & Fukazawa (2022: §3.3).

All rules described here are regressive: the coda segment adapts to the following onset. None are notated in the standard romanization orthography, which writes underlying forms (ran wa, not ram ma); both sandhi types are nonetheless fully productive in speech Nakagawa (2024: 37–38). The phoneme /r/ and its coda release are described in Chapter 19 (The Rhotic /r/ and Coda-r); glide insertion at vowel hiatus is treated in Chapter 29 (Glide Epenthesis and Vowel-Hiatus Resolution); the interaction of these rules with personal-affix clitics is treated in Chapter 31 (Personal-Affix Junctural Sandhi, =an/a= Allomorphy, and Connected-Speech Reduction).

28.2 Coda /r/-sandhi

When a /r/-final morpheme immediately precedes a consonant-initial morpheme, the coda /r/ is deleted and the following consonant is lengthened, yielding a surface geminate. The rule applies word-internally, at compound boundaries, and across word boundaries; it is not blocked by pauses within an utterance, and it crosses sentence boundaries without restriction (§4.1 below) Nakagawa (2024: 35–37); Shiraishi (2022: §6.1).

Shiraishi (2022: §6.1), drawing on Shiraishi (2001), characterizes the trigger consonants as the coronal class — /n/, /c/, /t/, /r/ — with /s/ and /y/ also attested as triggers in northern dialects (Chiri 1956: 343, cited in Shiraishi (2022: §6.1)). The assimilation can thus be stated as:

/r/ → C / _C[+coronal, ≠r]: the coda rhotic is deleted before a following coronal onset, which then geminates.

The earliest firsthand field record of this rule set is Simeon's 1967 survey of Hiratori, Horobetsu, and Higashi-Shizunai Simeon (1968: 22–24). Working from citation forms with echo vowels (oara, koro, kuru, asiri), he states the same alternations — oat tuye 'he cut completely', ot teeta 'very ancient times', kon nispa 'my master', an kun ne 'it is a man', asin-no 'newly', and dissimilated ku kot cise 'my house'. Two of his observations add texture the modern sources lack: the rules were mandatory in Higashi-Shizunai but optional in Hiratori and Horobetsu, where the unassimilated base form was the preferred variant; and his Hiratori informant — Kayano Shigeru — judged the assimilated variants "indicative of an older form of speech". Both points suggest that by the 1960s the sandhi was receding in the Saru valley as a live process while remaining categorical further east ‹speculative›.

28.3 Coda-nasal place assimilation

The coda nasal /n/ assimilates in place of articulation to a following labial or velar consonant: it surfaces as [m] before /p/ and /m/, and as [ŋ] before /k/. These are allophonic realizations fully predictable from the following context. The coda /m/ does not undergo velar assimilation before /k/ and remains [m] in that environment Nakagawa (2024: 30); Shiraishi (2022: §3).

The orthography consistently writes /n/ in all environments (an=pe, hon=kor, hanku), never the allophonic surface values [m] or [ŋ]. Nakagawa (2024: 30) and Shiraishi (2022: §3) both treat [m], [n], [ŋ] as allophonic realizations of the two phonemes /m/ and /n/, with /n/ conditioned by the following consonant and /m/ invariant.

Whether /n/ and /m/ are phonemically neutralized before labials is a point of orthographic debate. Forms such as matanpus ~ matampus are written with /n/ or /m/ before /p/ depending on the source; no consensus across dictionaries and authors has been established ‹contested› aynu-corpora Discord (2023–2026) (nukopoli, 2023-12-05). The loanword kampi 'paper' (from Japanese kami) raises the same issue: an etymological kanpi versus a phonetically assimilated kampi aynu-corpora Discord (2023–2026) (nukopoli, 2024-11-16) ‹contested›.

28.4 Coda /n/-sandhi

Before the [+continuant] consonants /s/, /y/, and /w/, a coda /n/ does not assimilate in place but is replaced by a glide or deleted: before /s/ and /y/ it surfaces as [j] (romanized y), yielding an apparent geminate with the following continuant; before /w/ the nasal is deleted and only /w/ remains. Shiraishi (2022: §6.1), following Shiraishi (2001), characterizes the trigger class as the continuants and describes n-sandhi as the spread of the feature [+continuant] from the following onset onto the preceding coda. The formal statement is:

/n/ → [j] / _{s, y}; /n/ → ∅ / _w

The coda /m/ does not undergo this rule. The form homsu (from hom-su) surfaces as [homsu], not *[hoysu], confirming that n-sandhi targets /n/ specifically and not the broader nasal class; /m/ and /n/ feed different morphophonological processes before continuants ‹corpus-suggested› aynu-corpora Discord (2023–2026) (nukopoli, 2024-03-06). The /n/+/s/ output [js] — written as y in poy seta — behaves phonologically as a coda /y/ in the resulting syllable.

28.5 Lexicalized sandhi outputs

Several Hokkaido Ainu lexical items preserve a sandhi output as their current lexical form, the underlying source either obsolete or reconstructed.

The best-documented case is attus ‘elm-bark robe’. The proto-form is reconstructed as at-rus; the Saru dialect attests an intermediate form ahrus (reflecting lenition rather than full assimilation); the general Hokkaido form attus reflects the lexicalization of /r/+/s/ → [ss] sandhi at the compound boundary aynu-corpora Discord (2023–2026) (nukopoli, 2024-12-26) ‹corpus-confirmed›. The word attus is attested 24 times in the corpus across multiple narrators and collections (小松哲郎, 神崎雅好, 相原典明, アイヌタイムズ; 平賀さだも, 平取町アイヌ口承文芸), confirming that the geminated form is the current lexical norm throughout Hokkaido.

A second piece of evidence for the productivity of /r/+/t/ → [tt] comes from a VITS-based Ainu TTS system: the model independently surfaced the sequence mosir ta 'land' + locative as [mositta] in generated speech, apparently learning the assimilation from training data without explicit phonological rules ‹corpus-suggested› aynu-corpora Discord (2023–2026) (nukopoli, 2026-05-20). This anecdotal ML observation does not add to the primary-source evidence for the rule — established in Nakagawa (2024: 36) and Shiraishi (2022: §6.1) — but illustrates how pervasive /r/-sandhi is in connected-speech corpora.

Among other compound-boundary cases, ar suy 'once more' surfaces as [assuy] via /r/+/s/ → [ss]; retar cikap 'swan' surfaces as [retat̚cikap] via /r/+/c/ → [t̚c]. The compound-internal t+i → ci rule — illustrated by mat-ikor → macikor — is the one sandhi rule that applies without exception in all Hokkaido dialects Nakagawa (2024: 36) and is treated in Chapter 19 (The Rhotic /r/ and Coda-r).

Additional lexicalized forms noted by the community include ketketcep (reconstructed from *ket~ket-se-pe) and nisatcawot (from nisat-sawot), both showing /t/+/s/ → [tc] assimilation at compound junctures ‹speculative› aynu-corpora Discord (2023–2026) (antitwilight, 2025-11-03).

28.6 Bihoro and eastern-dialect cluster simplification

Nakagawa & Fukazawa (2022: §3.3) document cluster simplifications in the Bihoro dialect — and to varying degrees in Obihiro, Kushiro, Asahikawa, and Nayoro — that go beyond the productive sandhi rules described above. In these varieties, specific heterorganic sequences of two obstruents, or an obstruent plus fricative, are simplified to a geminate in specific lexical items.

Nakagawa (2024: 36) treats these as historical word-form changes rather than synchronic productive rules. The diagnostic is substitution-productivity: a Kushiro speaker who produces hokke for 'lie down' does not produce *pek kari for pet kari 'the river did' in spontaneous speech, showing that the /tk/ → [kk] simplification is lexically specific and not a live rule generating [kk] from all /tk/ sequences Nakagawa (2024: 36). The community endorses this analysis: the changes are word-specific in Bihoro and may reflect a proto-phoneme issue rather than a productive synchronic rule ‹contested› aynu-corpora Discord (2023–2026) (antitwilight, 2024-11-17, citing Nakagawa & Fukazawa 2022: §3.3). The /rs/ → [ss] Bihoro cases are phonologically parallel to productive /r/+/s/ → [ss] sandhi in §2, but speakers do not generalize the alternation to new /rs/ sequences. The broader dialect typology is treated in Chapter 160 (Hokkaido Dialect Microvariation: Phonology and Morphosyntax).

28.7 Rule inventory and domain summary

The productive rules of this chapter are collected below. Formal statements follow the notation X → Y / Z (X undergoes change to Y in environment Z).

The principal structural asymmetry in the system — /r/-sandhi is domain-free, /n/-sandhi is phrase-bounded — is formulated by Shiraishi (2022: §6.2) as the distinction between a pure phonological rule and a P-structure rule. Neither is written in the standard romanization; both apply in connected speech Nakagawa (2024: 37–38).

These rules interact with the personal-affix system (ku= before vowel-initial stems; en= assimilation), treated in Chapter 31 (Personal-Affix Junctural Sandhi, =an/a= Allomorphy, and Connected-Speech Reduction), and with the possessed-form morphophonology of nouns. The syllable structure that sets coda position is described in Chapter 21 (Syllable Structure, Phonotactics, and Word-Edge Constraints), and the consonant inventory from which all triggering segments are drawn is described in Chapter 16 (The Consonant Inventory and Its Phonetic Realization).