Chapter 29Glide Epenthesis and Vowel-Hiatus Resolution

29.1 Overview

Morpheme composition in Hokkaido Ainu frequently places two vowels in adjacent position across a prefix or morpheme boundary. Four main strategies resolve such hiatus sequences, each conditioned by the type of boundary and the phonological shape of the adjacent morphemes: (1) glide epenthesis, insertion of a transitional [y] or [w] between the boundary vowels; (2) vowel weakening, in which a stem-initial high vowel /i/ or /u/ reduces to a glide after a preceding prefix; (3) glottal insertion, in which [ʔ] marks the syllable onset; and (4) vowel deletion, in which one of the adjacent vowels is elided. The present chapter covers strategies (1) and (2), together with the phonotactic constraint that blocks both of them in certain environments. The glottal-insertion strategy is treated in Chapter 20 (The Laryngeals: Glottal Stop, /h/, and the Final-h Question), and vowel deletion in the context of personal-prefix sandhi is covered in Chapter 31 (Personal-Affix Junctural Sandhi, =an/a= Allomorphy, and Connected-Speech Reduction).

Glide epenthesis applies only after specific prefixes — the indefinite/antipassive prefix i- ‘ANTIP’, the reciprocal u- ‘RECP’, and variably the reflexive si- ‘REFL’ — and only at a prefix boundary, not at a compound morpheme boundary. Vowel weakening is also prefix-conditioned. Both are blocked when a glide before a closed syllable would create an illicit onset cluster, because the (C)V(C) syllable canon of Hokkaido Ainu forbids consonant clusters within a single syllable: see Chapter 21 (Syllable Structure, Phonotactics, and Word-Edge Constraints) Nakagawa (2024: 41–42).

29.2 The glide transition (わたり) and its non-phonemic status

Nakagawa Nakagawa (2024: 34) defines わたり 'glide, transition' as the sound produced while the vocal tract reconfigures between two successive phonemic targets in normal-tempo speech. In his analysis of Saru and Chitose, this transition is not an independent phoneme. The word iyomante 'the bear-spirit-sending ceremony' — phonemically /i-omante/ (< the indefinite-object prefix i- ‘ANTIP’ + omante ‘send away’) — is pronounced [iyomante] in connected speech, but the [y] between prefix and stem is a nonphonemic transition that standard orthography does not treat as a separate segment Nakagawa (2024: 34). Likewise uwepeker 'tell each other stories' is phonemically /u-epeker/ (< u- ‘RECP’ + epeker, a stem posited by the analysis and not attested as a free word ‹speculative›), and the [w] at the prefix–stem juncture is a transitional sound rather than a grammatically independent consonant.

Two diagnostics establish non-phonemic status for Saru and Chitose. The nominalizer allomorphy test: morphemes ending in a genuine /y/ or /w/ take the consonant-allomorph -pe of the nominalizer (e.g. okay-pe, siw-pe), while vowel-initial stems take -p even when preceded by an epenthetic glide: iyomante-p, not *iyomante-pe Shiraishi (2022: §3); Nakagawa (2024: 34). The double-glide test (described in §5 below) provides a second criterion for vowel-weakening contexts. The general analysis of /y/ and /w/ as consonant phonemes in the inventory — and the diagnostics that establish root-phoneme /y/ /w/ — is developed in Chapter 22 (Glides, Vowel Hiatus, and the Diphthong Question). Whether the transition [y]/[w] at morpheme boundaries is phonemic in varieties other than Saru and Chitose is a genuinely contested question, treated in §8.

29.3 Glide epenthesis at the i- and u- prefix boundary

Glide epenthesis operates at the juncture of a vowel-final prefix and a vowel-initial stem. The relevant prefixes are the indefinite-object/antipassive i- ‘ANTIP’ and the reciprocal prefix u- ‘RECP’ Kindaichi & Chiri (1936: 13); Shiraishi (2022: §4.3). After /i/ a [y]-glide appears; after /u/ a [w]-glide. The rules:

R1 ∅ → y / i- ___ V  (at a prefix boundary)
R2 ∅ → w / u- ___ V  (at a prefix boundary)

The epenthetic glide is determined by the prefix vowel — /i/ triggers [y], /u/ triggers [w] — independently of the quality of the following stem vowel. Table 1 shows y-insertion before /o/, /oi/, /e/, and /u/, and w-insertion before /e/ and /o/. A derivation for the canonical pair:

29.4 The *CVVC constraint as a blocker of glide formation

The (C)V(C) syllable canon prohibits consonant clusters in onset position and disallows more than one vowel per syllable Nakagawa (2024: 41–42); Shiraishi (2022: §4.1). When glide formation at a morpheme boundary would produce a string that violates this canon, the process is blocked.

The first-person-singular agent prefix ku- ‘1SG.A’ before a stem whose first syllable is closed (CVC) shows the blocking most directly. In /ku-ikra/ 'I send it', weakening the stem-initial /i/ to [y] would place [y] immediately before the coda /k/ of ikra's first syllable, producing *[yk] — an illicit CC onset at the following syllable boundary. The vowel is therefore not weakened: ku-ikra is the surface form, with the vowel hiatus [ku.ik.ra] maintained Nakagawa (2024: 52–53); aynu-corpora Discord (2023–2026) (nukopoli, 2024-11-28) ‹corpus-confirmed›.

The applicative prefix e= ‘APPL’ before the same stem provides a parallel: /e=ikra/ cannot surface as *[eykra] because [yk] is an illicit onset. The actual form is eíkra [e.ík.ra], where the /i/ of ikra bears accent as a syllable nucleus — confirming it has not been reduced to a glide Nakagawa (2024: 52–53); aynu-corpora Discord (2023–2026) (nukopoli, 2024-11-28) ‹corpus-confirmed›. The open-σ1 stems itasa and iwanke contrast with ikra: e= + itasa → eytasa; e= + iwanke → eywanke. The open first syllable (CV) of those stems permits the glide without creating a cluster.

The constraint has direct practical consequences for corpus normalization. The io→iyo and ue→uwe patterns cannot be applied uniformly across all morpheme boundaries: community member thegodofneet (2025-04-24, aynu-corpora Discord (2023–2026)) notes that no form *eciyoka for eci-oka is attested in the corpus ‹corpus-confirmed›. A rule-based normalization that applied io→iyo or ue→uwe at every juncture would produce unattested forms. Glide-bearing spellings are appropriate only where the morphosyntactic conditions of §3 and the phonotactic licence of §4 are jointly satisfied and where corpus evidence confirms the variant.

29.5 Vowel weakening after personal and applicative prefixes

Vowel weakening (母音弱化) converts a stem-initial /i/ or /u/ to a glide after a preceding prefix. No new segment is inserted; the high vowel already present at the stem onset is converted to a consonantal element and loses the ability to bear accent Nakagawa (2024: 40). The process has been verified for Saru and Chitose; its distribution in other Hokkaido varieties requires further documentation Nakagawa (2024: 40).

The rules:

R3 /i/ → [y] / V= ___ (stem-initial /i/, open first syllable)
R4 /u/ → [w] / V= ___ (stem-initial /u/, open first syllable)

In koytak (< ko= + itak), the stem-initial /i/ of itak 'speak' weakens to [y]; in yaykowepeker, the stem-initial /u/ of uepeker weakens to [w]. The open-first-syllable restriction of §4 applies here: weakening is possible only when the stem's σ1 is open (CV), so no onset cluster results.

A double-glide test distinguishes R3/R4 from the presence of an underlying /y/ or /w/ in the stem. If the stem-initial segment were an underlying /y/ or /w/, weakening it to [y] or [w] would yield a geminate. The verb iunin 'ache' has a phonemic stem-initial /i/ (not */yi/); with the prefix ku- ‘1SG.A’ it surfaces in Chitose as [kuyunin], with a single [y], not *[kuyyunin] Nakagawa (2024: 40). The same test in Saru: somo eyunin ya? (not *eyyunin). The single [y] confirms that R3 weakens a stem vowel rather than inserting a new glide alongside an underlying one.

The accent consequences of vowel weakening — the nucleus shift when a formerly accentable σ1 vowel becomes a non-accentable glide — are treated in Chapter 26 (Accent in Compounds and under Affixation and Cliticization). For the related deletion of the prefix vowel in ku- and ci- before non-/i/ vowels (ku-omare → k=omare), see Chapter 31 (Personal-Affix Junctural Sandhi, =an/a= Allomorphy, and Connected-Speech Reduction) and Chapter 62 (Architecture of the Personal-Affix System: The Four Persons and the S/A/O Paradigms).

29.6 Glottal insertion when glide cannot appear

When the morphosyntactic conditions of §3 do not permit glide epenthesis and the adjacent vowel is non-high so that R3/R4 weakening does not apply, [ʔ] fills the hiatus. Tamura (1970: 586–587, 1996: xv, cited in Shiraishi 2022 §4.3) notes that the glottal onset is most clearly audible when the second vowel is accented or belongs to a closed syllable: aˈʔep 'food', aˈʔoypep 'tableware', eˈʔikka, teˈʔeta 'a long time ago'. In non-prominent environments the transition is gradual rather than sharply glottalized Shiraishi (2022: §4.3).

A morphologically restricted h-insertion applies within the affiliative (possessed) form of nouns when two vowels would be adjacent: onaa → ona-h-a 'his/her father'; sikii → siki-h-i 'his/her eyes' (Kindaichi & Chiri 1936: 14, cited in Shiraishi 2022 §4.3 n.8). This process is restricted to possessive derivation and is treated in Chapter 45 (Morphophonology of the Affiliative Suffix and Echo/Copy Vowels).

Whether [ʔ] at these junctures constitutes an underlying phoneme /ʔ/ or is a predictable boundary marker is contested; the debate is taken up in Chapter 20 (The Laryngeals: Glottal Stop, /h/, and the Final-h Question).

29.7 Vowel deletion as an alternative hiatus strategy

Vowel deletion provides a fourth resolution route, particularly for non-high vowels at morpheme boundaries where glide formation is unavailable. Shiraishi Shiraishi (2022: §4.3) documents: keran (< kera=an 'is delicious'), sisotta (< siso or=ta 'on the fireplace'), and noya noya (< noye a noye a). Deletion is conditioned by speech rate and prosodic context and is not in free variation with glide epenthesis at the same boundary. The full account, including the accent interactions that block resyllabification, is in Chapter 31 (Personal-Affix Junctural Sandhi, =an/a= Allomorphy, and Connected-Speech Reduction).

29.8 The phonemic status of the glide transition across dialects

Three analyses of the morpheme-boundary [y] and [w] bear on whether they are underlying phonemes or surface transitions, and the three rest on genuine cross-dialectal differences.

Nakagawa's non-phonemic analysis (Saru/Chitose). The transition [y] and [w] at prefix boundaries are non-phonemic in Saru and Chitose: their appearance is entirely predictable from the shape of the adjacent morphemes, and neither the nominalizer allomorphy diagnostic nor the double-glide test provides evidence for underlying /y/ /w/ at these junctures Nakagawa (2024: 34–35).

Okuda's phonemic analysis (Shizunai). In the Shizunai variety, which lacks contrastive pitch accent, the glide vs. glottal distinction at a morpheme boundary carries lexical weight that accent handles elsewhere. The minimal pair first observed by 志賀雪湖 (Okuda 2025 Okuda (2025)): [suwat] 'pothook' — an opaque, fused compound — vs. [suʔat] 'pot bail' — a transparent su- + at with the boundary maintained. In an accentless dialect this /w/ vs. /ʔ/ contrast is phonemically distinctive, giving /suwat/ vs. /suʔat/ (= /suat/). Okuda further documents the pairs /u'a/ ~ /uwa/, /i'o/ ~ /iyo/, /i'e/ ~ /iye/ as additional Shizunai minimal pairs, and concludes that "one cannot say, without specifying dialect, that わたり /y/ /w/ are non-phonemic" Okuda (2025) ‹contested›.

Satō's two-process analysis (Chitose). Satō (2003, cited in Shiraishi 2022 §4.3) proposes that in Chitose, [y] and [w] in sequences /ie/, /ia/, /io/, /iu/, /ue/, /ua/, /uo/ at a morpheme boundary occur in the same phonetic environment as their glottal-stop-bearing counterparts and contrast with them; he therefore posits underlying /y/ /w/ phonemes in those contexts. The glide after ku- and e= prefixes, by contrast, arises through surface glide formation (i → y, u → w) — a distinct process from lexical /y/ /w/. The analysis thus involves two interacting processes: underlying phonemes in some contexts, and surface-phonological glide formation in others Shiraishi (2022: §4.3) ‹contested›.

In Saru and Chitose the nominalizer-allomorphy and double-glide diagnostics do not reveal evidence for underlying /y/ /w/ at the prefix boundaries treated in §3; the non-phonemic analysis is adopted for those varieties. In Shizunai the accentless system exposes a contrast masked elsewhere, and the glides there are phonemically distinct from [ʔ]. The question is structurally linked to the /ʔ/-phoneme debate — positing /ʔ/ at all vowel-initial syllables reanalyses the Shizunai contrasts differently — and both are considered together in Chapter 20 (The Laryngeals: Glottal Stop, /h/, and the Final-h Question) ‹contested›.