Chapter 4Phonology

4.1.1 Obstruents and the absence of a voicing contrast

Voicing is not contrastive anywhere in the system. Chiri (1942: §2(1)) states flatly that no pair of words is distinguished by voicing: the plosives and the affricate are voiceless in careful speech, but voiced variants [b d ɡ dʑ] appear freely after nasals, as in his membe ‘grandmother’, unji for unci ‘fire’ and henge for henke ‘grandfather’. Piłsudski heard the same instability on the East coast: between vowels and after y the plosives “waver” between the voiceless and voiced series, which Dal Corso (2024: 61) renders as partially voiced [p̬ t̬ k̬] (e.g. pájgara for paykara ‘spring’). The two coasts are reported to differ in degree: Piłsudski found voicing commoner in the west, Chiri likewise heard more voiced tokens on the West coast, while in Murasaki's West-coast recordings of the 1960s voicing of p is unattested and even t k are rarely voiced ((Dal Corso 2024: 61)). Dal Corso suggests contact with Japanese as one reason for the retreat of voicing in the later data.

The affricate c is alveolo-palatal [tɕ]: Piłsudski insists that it is neither [ts] nor English-type [tʃ] but “akin to the sound of the Polish ć” ((Piłsudski 1912: 7); (Dal Corso 2024: 63–64)). After nasals it has the voiced variant [dʑ], as in únʑi for unci ‘fire’. A regular source of c is /t/ before i, where t shifts to c without exception, e.g. kuci ‘his belt’ from kut-i ((Chiri 1942: §16,6)); this feeds the coda alternations described in Chapter 5. In the East-coast dialects of Ay and Tunayci, p and k (but not t) are palatalised before e [ɛ], e.g. emújḱe for emuyke ‘all’, tamb́e for tanpe ‘this thing’ (Dal Corso 2024: 61–62); nothing comparable is recorded for the West coast.

The sibilant s varies between [s] and [ɕ]. Before and after i, and generally in coda position, it is [ɕ] on both coasts (śine for sine ‘one’ and iśo for iso ‘bear’ next to i, and ahkaś for ahkas ‘walk’ in coda); elsewhere word-initially the two sounds vary freely, Piłsudski recording both suj and śuj for suy ‘again’ ((Piłsudski 1912: 5); (Dal Corso 2024: 62–63)). Chiri adds that plain [s] is the usual West-coast value, [ɕ] being obligatory only next to i and frequent in codas (Chiri 1942: §2(2)).

The glottal fricative h occurs both as onset and — its most distinctive role — as coda. As a coda it is a fricative whose place is determined by the preceding vowel: velar [x] after a o e (mah [max] ‘woman’), palatal [ç] after i, and labialised, near [ɸ], after u (uh [uɸ] ‘take’); Chiri, who writes this segment x and counts it a separate phoneme, notes that utterance-finally it is so weak as to escape notice Chiri (1942: §2(3)). We follow Hattori and Murasaki in treating onset [h] and the coda fricatives as one phoneme /h/. As onset before or after u, /h/ is [ɸ] systematically on the West coast but only variably in Piłsudski's data (fura beside húxkara), and he also heard voiced [ɦ] between vowels ((Dal Corso 2024: 63)).

4.1.2 The tap r

The single liquid is a tap [ɾ] between vowels. Word-initially, and after n, Piłsudski describes an articulation made “simultaneously” with a t placed further back than usual, pointing to a retroflex [ʈ] or affricated [ʈɽ], with much variation between speakers ((Piłsudski 1912: 8–9); (Dal Corso 2024: 64)). The hardening is lexicalised in part of the Sakhalin area: tetara for retara ‘be white’ is widespread, and the northern East-coast dialect of Nairo turns initial r into t in a whole set of words (tup for ruh ‘ice’) (Tangiku 2022: §2.6). In codas r does not occur: where Hokkaidō has syllable-final r, Sakhalin shows either an echo vowel or h (§4.4.4).

4.1.3 Nasals and approximants

The nasals are m and n. Coda m survives only in careful speech: Chiri (1942: §16,2) observes that word-final m is normally pronounced n (isan for isam ‘not exist’, aman for amam ‘grain’), and that before k t n the shift is categorical (enko ‘half’, kinta ‘in the mountains’). Murasaki's corpus bears this out: Asai Take's tales have isan teh an beside slower isam. Conversely n assimilates to following labials and glides (pompe from pon-pe ‘little thing’; (Chiri 1942: §16,3)); the fuller sandhi picture is taken up in Chapter 5. Piłsudski additionally records palatal [ɲ] for /n/ before e and word-finally (etunńe ‘not want’), and a rare [ŋ] (Dal Corso 2024: 62).

The approximants are w and y. In the East-coast corpus /w/ is regularly written v, a labiodental value Piłsudski reports for all the dialects he documented; Dal Corso treats [v] as an East-coast allophone of /w/ rather than a separate phoneme (Dal Corso 2024: 63). Syllable-finally the two approximants form the falling diphthongs ay uy ey oy and aw iw ew ow; that these offglides count as consonants, not as vowel sequences, is shown by morphology — stems in -y -w select the consonant-stem allomorphs of suffixes, e.g. nominalising -pe rather than -h (ray-pe ‘dead one’), a point already made as a “morphophoneme” argument by Chiri (1942: §2(5)).

4.1.4 The glottal stop

Every vowel-initial word begins phonetically with a glottal stop, strongest in accented syllables; Chiri regards it as fully predictable and therefore not a phoneme Chiri (1942: §1(3)). Hattori's phonemicisation of Rayciska includes /ʔ/ in the consonant chart but bars it from codas (reported in (Tangiku 1998: §2-1)), and Dal Corso admits it for West Sakhalin only “provisionally, in free variation” (Dal Corso 2024: 59–60). Piłsudski never marks it, which may reflect his self-confessed “Slav ear” rather than its absence in the East (Dal Corso 2024: 60). We write no glottal stop: word-initially it is automatic, and word-internally its only role is to keep adjacent vowels apart at morpheme boundaries (see Chapter 5). Murasaki's apostrophes ('oman, 'e'ani) are accordingly not reproduced in our examples.

4.4.1 Syllable types and the bimoraic minimum

Five syllable shapes occur: V, CV, VC, CVC, and — the shape that sets Sakhalin apart from Hokkaidō — CVV with a long nucleus ((Itabashi 2001: 89)). The two vowel-initial types are restricted: in native simplex words V and VC syllables stand only word-initially, and where word formation would create one elsewhere it is absorbed into the preceding syllable (sik-o ‘eye attaches’ > siko ‘be born’, rup-us ‘ice attaches’ > rupus ‘freeze’) or removed by haplology or vowel deletion (kera-an > keran ‘be tasty’). Chiri accordingly reduces the basic templates to CV and CVC, with the onsetless types as word-initial variants carrying a zero onset (Chiri 1942: §5).

Following Murasaki's moraic description, Itabashi (2001: 89) treats Sakhalin as a mora-counting language: a CV sequence, a bare V, and a coda consonant each contribute one mora, and the minimal accent-bearing free form is two moras. This bimoraic minimum is what lies behind the lengthening of open monosyllables noted in §4.3: a content word of the bare shape (C)V would be a single mora, too light to carry the accent it requires, and so surfaces as CVV — nii ‘tree’, kuu ‘bow’, too ‘day, lake’ — while clitics and affixes, which are unaccented, escape the requirement and stay short. The same arithmetic excludes long vowels from closed syllables: CVVC would run to three moras, and the language repairs it by shortening (imii → imi-yehe above).

4.4.2 The coda inventory

The coda is where Sakhalin phonotactics departs most sharply from Hokkaidō. Hokkaidō Ainu closes syllables with any of nine consonants, p t k s r m n w y; Sakhalin permits only six, s n m w y h, with coda m marginal (§4.1.3) — the plosives, c, r and the glottal stop are excluded (Hattori 1967, as reported in (Tangiku 1998: §2-1); (Itabashi 2001: 89)). The northern East-coast dialects from Taraika to Nairo, together with Tarantomari on the West coast, stand outside the restriction and keep Hokkaidō-style p t k (Tangiku 2022: §2.3); they are in this respect the most conservative Sakhalin varieties (§4.4.3).

Chiri drew the quantitative consequence of this restriction in a small piece of arithmetic Chiri (1942: §6). The vowels supply 5 onsetless V syllables; the 11 onset consonants combine with the 5 vowels into 51 CV syllables (11 × 5 = 55, minus the four non-occurring ti wi wu yi); these figures are common to both areas. The codas then multiply out differently: VC gives 5 × 9 = 45 syllables in Hokkaidō but 5 × 6 = 30 in Sakhalin, and CVC gives 51 × 9 = 459 against 51 × 6 = 306. In total Hokkaidō has 560 possible basic syllables and Sakhalin 392 — the coda restriction deletes nearly a third of the syllable space. (Chiri counts his Sakhalin-only coda fricative x, our h, among the six; and his count has no CVV type because he treated length as non-distinctive, §4.3. Adding the long-vowel syllables would not affect the comparison, which turns entirely on the codas.)

4.4.3 Coda neutralisation

The signature phonotactic rule of Sakhalin Ainu is the neutralisation of coda obstruents. Syllable-final *p *t *k all surface as h (phonetically [x ç ɸ] by vowel environment, §4.1.1), except that the coronal and dorsal *t *k after i palatalise to s (sik → sis ‘eye’); labial *p gives h even after i (cip → West cih ‘boat’, East cis) ((Chiri 1942: §16,1); (Tangiku 2022: §2.3)). Final *r goes its own way and is treated in §4.4.4. The rule applies word-internally as well as finally (ahkas ‘walk’ : Hokkaidō apkas; yuhke ‘be strong’ with root yup, (Dal Corso 2024: 62)), and it is the engine of the stem alternations treated in Chapter 5, since the underlying stop reappears before vowel-initial suffixes.

A second source of coda h is the resolution of geminates and other stop clusters. Where Hokkaidō has a long stop or a heterorganic stop sequence, Sakhalin has h plus the following stop: the geminates wakka : wahka ‘water’, satte : sahte ‘dry’, and the cluster hotke : hohke ‘lie down’ ((Itabashi 2001: 93–94); (Chiri 1942: §16,8)). Nothing new needs to be stated: the first stop has no following vowel of its own, so it debuccalises by the rule just given, while the second survives as the onset of the next syllable. Geminates in loanwords are reshaped the same way (§4.4.5).

The northern East-coast dialects from Taraika to Nairo, and Tarantomari on the West coast, stand outside the rule and retain tek, sik, rup like Hokkaidō (Tangiku 2022: §2.3). In Piłsudski's other East-coast dialects the final fricative was evidently faint: he usually writes nothing at all (pate for pateh ‘only’) and admits confusing ee ‘eat’ with eh ‘come’ and ruu ‘road’ with ruh ‘sea ice’, an error Chiri diagnoses precisely as the weakness of utterance-final [x] ((Chiri 1942: §16,1); (Dal Corso 2024: 61–62)).

4.4.4 Coda r: echo vowel and debuccalisation

Coda r patterns apart from the stops. Chiri gives two resolutions for syllable-final *r: either the preceding vowel is copied, opening the syllable — utara ‘people’, kisara ‘ear’, mukara ‘axe’, koro ‘have’ against Hokkaidō utar, kisar, mukar, kor — or the r debuccalises like a stop, word-finally in utah and before a consonant in unahpe ‘aunt’ (Hokkaidō unarpe) (Chiri 1942: §16,4). He presents the routes as alternatives, citing utara and utax from the very same page of Piłsudski's Materials — so no purely segmental conditioning can decide between them.

The corpus shows the choice to be in large part dialectal. In the East-coast texts the echo vowel is all but categorical. In the tale Piłsudski took down at Takoye from the nineteen-year-old Kusurikoya, utara recurs from one end to the other — onneru utara ‘the old people’, mahnekuh utara ‘the women’, ohkayo utara ‘the men’ — and debuccalised utah never appears; Sakaguchi, surveying the whole East-coast folktale corpus, states the generalisation directly: the variants uta ~ utah are West-coast forms, his eastern materials showing almost exclusively utara ((Sakaguchi 2020a: 187); (Sakaguchi 2020c)). The h-forms cluster in the West-coast records — Murasaki's conversational material and the modern revitalisation register that continues that norm. We therefore read the second route as a West-coast extension of debuccalisation from the obstruents to r: having converted *p *t *k to h, the western varieties drew the one remaining eligible coda into the same lenition, while the East keeps r distinct by epenthesis. Even in the west the two forms coexist word by word (utara and utah are both current), so the residue is lexical variation rather than rule.

The echo vowel is phonologically real, not an orthographic convenience: stems that acquired it (koro ‘have’, nukara ‘see’, Hokkaidō kor, nukar) still behave as consonant-final under suffixation, the hidden r resurfacing or assimilating (koro + -te → konte ‘let have, give’), in contrast with stems whose final vowel is original, such as tura ‘be with’ (Tangiku 1998: §§2–3). These alternations belong to the stem morphophonology and are taken up in Chapter 5.

4.4.5 Loanword adaptation

Borrowings are reshaped to this template. Initial r and foreign l are avoided: Tungusic luca ‘Russian’ was borrowed as nuca (Chiri 1942: §17,2), and 蠟燭 ‘candle’ (Chinese, via Japanese rōsoku), through the intermediate ratcaku, appears as cahraku ‘lamp’ with metathesis of the initial r ((Chiri 1942: §20); Chapter 25 (Language contact and loanwords)). Japanese accent and vowel quantity are mapped onto length: kane ‘metal’ gives kaani, where Hokkaidō has accented káni (Tangiku 2022: §2.1), and tuki ‘cup’ gives tuuki (Bugaeva 2022). Geminates of the source language are resolved like native stop clusters: Japanese teppō ‘gun’ yields tehpo, the geminate pp surfacing as hp (Chapter 25). Contact vocabulary also flowed from the Amur–Sakhalin languages, e.g. tanku ‘hundred’ from Tungusic (Tangiku 2022: §3.3). Russian, by contrast, left almost no integrated lexicon: its mark on Sentoku Tarōji's Cyrillic letters is chiefly the written apparatus of dates and sums and the personal and place names adapted to Ainu phonology — Aleksandr → Ariksandoru (with the cluster broken by an echo vowel and l → r), lesnichiy ‘forester’ → Lesneciy — rather than borrowed words (Chapter 25) (Tangiku & Ogihara 2001).

4.5.1 The glottal onset as a prosodic signal

The glottal stop of §4.1.4 belongs in part to prosody. Chiri's formulation is prosodic from the outset — the onset appears on the initial vowel of an accented syllable, and is “especially conspicuous” in Sakhalin Chiri (1942: §1(3)) — and the West-coast orthographic tradition tracks the same conditioning: Murasaki's apostrophes sit precisely on accent-bearing vowel-initial morphemes ('itah, 'oman, and word-internally at junctures such as 'e'ani), not on every orthographic vowel. For the East coast, where Piłsudski wrote nothing, the segment betrays itself indirectly, through sandhi. A word-final stop debuccalises only when no vowel follows; before a genuinely vowel-initial word it should resyllabify as an onset and keep its stop value. At the juncture nah an ‘being so’, however, Piłsudski hears [nax án]: the coda of nah is treated as pre-consonantal, which means the accented an must begin with a consonant — the glottal onset (Dal Corso 2024: 61). The accent-linked glottal stop is thus recoverable for the East from its phonological effect even though no transcriber recorded it directly.

4.5.2 Pitch, length, and Proto-Ainu

Set against Hokkaidō, the Sakhalin system looks like one half of an old inheritance. The correspondences of §4.3 — Hokkaidō pitch matching Sakhalin length in the regular cases, with residues like niná : niina ‘crush’ that no single-feature account derives — point, with (Vovin 1993) and (Itabashi 2001: 95–98), to a proto-language in which vowel length and pitch were independent distinctive features (against Hattori's length-only reconstruction, §4.3). Each daughter branch then kept one feature and made the other predictable. Hokkaidō phonemicised pitch, length surviving only as a phonetic correlate of the accented open syllable; Sakhalin phonemicised length, pitch surviving only as the automatic second-mora contour described above. The two systems carry the same old prominence by different means, which is why the cognate sets translate so regularly between an accent mark on one side of the strait and a doubled vowel on the other.

Kuril Ainu can be placed in this scheme only tentatively. Its record is confined to early wordlists — Krasheninnikov, Voznesenskii, Dybowski, Torii — which are fragmentary and mutually inconsistent on quantity: ‘two’ appears as tuup ~ tuuh ~ tubic across collectors. From this material Itabashi (2001: 90–91) nonetheless argues that Kuril had a pitch accent of the Hokkaidō type, a distinctive low-to-high rise, while the irregular long-vowel spellings leave open whether length was also contrastive. If that is right, the length-for-pitch reanalysis is a specifically Sakhalin innovation, not a shared northern development — and the pieces described in this chapter, the length contrast (§4.3), the bimoraic minimum (§4.4.1) and the accent-conditioned glottal onset, form a single prosodic complex organised around quantity.