🗣️ 4. Connected speech

Ukrainian adjusts the shape of words to their neighbours more than most European languages, and it writes many of these adjustments. The best known are the euphonic alternations (чергування для милозвучності): the preposition and prefix у ~ в, the conjunction і ~ й, the preposition з ~ із ~ зі ~ зо, and pairs like б ~ би, ж ~ же, chosen by whether the neighbouring sounds are vowels or consonants. They all serve one tendency, the avoidance of clusters of consonants and of vowels in hiatus, which keeps the stream of syllables open (Bas-Kononenko 2019; 3. Syllable, stress and intonation §1.2). Inside words, the language assimilates consonants in voicing, place and softness, simplifies clusters, and alternates consonants at morpheme boundaries; these are partly written, partly only pronounced.

This chapter gives the rules and measures how the written language follows them: §1 the principle, §§2–5 the alternating words with counts, §6 assimilation, §7 cluster simplification, §8 length at junctions, §9 consonant alternations at morpheme boundaries, §10 verse and song, and §11 what Mova does. Notation as in 1. Consonants.

The counts. The euphonic choices were counted on 2026-11-06 in three bodies of text, each split into words and punctuation the same way (a word is a run of Cyrillic or Latin letters, apostrophes and hyphens):

The left context of the alternating word is V (the previous word ends in a vowel letter), C (it ends in a consonant letter or ь) or P (start of the sentence or after a punctuation mark: the "pause" of the rules); the right context is V or C by the first letter of the next word, with special classes named in each table. The counts measure written usage. They do not show what the writer pronounced, and editing practice differs between the corpora: the ParlaMint transcripts, for instance, almost never write й for і.

1. The principle

The 2019 orthography opens each euphony rule with the same purpose: "to avoid a cluster of letters that stand for consonants hard to pronounce and to achieve euphony" (Pravopys 2019 §§23–25). Two things follow. First, the alternations are rules of writing that reflect a rule of speech, and the code accepts that writers deviate from them: each paragraph ends with a note that "deviations occur, caused by the demands of rhythm and melody or by the author's preferences". Second, the rules are local: they look at the last sound of the previous word and the first sound of the next, and at a pause between them. Smal-Stockyj (1913 §3) observes the same mechanism at work in voicing across word boundaries (пас би 'would graze' pronounced [pɑzbɪ], §6.1): in Ukrainian the phrase, not the word, is the domain of these processes.

2. у ~ в

The preposition у/в 'in, into, at' and the prefix у-/в- alternate (Pravopys 2019 §23).

Some words have only one form in any position: only у in увага 'attention', умова 'condition', установа 'institution', указ, ультиматум, уява 'imagination', Угорщина, Урал; only в in влада 'power', власний 'own', вплив 'influence', взаємини 'relations' (§23.3). A few pairs differ in meaning: вдача 'character' / удача 'luck', вклад 'deposit' / уклад 'way of life', вправа 'exercise' / управа 'board, administration', вступ 'introduction' / уступ 'ledge'. The name of the country is Україна in any position; poetry uses Вкраїна (§23 n.: до Вкраїни свій погляд я з любов'ю повертав, F. Malyc'kyj). In the news sample Україн- occurs 101 430 times and Вкраїн- once.

Usage. Share of у among у + в (prepositions only, written as separate words), by context:

(News: 469 706 tokens of у/в, 53.5% у; ParlaMint 2 272, 36.6% у; IU 3 071, 51.4% у.)

What the table shows:

• The strongest rules hold almost without exception: у before в, ф (97–99% in the news) and в between vowels (96.6%). These are the cases where the "wrong" choice produces a real cluster (в вагоні [u̯ʋɑɦɔnʲi]) or a hiatus (у Одесі).
• "Between consonants" is a weaker rule in practice: 17% of the news tokens have в there (брат в Києві), and the ParlaMint transcripts are evenly split. The transcribers seem to write в as the default form whenever no vowel–vowel or в/ф conflict forces a choice.
• After a vowel before a consonant the rule asks for в, but 40% of the news tokens have у. The rule itself makes an exception for clusters (у своїй, у двері), and the counts follow it: after a vowel, у makes up 38.5% of tokens before a single consonant + vowel (61 899 against 98 771 в), but 59.8% before a word beginning with two consonant letters (23 931 against 16 106).

The same alternation runs in the prefix: вчитель / учитель, вже / уже 'already', вчора / учора 'yesterday', вперше / уперше 'for the first time', вдома / удома 'at home'. Here the written language has settled on the в-form as the default: in the news sample вже 24 613 against уже 3 311, вперше 3 033 against 235, вчитель 197 against 43 (lower-case forms). уже stands mostly after a consonant (2 317 of 3 302 tokens), but even after a consonant вже is more frequent (3 937 against 2 317). In this part of the system usage has moved ahead of the rule: the prefix forms behave like separate words chosen by habit, and only у before в (уві сні, увійти) is still regular.

A doubled prefix вв- appears after a vowel before в: вважати 'to consider', ввійшли 'came in', ввічливі 'polite', вві сні 'in sleep', ввісьмох 'eight together' (§23 n.); after a consonant the same words begin with у: Він увійшов.

3. і ~ й

The conjunction і 'and' (and the particle і 'also, even') alternates with й; so does the initial і- of some words (іти / йти 'to go', імовірно / ймовірно 'probably', ім'я / ймення) (Pravopys 2019 §24).

After й, я, ю, є, ї the orthography recommends та or і rather than й (гай і долину, гай та долину; §24.4). та 'and' is the third member of the system and the usual way to avoid both і after a vowel and й after a consonant.

Usage. Share of й among і + й:

(News: 218 983 tokens, 11.4% й; ParlaMint 2 397, 1.5% й; IU 2 780, 19.5% й.)

The rule against й is obeyed: й between consonants, at a pause and before an iotated letter is almost absent. The rule for й is not: even between vowels, where the orthography calls for й, the news writes і in half the cases and ParlaMint in 91%. The written language treats і as the unmarked form and й as a stylistic choice, more frequent in fiction and in careful writing (the IU treebank, with fiction in it, has the highest й rates). This is a case where the code describes a literary ideal that most written registers do not follow; the euphony it aims at is achieved in practice also by та.

4. з ~ із ~ зі ~ зо

The preposition з 'from, with, of' has four forms (Pravopys 2019 §25):

із and зі are interchangeable before some clusters (разом із сходом сонця = разом зі сходом сонця). The prefix has the same set, with с- before к, п, т, ф, х (§5.3).

Usage. Shares of the four forms (news / ParlaMint / IU):

(News: 231 597 tokens: з 184 580, із 32 671, зі 14 308, зо 38.)

The pattern is clear in the news and IU: з before vowels and simple consonants, із between consonants, із or зі before sibilants, with зі preferred before clusters (зі стола, зі Львова). In the ParlaMint transcripts з is the default even before sibilants (two thirds of the cases).

зо. зо is the rarest form and the one where usage has left the norm behind. In the news sample it occurs 38 times in all, mostly as the approximative зо два 'about two' (9), зо три (7), зо дві (3). The phrase зо мною 'with me', which the orthography gives as an equal variant of зі мною, does not occur once, against 391 tokens of зі мною. Зо мною is now a literary and dialect form; a description of today's written usage would list зі мною alone.

5. Other alternating pairs

5.1 Short and full particles and conjunctions

The particles б/би and ж/же follow the vowel–consonant rule almost perfectly: 99.7% би after a consonant, 99.85% ж after a vowel. They are enclitics, and the choice is part of the phonology of the phrase (зробив би, зробила б; він же, вона ж). The conjunctions щоб/щоби and хоч/хоча are lexical choices: щоби is the Galician and literary-older form, хоча the neutral one.

5.2 The reflexive -ся ~ -сь

The reflexive postfix has a short form -сь after a vowel, in verse, colloquial speech and Western usage (умивсь 'washed himself', сміялась 'laughed'), beside the standard -ся (умився, сміялася). The literary norm treats -сь as stylistic; 13. Verbs gives the forms.

5.3 The prefix з- ~ с- ~ зі- ~ із-

The prefix з- becomes с- before к, п, т, ф, х: сказати 'to say', спалахнути 'to flash', стовкти, сфотографувати, схилити; before other letters it is з- (збавити, звести, зжитися, зсадити, зчепити, зшиток), rarely із- (ізнов, ізсередини), and before two consonant letters of the root mostly зі- (зібгати, зігнути, зіграти, зіпсувати, зіткнення) (Pravopys 2019 §31.1). The other prefixes ending in a voiced consonant, без-, від-, між-, над-, об-, перед-, під-, понад-, роз-, через-, keep it before a voiceless consonant: безкорисливий, відкриття, підтримка, розтягнути (§31.2), in writing and in the standard pronunciation (§6.1).

5.4 н after prepositions

The third-person pronouns він, вона, воно, вони take a prothetic н- after a preposition: його 'him' but до нього, з ним, на неї, у них, біля неї, від нього. The н is a morphological rule that continues an old preposition-final n (вън, сън, кън); it applies to prepositions governing the pronoun, not to other words before it (без нього, but біля його хати 'near his house', with the possessive). 12. Pronouns, determiners and numerals has the paradigms.

6. Assimilation

6.1 Voicing

Regressive voicing. A voiceless obstruent before a voiced one becomes voiced, inside words and across the boundary of a phrase:

The last two are Smal-Stockyj's (1913 §3): "where voiceless stops are written before б, д, ж, з, they sound voiced, e.g. також [taɡʒe], and so also in connected speech, e.g. пас би". The spelling does not show this voicing.

No regressive devoicing. A voiced obstruent before a voiceless one stays voiced: дуб [dub], ніжка [ˈnʲiʒkɑ], берізка [bɛˈrʲizkɑ], розповідь [rɔzˈpɔʋʲidʲ] 'story', безпека [bɛzˈpɛkɑ] 'safety' (1. Consonants §8). The exception is г before a voiceless consonant, which becomes [x]: легко [ˈlɛxkɔ], нігті [ˈnʲixtʲi], кігті, вогкий 'damp'. In fast speech the prefixes роз-, без- can lose their voice before voiceless consonants; the norm keeps [z].

6.2 Place and manner

Dental sibilants and affricates assimilate to a following hushing consonant, and hushing consonants to a following dental, with lengthening where two sounds merge. The standard orthoepic rules, also the core of the rule-based phonemizer of Senyk et al. (2025):

The forms with -шся and -ться are the most frequent: in the news sample -ться ends 41 640 words in 300 000 sentences, -шся 130 (the second person is rare in news). Senyk et al. (2025) needed a separate exception for -ться in their phonemizer, because the general rule order would produce the wrong result for it.

Bas-Kononenko (2019) points to a disagreement in the orthoepic dictionaries over how to describe коротший [kɔˈrɔʧːɪj] 'shorter' and ворітця [ʋɔˈrʲiʦʲːɑ] 'little gate': one dictionary treats the result as a "fusion of two sounds into an affricate", while it is regressive assimilation by place and manner, which turns the stop into an affricate; she also notes that коротший with [ʧː] is possible "only at a quicker tempo".

6.3 Softness

A dental before a soft dental or before і softens in pronunciation without being written soft: пісня [ˈpʲisʲnʲɑ] 'song', кузня [ˈkuzʲnʲɑ] 'smithy', радість [ˈrɑdʲisʲtʲ] 'joy', гостя [ˈɦɔsʲtʲɑ] 'female guest'. The orthography accordingly writes no ь between the two (Pravopys 2019 §27.3: світ, свято, слід, сніг, цвях, щастя), except after л (їдальня, кільце, §26.1.3). The degree of this softening varies with region, generation and tempo: it is consistent in пісня, радість, weaker in свято and сміх, where the first consonant is at most semi-palatalized. The opposite case, a soft consonant before a hard one, keeps its softness and is written with ь (батько, різьба, Грицько).

7. Cluster simplification

Three-consonant clusters with a stop in the middle lose the stop, and the orthography writes the loss (Pravopys 2019 §28):

Exceptions keep the stop in writing: зап'ястний, кістлявий, пестливий, хвастливий, хвастнути, шістнадцять, дев'яностники; випускний, вискнути (§28). In adjectives from foreign nouns in -ст the т is written but not pronounced: баластний, компостний, контрастний, форпостний. The same silent т appears in the native шістнадцять [ʃʲisˈnɑʦʲːɑtʲ] 'sixteen'.

Where the cluster arises at a suffix boundary the orthography keeps the letters and the pronunciation simplifies: студентський [studɛnʲsʲkɪj], туристський [tuˈrɪsʲkɪj], пропагандистський (§19.2 keeps -стськ- in writing). 6. Spelling rules and the 2019 orthography covers the spelling.

8. Length at junctions

Where a morpheme ends and the next begins with the same consonant, both are kept and the consonant is long: відділ [ʋʲidʲːiɫ], беззвучний, ввічливий, законний [zɑˈkɔnːɪj], розрісся (Pravopys 2019 §29). The long consonants before an old j (життя, знання, ніччю) are described in 1. Consonants §9. In connected speech the boundary of two words gives the same lengthening without any written trace: від дому [ʋʲidːɔmu], з зошитом.

9. Consonant alternations at morpheme boundaries

The following alternations are not euphonic and not automatic; they belong to particular inflections and suffixes and are written. They are listed here because they are the morphophonology every paradigm in chapters 11–17 relies on (Pravopys 2019 §§12–22; Smal-Stockyj 1913 §7, who traces them to the old palatalizations).

The comparatives and the -н- adjectives have their own lists of exceptions (легший; помічний from поміч, нічний from ніч; рушник, соняшник, торішній; §§20–21). The к ~ ч and г ~ ж alternations in vocatives are among the reasons the vocative is morphologically distinct (11. Nouns).

10. Verse and song

The euphony rules are rules of prose. Verse uses the alternating forms to fit the metre, and the orthography allows it explicitly (§§23–25: deviations "caused by the demands of rhythm and melody"): у or в, і or й, з or зі add or remove a syllable, so the poet's choice is driven by the line. Вкраїна (four syllables, stressed on the third) beside Україна (five) is the best-known case: Shevchenko and the poets after him use both. Song lyrics go further: the folk song uses full and short forms freely (-ся/-сь, мені/мні), elides vowels, and keeps dialect forms for rhyme and metre. Songs are a register with their own grammar; 37. Register and variation and 8. Regional varieties, diaspora and surzhyk describe it.

11. How Mova handles connected speech

Mova has no rules of euphony, assimilation or alternation for Ukrainian.

• The alternating forms are separate words. UD Ukrainian lemmatizes each variant as itself: в (lemma в), у (у), і, й, з, із, зі, зо each carry their own lemma in both treebanks (counted 2026-11-06: e.g. в ADP 2 753, у ADP 2 026, й CCONJ 316 and PART 258, in lower-case forms). The English tree parser trained on them through en xpos-fill sees eight unrelated prepositions and conjunctions; nothing tells it that у and в are one word.
• The kg crate groups some of them. kg/src/uk.rs treats у and в together when it decides the case after the preposition (an animate noun with identical genitive and accusative after у/в is taken as genitive: у друга), and з, із, зі, зо together (genitive first: зі стола). But the relation label it stores keeps the word itself (adverbial:у+v_mis, adverbial:в+v_mis), as kg — the Mova knowledge graph specifies, "needed to restore the sentence". So в Одесі and у Одесі produce different relations, and the generator (kg/src/textgen.rs) restores the original variant rather than choosing it: after a reordering of chunks the restored у/в may no longer fit its new neighbours. The round-trip gate counts such a sentence as "same words, different order", not as an error.
• No pronunciation. Assimilation, voicing and softness (§6) exist only in this manual. The phonemizer of Senyk et al. (2025) is the open model of these rules; Mova has none.
• Alternations at morpheme boundaries (§9) reach kg only as forms listed in VESUM (dict_corp_vis.txt.bz2, CC BY-NC-SA, research and tests only).

What would be needed. A small, testable Rust module could (1) map the euphonic variants to one lemma for analysis (у/в → у, і/й → і, з/із/зі/зо → з, б/би, ж/же, -ся/-сь), and (2) choose the variant in generation from the neighbouring letters by the rules of §§2–5, which are local and deterministic. The counts of this chapter are its test data: a generator should reproduce the robust rules (у before в, ф; в between vowels; би/б, же/ж) at the 97–99% the news shows, and should not be held to the weak ones (й between vowels), where usage itself is split. Neither part is built.

Sources for this chapter

• Ukrajins'kyj pravopys (2019). §§12–31. Official document, not protected by copyright. Library books/u/uk/ukrainskyi-pravopys-2019 (uk-gram-pravopys-2019).
• Smal-Stockyj, Stephan. 1913. Ruthenische Grammatik. §3 (voicing in connected speech), §7 (palatalization). Public domain. Library books/s/st/stockyj-1913-ruthenische-grammatik (uk-gram-smal-stockyj-1913).
• Bas-Kononenko, O. V. 2019. Vymovni normy jak mirylo mylozvučnosti ukrajins'koji movy. Ukrajins'ka mova 2019 (1): 103–114. Library card CC BY 4.0; the article prints CC BY-NC-ND 4.0, so paraphrased only (bas-2019-kononenko-orfoepichni-normy); doi.org/10.15407/ukrmova2019.01.103
• Senyk, Anastasiia, Mykhailo Lukianchuk, Valentyna Robeiko & Yurii Paniv. 2025. Context-Aware Lexical Stress Prediction and Phonemization for Ukrainian TTS Systems. UNLP 2025. CC BY 4.0 (senyk-2025-context-aware-lexical-stress-prediction).
• Counts (2026-11-06): Leipzig ukr_news_2024 (1M sentences), CC BY (Goldhahn, Eckart & Quasthoff 2012), leipzig-news-2024; UD Ukrainian-ParlaMint 2.18 (CC BY-SA 4.0) and UD Ukrainian-IU 2.18 (CC BY-NC-SA 4.0, counted only), ud-treebanks-v2.18, # text lines of train, dev and test.
• Mova: kg/src/uk.rs (preposition handling), kg/src/textgen.rs, kg — the Mova knowledge graph; en/src/main.rs (xpos_fill).

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/uk/04-connected-speech/