🗣️ 10. Word classes

Ukrainian school grammar counts ten parts of speech (частини мови, častyny movy): six notional, or "independent", classes that name and inflect, three function classes that serve them, and the interjection, which stands apart. Universal Dependencies (UD) divides the same words into seventeen tags, and the large morphological dictionary VESUM uses a third scheme. This chapter sets the three side by side (§§1–3), gives the tests that tell the classes apart (§§4–11), lists the words that belong to more than one class (§12), and ends with what Mova's two Ukrainian paths, the English parser trained on UD and the kg crate built on VESUM, do with them (§13).

Examples are in Ukrainian with a transliteration on first use (the scholarly system of chapter 00) and a gloss in single quotes. Counts were made on 2026-11-06 over UD Ukrainian 2.18 (IU, CC BY-NC-SA 4.0, counted but never quoted; ParlaMint, CC BY-SA 4.0), the VESUM lemma list (dict_corp_vis.txt.bz2, CC BY-NC-SA 4.0, counts only) and the Leipzig ukr_news_2024 sample of one million news sentences (CC BY). Unmarked examples are constructed.

1. The ten classes of the school tradition

The first six are the notional classes (самостійні, samostijni), the next three the function classes (службові, službovi), and the interjection is the tenth. Three further groups are recognised in most modern descriptions, either as classes of their own or as subclasses:

• Verb forms that look like other classes. The participle (дієприкметник, dijeprykmetnyk: написаний 'written') inflects like an adjective, and the adverbial participle (дієприслівник, dijepryslivnyk: написавши 'having written', пишучи 'writing') is invariable like an adverb. School grammar treats them as forms of the verb; some older grammars listed them separately (13. Verbs).
• Predicatives, also called "words of the category of state": треба 'it is necessary', можна 'one may', шкода 'it is a pity', жаль 'sorry', варто 'it is worth', слід 'one should', видно 'one can see'. They are the predicate of a subjectless clause (§9).
• Modal words (мабуть 'probably', звичайно 'of course', на щастя 'fortunately') and onomatopoeia (гав 'woof', бах 'bang'), which some authors take out of the adverbs and interjections.

The count of ten is a convention, not a discovery. Simovych's grammar of 1919 has nine: five inflected classes (nouns, pronouns, adjectives, numerals, verbs) and four uninflected (prepositions, adverbs, conjunctions, interjections); the particle has no chapter of its own, and words like не, ж are treated with adverbs and conjunctions (Simovych 1919 §§122–123). Smal-Stockyj's Ruthenische Grammatik of 1913 is arranged by inflection (nouns, adjectives, pronouns, numerals, verbs) and gathers the particles, prepositions and conjunctions in its syntax part, where §58 treats a set of particles (було, бувало, нехай, хай, най, бодай, би, не, ні) as a group (Smal-Stockyj 1913, contents and §58). The principle has not changed in a century: a class is defined first by inflection and then by syntax.

2. Three classifications side by side

Four points of difference matter in practice.

1. UD splits the pronoun. Substantival pronouns (я, хто, ніщо) are PRON; adjectival ones (мій, цей, весь, який, жодний) are DET, because they modify a noun the way English determiners do. Traditional grammar keeps them in one class, and VESUM does not have a pronoun part of speech at all: a pronoun is a noun or an adjective with the tag pron and a subtype (pers, pos, dem, int, rel, ind, neg, def, gen, emph, refl).
2. VESUM makes the adverbial participle a lemma of its own. Of the 442 458 lemma lines in the VESUM file, 70 959 are advp (написавши, пишучи), against 49 753 verb lines. UD keeps them under VERB, lemmatised to the infinitive.
3. Ordinals are adjectives everywhere except in school. Перший 'first', п'ятий 'fifth' decline exactly like новий 'new'; UD tags them ADJ with NumType=Ord (1 688 tokens in the two treebanks), VESUM adj.
4. Large numbers are nouns. Тисяча 'thousand', мільйон 'million', мільярд 'billion' decline as nouns and have a plural; UD tags them NOUN with NumType=Card (134 tokens). Indefinite quantifiers кілька 'several', багато 'many', декілька 'a few' are DET (261 tokens with NumType=Card), although багато also appears as ADV and NUM (§12).

3. How often each class occurs

UPOS distribution in the two UD Ukrainian treebanks (all three files of each, 2026-11-06):

IU mixes fiction, news, opinion, Wikipedia, legal texts, letters and comments; ParlaMint is transcribed speech of the Verkhovna Rada from 2003, 2004 and 2014. The two profiles are close. The parliament has more pronouns (ми 'we' is its commonest pronoun lemma with 1 361 tokens, against 477 in IU, where він 'he' leads with 818) and more proper nouns; IU has more interjections, particles and foreign material (X). The stable figure is the noun: about a quarter of all tokens, and with adjectives and prepositions more than four tokens in ten. Ukrainian text is nominal, and its case morphology carries much of the syntax.

The lexicon is even more nominal. VESUM's 442 458 lemma lines (a lemma with two genders or two parts of speech has two lines) divide into 191 259 noun, 108 690 adj, 70 959 advp, 49 753 verb, 19 462 adv, 1 081 noninfl, 495 intj, 232 prep, 209 part, 159 conj, 142 numr and 17 number (count of 2026-11-06). The closed classes are small: there are fewer than 250 prepositions and fewer than 200 conjunctions in the whole dictionary.

4. Nouns

A Ukrainian noun has a fixed gender (masculine, feminine, neuter; plural-only nouns have none), a number and seven cases, and is either animate or inanimate, which decides the form of the accusative (11. Nouns). The tests:

• It takes adjectives that agree with it: нов-а хат-а 'new house', нов-ий дім 'new house', нов-е вікно 'new window'.
• It can be the subject or object without a preposition and stands in a case governed by a verb or preposition: бачу хату 'I see the house' (accusative), біля хати 'by the house' (genitive).
• It answers хто? 'who?' (animate) or що? 'what?' (inanimate).

Proper nouns (UD PROPN) are capitalised and inflect like common nouns: Київ — Києва — у Києві, Тарас — Тараса — Тарасе!. Surnames in -енко are second-declension masculine for men (Шевченка) and invariable for women (з Оксаною Шевченко); surnames of adjectival form decline as adjectives (Ковальський — Ковальського, Ковальська — Ковальської).

Substantivised adjectives are adjectives used as nouns, inflecting as adjectives: черговий 'person on duty', вартовий 'sentry', знайомий 'acquaintance', військовий 'serviceman', уповноважений 'authorised representative', головуючий 'person presiding', вітальня 'living room' (once an adjective), пальне 'fuel', морозиво 'ice cream'. UD tags those with a fixed nominal meaning NOUN: військовий (42 tokens), уповноважений (26), головуючий (11, almost all in ParlaMint); the month name лютий 'February' (38) is the adjective 'fierce' turned noun. Mova's kg parser does not yet treat these: the kg README lists "substantivized adjectives (першим as the noun перше)" as a known failure.

5. Adjectives

An adjective agrees with its noun in gender (singular only), number and case: зелен-ий ліс, зелен-а трава, зелен-е поле, зелен-і луки. It has a hard and a soft declension (новий / синій), degrees of comparison (новіший, найновіший), and two subtypes that school grammar keeps inside the class:

• Possessive adjectives in -ів/-їв and -ин/-їн: батьків 'father's', мамин 'mother's', Шевченків 'Shevchenko's'. They answer чий? 'whose?'. UD tags them ADJ with Poss=Yes. They are alive in speech and in literature (мамин борщ, батькова хата) and common in place names (Васильків, Борисів, Іванівка from Іванів), and they are one of the old forms that Ukrainian keeps where the neighbouring standards replaced them by the genitive (9. Older Ukrainian and the old forms alive today).
• Short forms: a handful of masculine adjectives have a short nominative used as a predicate: повинен 'obliged' (26 tokens in UD, against the long повинний), потрібен 'needed' (9), згоден 'agreeing' (5), здатен 'capable' (4), годен 'fit, able', винен 'guilty', ладен 'ready'; in poetry and song also зелен, ясен, молод (зелен гай). UD marks them Variant=Short.

Full detail is in 15. Adjectives and adverbs.

6. Numerals

The numeral class is the least homogeneous:

The defining property of the cardinal is not its meaning but its government: a cardinal in the nominative or accusative decides the case and number of its noun (два брати, п'ять братів) and in the other cases agrees with it (з п'ятьма братами). That is why UD keeps тисяча (which governs the genitive plural like a noun, тисяча років) apart from п'ять. The rules are in 12. Pronouns, determiners and numerals §§9–14.

7. Pronouns

Pronouns point to an entity or property instead of naming it. By meaning they are personal (я, ти, він), reflexive (себе), possessive (мій, свій, їхній), demonstrative (цей, той, такий), interrogative and relative (хто, що, який, чий, котрий), definite and generalising (весь, кожен, сам, самий, інший), negative (ніхто, ніщо, жоден, ніякий) and indefinite (хтось, дехто, будь-який, казна-що). By form they are noun-like (хто, ніщо: case only) or adjective-like (мій, цей, жодний: gender, number, case, agreement).

UD's PronType counts for the two treebanks (PRON + DET):

VESUM's 491 pronoun lemma lines divide by subtype into 233 ind, 87 dem, 45 neg, 39 int, 37 rel, 30 gen, 20 def, 18 pos, 11 emph, 9 pers and 1 refl: the open end of the pronoun is the indefinite series, built productively with -сь, де-, будь-, -небудь, казна-, хтозна-, аби- (12 §8).

8. Verbs and their non-finite forms

The verb is the only class with person, tense, mood and aspect. Every verb is imperfective or perfective, and most come in aspect pairs (писати — написати, вирішувати — вирішити). Its forms fall into four groups, which taggers and dictionaries assign differently:

The impersonal forms in -но/-то (Лист написано 'The letter has been written', literally 'letter-ACC written-IMPERS') are a Ukrainian speciality among the East Slavic languages: they take an accusative object and no subject, and UD files them as finite verbs with Person=0. Their syntax is in 13. Verbs and chapter 28; the point here is that they are verbs, not short participles, although historically they come from the neuter short participle.

AUX in UD Ukrainian is бути 'be' in its auxiliary and copular uses (the future буду писати, the copula є; 1 887 tokens in the two treebanks), the iterative бувати (6) and the conditional particle би/б (303), which UD treats as an auxiliary. Modal verbs (могти 'can', мусити 'must', хотіти 'want') are VERB.

9. Adverbs and predicatives

An adverb does not inflect and modifies a verb, adjective, adverb or whole clause: швидко біжить 'runs fast', дуже гарний 'very beautiful', вдома 'at home', вчора 'yesterday', по-українськи 'in Ukrainian', навпаки 'on the contrary'. Most qualitative adverbs are made from adjectives in -о/-е (швидкий → швидко, добрий → добре) and have degrees (швидше, найшвидше) (15).

Predicatives are invariable words that serve as the predicate of a clause with no nominative subject. The person concerned is in the dative, and the clause often has an infinitive or a що-clause:

• Мені треба йти. (Meni treba jty.) 'I have to go', lit. 'to-me necessary to-go'.
• Тут можна курити. 'One may smoke here.'
• Шкода, що ти не прийшов. 'Pity that you didn't come.'
• Йому було жаль сестри. 'He felt sorry for his sister.'
• Варто спробувати. 'It's worth trying.'
• Нам слід поспішати. 'We should hurry.'
• Було видно гори. 'One could see the mountains.'

The tense is carried by бути: треба було, треба буде. VESUM marks 1 390 lemma lines with the tag predic (919 of them adverbs, 416 noninfl like треба, 40 nouns like час, 12 interjections) and records their government in comments (треба: rv_inf:rv_oru:rv_dav, that is, infinitive, instrumental, dative; можна: dative; шкода: infinitive). Most are adverbs in -о used predicatively (холодно 'it is cold', байдуже 'it doesn't matter', соромно 'it is shameful'), which VESUM tags adv:predic; треба, шкода (predicative), можна are noninfl:predic.

UD Ukrainian does not have a separate tag and files predicatives as ADV: треба 242 tokens, можна 208, потрібно 120, необхідно 51, слід 40 (and 4 as the noun слід 'trace'), варто 9, шкода 5. Жаль is ADV 3 times and NOUN 66 times (mostly the noun 'regret' in на жаль 'unfortunately'). A parser that learns from UD therefore sees треба as an adverb with no head verb, attached as the root of its clause.

Predicatives are distinct from the short adjectives повинен, потрібен, згоден (§5), which agree with a nominative subject (він повинен, вона повинна, вони повинні), and from the impersonal -но/-то verbs (§8), which govern an accusative.

10. Function words

10.1 Prepositions

A preposition governs one or more cases (there is no preposition with the nominative except in fixed patterns like піти в солдати 'become a soldier', with an old accusative identical to the nominative, 11 §6.4):

The locative is used only after a preposition: that is why its other name is "prepositional" in some traditions (Smal-Stockyj 1913 §16 already notes that the locative "is now used only with prepositions"). Prepositions with several cases distinguish meaning by case: на стіл 'onto the table' (acc.) vs на столі 'on the table' (loc.); з хати 'out of the house' (gen.) vs з братом 'with the brother' (instr.); за хату 'behind the house (motion)' vs за хатою 'behind the house (place)'.

The forms of в/у, з/із/зі/зо and між/поміж alternate for euphony (4. Connected speech); the pronouns of the third person take the prefix н- after a preposition (до нього, з нею, 12 §2).

10.2 Conjunctions

Coordinating conjunctions (UD CCONJ) join elements of equal rank: і/й/та 'and', а 'and, whereas', але/проте/однак/зате 'but', або/чи/або… або 'or', ні… ні 'neither… nor'. Subordinating conjunctions (SCONJ) introduce a clause: що 'that', щоб 'so that', бо/тому що/оскільки 'because', якщо/коли/як 'if', хоча/хоч 'although', ніби/наче/мов 'as if'. Та is both 'and' and 'but' and is also the feminine of той; і/й alternate by sound (04). Relative pronouns and adverbs (який, де, коли) also introduce clauses but keep their own class and function inside the clause.

10.3 Particles

Particles (UD PART) are invariable words that change the meaning or force of a word or clause without being a member of it:

In the treebanks the commonest particles are ж (277), лише (152), то (174), тільки (220), не (by far the most frequent, tagged PART). The particle би/б that forms the conditional (13) is a particle for school grammar and an auxiliary (AUX, 303 tokens) for UD Ukrainian. Не is PART 2 513 times.

11. Interjections

Interjections (ой, ах, ого, агов, ну, ура, цить 'hush', гей, овва) stand outside the clause and are set off by commas. Some take an object: Ой, лишенько! Some inflect for number like imperatives: цить — цитьте 'hush (sg./pl.)', нате 'here you are (pl.)', ану́те. Interjections are rare in written text (123 INTJ tokens in IU, 45 in the parliament) and frequent in dialogue, songs and folk poetry, where ой opens a line (Ой у лузі червона калина). VESUM has 495 intj lemma lines, 19 of them marked bad as non-standard.

12. Words in several classes

A form can belong to several classes; the class is decided by its syntax. UD Ukrainian 2.18 counts (both treebanks, all files, lowercased forms):

Some of these splits are annotation choices more than facts of the language. Багато in багато людей 'many people' is DET in one sentence and NUM in another; то in якщо…, то is SCONJ in some documents and PART in others. A parser trained on the treebank inherits these inconsistencies.

Homonymy across classes is also common in inflected forms: мати 'mother' and мати 'have'; три 'three' and три 'rub!' (imperative of терти); мила 'dear (fem.)', 'soap (gen.)' and 'she washed' (мити); стали 'they stood' and 'steel (gen.)'. VESUM gives all analyses; choosing among them is the tagger's task.

13. How Mova handles word classes

Mova has two separate paths for Ukrainian, and neither has a Ukrainian tag set of its own.

The English parser on UD Ukrainian. en xpos-fill (xpos_fill and xpos_of in en/src/main.rs) writes a Penn Treebank tag into the XPOS column of the treebank, computed from UPOS and FEATS, and the English tagger and parser are then trained on the result. The mapping is the one in §2: NOUN → NN/NNS by Number, ADJ → JJ, NUM → CD, DET → DT/WDT/PRP$, PRON → PRP/WP/PRP$, PART → RB, VERB/AUX → VB/VBD/VBZ/VBP/VBG/VBN by VerbForm, Tense, Person and Number. On 2026-11-06, with the binary of 2026-10-22, training on train and testing on dev, run twice with identical results:

The commonest tagger errors are number confusions (NNS → NN 148 times and NN → NNS 126 times on IU dev), then noun/adjective confusions. They show what the mapping loses: the PTB tag keeps only number for nouns and nothing for adjectives, so a Ukrainian noun's seven cases, three genders and animacy, and an adjective's agreement, are invisible to the tagger. A word in -и can be a genitive singular or a nominative plural (книжки), and the tag has to guess the plural/singular split without the case that decides it. Predicatives come out as RB (they are ADV in UD), participles as JJ, adverbial participles as VBG, the -но/-то forms as finite verbs (VBP, since Person=0 is neither third singular nor past). The FEATS column is carried along in the file but the tagger does not use it.

The kg crate on VESUM. kg/src/gram.rs packs a word-form's grammar into 32 bits; the 5-bit part-of-speech field takes the VESUM values in POS_NAMES (noun, verb, adj, adv, prep, conj, part, intj, numr, noninfl, advp, onomat, other; onomat never matches, because VESUM writes it as a second tag, noninfl:onomat). Gram::from_vesum reads a VESUM tag string field by field. What this keeps and loses, read from the code:

• Pronouns are not a part of speech but a flag (PRON, bit 23) on a noun or adjective, as in VESUM. The subtype (pers, dem, neg…) is not stored.
• Predicatives fall under adv or noninfl (whichever VESUM gives first); the predic tag itself is not stored. The parser recognises the commonest ones by list instead: CONJ in kg/src/uk.rs includes треба, потрібно, можна, слід, варто, необхідно, неможливо, and a sentence containing one of them is rejected as not a simple sentence ("Треба було піти").
• Adverbial participles have their own value advp, so пишучи is never confused with a finite verb.
• Quantity words that take the genitive are a closed list QUANT (багато, мало, кілька, декілька, чимало, скільки, трохи…), and a "quantity group" becomes a quantity:<word> edge of the fact. Cardinal numerals have no such rule; sentences with digits are rejected outright.
• Substantivised adjectives are not handled (README, "Not handled").
• The VESUM flags for indeclinable (nv), plural-only (ns), archaic (arch), non-standard (bad) and slang are not stored in Gram; the information is lost after lookup.

Neither path has a model of the ten school classes as such, and neither reads the treebank's FEATS for tagging. The kg parser covers only sentences of up to eight words without commas, conjunctions or digits, and is checked by hand on Tatoeba samples (kg/examples/gold-v1.tsv, 31 sentences: 19 judged correct, 2 half-correct, 9 wrong on 2026-09-24; two of the nine errors are the adverb ніколи 'never' read as the name Нікола, a word-class error caused by proper-name lookup). VESUM is non-commercial, so the kg path is a research and test path, not one that can ship in a release (Data and model licences).

What would be needed. A Ukrainian tag set that keeps case, gender, number, animacy and aspect (UD FEATS or the VESUM tags themselves) in the tagger's input and output; a predicative class; and a lexicon with open terms to replace VESUM in what is built (Wiktionary is CC BY-SA, so learning and tests only; the 2019 orthography is free to reuse but has no lexicon).

Sources for this chapter

• Simovych, Vasyl. 1919. Граматика української мови для самонавчання та в допомогу шкільній науці. Kyiv–Leipzig: Ukrainska nakladnia. §§120–126. Public domain. Library books/s/si/simovych-1919-hramatyka-ukrainskoi-movy-dlia (slug uk-gram-simovych-1919).
• Smal-Stockyj, Stephan. 1913. Ruthenische Grammatik. Berlin–Leipzig: Göschen. §§16, 35–37, 58. Public domain. Library books/s/st/stockyj-1913-ruthenische-grammatik (uk-gram-smal-stockyj-1913); archive.org/details/ruthenischegramm00smaluoft
• Український правопис (2019), approved by Cabinet of Ministers resolution No. 437 of 22.05.2019; morphology part §§66–120. Official document, not protected by copyright. Library books/u/uk/ukrainskyi-pravopys-2019 (uk-gram-pravopys-2019).
• Babych, Bogdan. 2019. Unsupervised induction of Ukrainian morphological paradigms for the new lexicon. CC BY 4.0 (ACL Anthology) (babych-2019-unsupervised-induction-ukrainian-morphological-paradigms): Ukrainian inflected classes have between 7 and 28 distinct feature combinations per lemma.
• Shvedova, Maria, Arsenii Lukashevskyi & Andriy Rysin. 2025. Developing a Universal Dependencies treebank for Ukrainian parliamentary speech. CC BY 4.0 (ACL Anthology) (shvedova-2025-developing-universal-dependencies-treebank-ukrainian).
• UD Ukrainian IU 2.18 (CC BY-NC-SA 4.0, counts only) and ParlaMint 2.18 (CC BY-SA 4.0), ud-treebanks-v2.18; counts of 2026-11-06.
• VESUM (dict_corp_vis.txt.bz2, release v6.8.6), BrUK project, CC BY-NC-SA 4.0: counts only.
• Leipzig Corpora Collection, ukr_news_2024 (1M sentences), CC BY (Goldhahn, Eckart & Quasthoff 2012).
• Mova: en/src/main.rs (xpos_fill, xpos_of, tag_eval, parse_eval), kg/src/gram.rs (POS_NAMES, Gram::from_vesum), kg/src/uk.rs (CONJ, QUANT), kg — the Mova knowledge graph, kg/examples/gold-v1.tsv.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/uk/10-word-classes/