🗣️ 35. Reference and coreference

A Ukrainian text keeps track of its people and things with fewer words than English and with more grammar. A noun, once introduced, is taken up by a personal pronoun that copies its grammatical gender — a table is він 'he', a book is вона 'she', a window is воно 'it'; by a zero subject, which the verb ending identifies; by a demonstrative (цей, той), which points to a less expected referent; or by the reflexives себе and свій, which point back to the subject. The verb helps unevenly: the present and the future carry person and number (читаю can only be 'I read'), while the past carries gender and number but not person (прийшла is 'I, you or she came', said of a woman). This chapter describes the referring expressions, how gender, the verb and the order of the discourse resolve them, what the UD treebanks show about overt and zero subjects (counted on 2026-11-06), the variation between the literary standard, speech and the dialects, and what Mova does with Ukrainian reference today: it records a dropped subject as a member without a word, and does nothing more.

Contents

1. Referring expressions: the inventory
2. Gender as the main cue
3. Zero subjects and the tenses
4. Tracking referents: zero, він, той
5. Reflexives: себе, -ся, свій
6. Demonstratives and це
7. Person in reported speech
8. Generic, indefinite and unknown referents
9. Abstract anaphora and the resumptive pronoun
10. Varieties and styles as deltas
11. Quick reference: reference in UD
12. How Mova handles it
13. Sources

Related chapters: the forms of the pronouns in 12. Pronouns, determiners and numerals; verb agreement and the counts of dropped subjects in 21. Clause structure; relative pronouns and the resumptive pronoun after що in 25. Relative clauses; word order and topic in 29. Ellipsis, word order and information structure; the polite Ви in 32. Politeness and hedging; connectives in 36. Discourse relations; register in 37. Register and variation.

Conventions as in chapter 20: examples in Cyrillic italics, romanized by the national system (Cabinet of Ministers resolution No. 55 of 2010) on first use, glossed in quotes; sentences from UD Ukrainian-ParlaMint (CC BY-SA 4.0) quoted with their sent_id; UD Ukrainian-IU (CC BY-NC-SA 4.0) used for counts only. Counts: UD 2.18, all splits, awk over the CoNLL-U files, 2026-11-06; IU has 99,495 and ParlaMint 89,858 syntactic words without punctuation.

1. Referring expressions: the inventory

Two properties set the Ukrainian system apart from the English one. First, there is no gender-neutral pronoun for things: every noun has a gender and is resumed by the pronoun of that gender, so він and вона are as often things as people. Second, the subject pronoun is optional wherever the verb identifies the subject, and the absence of a pronoun is itself a referential signal (§ 4).

The UD treebanks give the overall shape. Personal pronouns of the third person as subjects (nsubj), 2026-11-06:

Written prose (IU: fiction, news, letters, posts) refers to third persons about twice as often as parliamentary speech, where the first and second persons dominate (я, ми, ви — chapter 21, § 3). Воно is the rarest: it resumes neuter nouns (вікно, питання, дитя), and in IU 4 of its 94 tokens are expletive (expl).

2. Gender as the main cue

2.1 Grammatical gender decides the pronoun

A pronoun takes the gender and number of its antecedent noun, whatever the noun denotes:

Gender is therefore a strong filter for coreference — stronger than in English, where it covers every thing. A sentence such as Чоловік не міг підняти сина, бо він був дуже важкий 'The man could not lift his son because he was very heavy' does not have this advantage: both nouns are masculine. Kuchmiichuk (2023, kuchmiichuk-2023-silver-data-coreference-resolution-ukrainian), translating the 285 Winograd schemas into Ukrainian, found that the opposite problem arises too: where English separates a person from a thing by he versus it, Ukrainian may use він for both, creating ambiguities the English had not; 22 of the 285 schemas could not be translated with their ambiguity intact, leaving 263.

2.2 Natural gender against grammatical gender

When a masculine noun names a woman, agreement and reference may follow the person rather than the noun:

Feminitives, increasingly used since the 2010s and codified by the 2019 orthography, remove most of these conflicts: директорка, міністерка, депутатка, речниця resume with вона and agree in the feminine. Synchak and Starko (2022, synchak-2022-feminine-dictionaries) describe how they are collected from the GRAC corpus into a dedicated web dictionary of feminine personal nouns (2022). For reference this means that a feminine past tense next to a masculine title (Міністр заявила) is a reliable clue that the referent is a woman; in the Leipzig news corpus of 2024 such mixed agreement is rare (one міністр зазначила in 1 M sentences), because news writers now use the feminitive.

2.3 Neuter gender for the familiar, the unknown and the disparaged

Smerechynskyi (1932, smerechynskyi-1932-narysy-z-ukrainskoi-syntaksy, ch. V) describes a use of the neuter that has no English counterpart. Folk speech refers with воно and neuter verb forms to a person whose identity or sex does not matter or is unknown, and, after Sulyma, to a person spoken of with familiar, slightly condescending tenderness by older people (Воно ще мале, нічого не тямить 'It (the child) is still small, it understands nothing'). His example of the unknown thief: Хто ж це в мене їх підчистив. Воно й бере, та не заразом, а потрошку… — 'Who has been thinning them out on me? It takes them, not all at once but bit by bit…' (a folk text cited by Smerechynskyi). The indefinite щось does the same work when nothing at all is known of the agent: Щось прийшло і черешні обірвало 'Something came and stripped the cherries'. The neuter is thus a referential device for low identifiability, not only a class of nouns.

3. Zero subjects and the tenses

3.1 What the verb tells

The past tense is the weak point. Прийшла додому пізно is complete and grammatical, but whether the woman is the speaker, the addressee or a third person is decided only by the preceding discourse. The present has the opposite profile: the first and second persons are unambiguous without a pronoun, the third person is ambiguous among all third-person referents of the right number.

3.2 How often the subject is dropped

The counts below are finite indicative verbs that are not conj (a conjunct shares the subject of the first verb) and not impersonal (Person=0), with no nsubj or csubj (UD 2.18, 2026-11-06; the method and totals of chapter 21, § 3.1, here split by the type of subject and, in the past, by gender):

("Other subject" is a noun, a non-personal pronoun or a clause. ParlaMint's past totals here include a few AUX roots and so differ by five from chapter 21.)

Three things stand out:

• The first person drops its pronoun most, because the ending is unique: three in five in writing, nearly half in parliamentary speech, where я is often kept as a formula of public speech (Я прошу…, Я вважаю…).
• The past with a person keeps the pronoun. Of the past-tense verbs with an overt personal-pronoun subject in IU, 380 are first person, 53 second and 441 third. In the masculine and feminine past the pronoun outnumbers the zero subject (410 vs 278, 170 vs 162); in the present first person the zero outnumbers the pronoun (222 vs 130). The pronoun does in the past what the ending does in the present.
• The neuter past is mostly subjectless. Its 43% zero subjects in IU are largely not dropped pronouns but impersonal verbs: вдалося 'one managed', сталося 'it happened', було 'there was', стемніло 'it got dark' (chapter 21, § 5).

Fellerer (2022, fellerer-2022-pro-drop-sw), counting by hand the clauses where a pronoun could compete with its absence, found the same contrast in standard Ukrainian memoir prose — subject pro-drop in 60% of present-tense clauses and 20% of past-tense clauses —. The treebank counts agree in direction and roughly in size.

3.3 Where the zero subject sits

Zero subjects of the past are more frequent in dependent clauses, which continue the subject of their main clause, than in main clauses (IU / ParlaMint, past finite verbs, 2026-11-06):

Of the IU adverbial and complement clauses in the past without a subject, 78 have a verb whose gender and number match the subject of the main clause (consistent with "same subject"), 47 do not (a different referent recoverable from context, or an impersonal neuter), and in 33 the main clause has no determinable subject (ParlaMint: 19, 13, 9). The same-subject reading is the default, not a rule. Kuchmiichuk (2023) makes the point from the other side: in a fluent Ukrainian version of The man couldn't lift his son because he was so weak, the subordinate clause has no subject at all (…бо був такий слабкий), and the zero is read as the main-clause subject, the man.

4. Tracking referents: zero, він, той

Ukrainian has a three-step scale of referential expressions inside a stretch of discourse:

The second line is ambiguous because він can resume either masculine noun; the third is not, because the demonstrative той points away from the topic (Petro) to the other referent. English needs the latter, the other or a repeated noun. Kuchmiichuk (2023) gives the Winograd example in this form: Чоловік не міг підняти сина, бо той був такий важкий — with той, the heavy one is clearly the son. The demonstrative is "grammatically correct and fluent" in Ukrainian where English allows only the personal pronoun.

Folk narrative builds its relative links in the same way. Smerechynskyi (1932, ch. XXIII and XXIX, II) shows that the popular language prefers a paratactic relative link with а or і and the demonstrative or personal pronoun to a relative clause with котрий: Прийшов до моря, а на морі рибаки рибу ловили 'He came to the sea, and on the sea fishermen were fishing'; І там через море лежить золотий міст, а на тому мості різні дерева стоять 'And there across the sea lies a golden bridge, and on that bridge stand various trees'; Ой у полі билина, її вітер колише 'Oh, in the field a blade of grass, the wind sways it' — where the second clause resumes the noun with той or with the personal pronoun in the needed case, without a relative word. Chapter 25 treats these as alternatives to the relative clause; for reference they show that той + the repeated noun (на тому мості) is a normal anaphor, not a stylistic accident.

The intensifying і той after a noun is another referential construction of the standard: Навіть батько і той злякався 'Even father — even he — got scared' (Dovzhenko, quoted in the 2019 orthography, § 158, which says it takes no comma). The demonstrative here repeats the subject to stress it, as a resumptive pronoun does.

5. Reflexives: себе, -ся, свій

5.1 Себе and -ся

The reflexive pronoun себе has no nominative and serves every person and number: Я бачу себе, Ти бачиш себе, Вони бачать себе. It refers to the subject of its clause, or, in an infinitive or nominal phrase, to the implied subject:

Its clitic counterpart -ся (a verbal suffix since the old enclitic ся fused with the verb; still mobile in the south-western dialects, § 10) marks reflexive, reciprocal, passive, anticausative and impersonal readings (chapter 28). Only in the reflexive reading (митися 'wash oneself') is -ся a referring element; себе is preferred when the object is stressed or contrasted: Він хвалить себе, а не інших 'He praises himself, not others'.

Counts (lemma себе, all relations): IU 172 (obl 67, obj 36, iobj 36, expl 13), ParlaMint 69 (obl 30, obj 18, iobj 15). The expl uses in IU are fixed expressions with собі that do not refer: Жили собі дід та баба 'There lived an old man and an old woman', the "ethical dative" of tales.

5.2 Свій in all three persons

Свій means 'one's own' and refers to the subject of its clause. Unlike English his/her/their, it is not tied to the third person: Simovych (1918, simovych-1918-praktychna-hramatyka-ukrainskoi-movy) notes that it "properly refers to the third person … but we use it in all three persons", and gives Орю свій переліг, убогу ниву (Shevchenko, 1st person), Своєї недолі й конем не об'їдеш (proverb, 2nd person) and Свою рідну Україну на віки покинув (song, 3rd person). With a first- or second-person subject the plain possessive is also correct: Я люблю свою матір and Я люблю мою матір are both standard, the first more usual and the second slightly more emphatic or personal.

With a third-person subject the choice changes the meaning:

The treebanks show how often the choice arises. Possessive determiners (det) by the person of the nearest governing verb's subject (from an explicit pronoun subject or, failing that, the verb's ending; a noun subject counts as 3rd person; a heuristic count, 2026-11-06):

The cells with a first-person verb mix the choice point (Я люблю свою / мою матір) with possessives that are not bound by the subject (Я бачив твого брата); the comparison is still telling. Written Ukrainian uses свій and мій/наш about equally with a first-person verb (27 vs 25); parliamentary speech prefers наш four to one (62 vs 15): наша держава, наш комітет, where the speaker speaks for a group and свій would narrow the reference to the subject alone. With infinitives (захищати свою державу) свій is frequent because it refers to the implied subject, which no possessive pronoun could name. Overall свій (IU 355, ParlaMint 226) is about as frequent as all third-person possessives together in IU (364) and twice as frequent in ParlaMint (110).

Свій also has lexical uses that are not anaphoric: свої 'our people, one's own folk' (Не стріляйте, тут свої! 'Don't shoot, we're on your side'), своя людина 'one of us', у свій час 'in due time'.

6. Demonstratives and це

6.1 Цей and той

ParlaMint's 881 цей determiners against 236 той reflect the speech of debate: цей законопроект, ця поправка, це питання point to what lies before the house. In written narrative той is relatively more frequent (IU 291 vs 457) because it tracks referents (§ 4) and introduces correlatives.

The correlative той, хто / те, що 'he who, that which' is the usual Ukrainian free relative with a head (chapter 25): Той, хто прийшов, мовчав; Те, що ти сказав, правда. In the treebanks той heads a relative or clausal modifier introduced by хто 24 times in IU and 53 times in ParlaMint (acl 33 + acl:relcl 20), and one introduced by що 22 times in IU and 3 times in ParlaMint.

6.2 Це as subject and pointer

Це is the most frequent subject pronoun of Ukrainian after the personal pronouns: nsubj 232 in IU and 418 in ParlaMint, with 89 / 104 expl uses (Це я винен 'It is I who am to blame'; chapter 29 on це in clefts). It refers to an entity (Це мій брат 'This is my brother'), to the situation (Це було давно 'That was long ago'), or to a proposition (Це правда 'That's true'). Unlike воно, це does not agree with the noun it presents: Це мій брат, Це моя сестра, Це мої друзі. The colloquial то (То правда; То мій брат) is its distal or western counterpart.

7. Person in reported speech

The standard reports speech with a shift of person, as English does: Вона сказала, що вона знає 'She said that she knew'. Smerechynskyi (1932, ch. XVII) argued that the folk language knows essentially one way of reporting speech: it keeps the forms of the direct speech and adds only the introducing що: А вона почала казати, що я знаю поганеньку 'And she began to say that I [= she] know a bad little one' (a folk text cited there); От, інші кажуть, що «даймо», другі — «ні» 'Well, some say "let's give", others "no"'. The я of the subordinate clause is the reported speaker, not the narrator. He contrasted this with the schoolbook rule (shift to the third person) and saw the що as working "like a colon in writing".

For reference resolution this means that in folk narrative, dialect texts and much colloquial speech, a first-person pronoun after каже, що may refer to the subject of казати. In the standard written language the shifted form is the norm and the unshifted one reads as free indirect or colloquial speech.

Simovych (1919, § 347) lists the inserted каже, кажуть, мовляв among the parenthetical words that mark "other people's thoughts": Мала, каже, нехай дожду (Shevchenko).

8. Generic, indefinite and unknown referents

The indefinite third plural is a zero subject without an antecedent: the reader must not link it to the last plural noun. Пішли до лікаря, а там кажуть: приходьте завтра 'We went to the doctor's, and there they say: come tomorrow' — кажуть is the staff, not we.

9. Abstract anaphora and the resumptive pronoun

Abstract anaphora. Propositions, events and facts are taken up by це, то, так, таке, цим, з цим, через це, тому: Він запізнився. Це всіх розсердило 'He was late. That made everyone angry'; Через це ми не встигли 'Because of that we didn't make it'. The connective use of these anaphors is the subject of chapter 36 (тому, через це, у зв'язку з цим).

The resumptive pronoun after a dislocated subject. In speech a noun phrase is often fronted as the topic and resumed by a pronoun in the clause: Такі закони, вони є не, знаєте, неґрунтовні і… (ParlaMint-UA_2022-01-25-m0.u262.p1.lang1.s6) 'Such laws, they are not, you know, ungrounded and…'. ParlaMint marks such topics with dislocated 49 times (IU 19). The standard written norm avoids the double subject; in speech it is the normal way to introduce a new topic before saying something about it (chapter 29). A related resumptive appears inside relative clauses with що (хлопець, що я його бачив 'the boy that I saw him'; chapter 25), where the pronoun carries the case that the invariable що cannot.

10. Varieties and styles as deltas

Each variety or style is listed only by what it changes in the reference system of the literary standard.

Fellerer (2022) argues that the higher pro-drop of the Dnister dialect is not simply caused by the clitics, which are themselves minority forms, but was reinforced by long contact with South-Eastern "Borderland" Polish, a pro-drop variety with similar clitics; the dialect's drop extends even to bare past forms where the clitic gives no help, as in his example of a train referred back to across two clauses with other subjects.

11. Quick reference: reference in UD

UD itself does not annotate coreference. Kuchmiichuk (2023) knew of one public Ukrainian coreference dataset, Kramov's Coreferent Clusters (2021; 2,528 documents, 361,534 tokens, of unclear annotation method), and added two silver sets: a manually translated Winograd set of 263 schemas and a machine-translated OntoNotes (3,493 documents, 94,269 sentences, 1.46 M tokens, 201,700 mentions) with the clusters projected through the attention of the translation model, which the author describes as noisy where the translation is poor.

12. How Mova handles it

Mova reads Ukrainian only through kg/src/uk.rs, the conservative extractor of facts from simple sentences (chapter 20, § 12; kg — the Mova knowledge graph). For reference it has exactly one device:

• The implicit subject. When no nominative agrees with the predicate, fn verbal adds a subject member with no word, the lemma ø|noun and the grammar Gram::implicit_subject(pred) (kg/src/gram.rs): part of speech noun, nominative, and the number, gender and person copied from the predicate, with the flags PRON and IMPLICIT. For a past-tense verb the person field stays empty, because VESUM gives the past no person; for a present verb the gender stays empty. The implicit subject is therefore exactly as informative as the verb ending, which is linguistically right — and it is never resolved to an antecedent.
• No links between facts. Each fact points to its source sentence (from_sentence); nothing links a pronoun or an implicit subject to a noun of another fact. Gender agreement is checked only inside a clause (fn agrees, fn adj_agrees).
• Possessives as attributes. свій, його, мій become attribute:<relation> members of their noun, with their own lemma; the binding difference between свій and його is not represented.

Measured on 2026-11-06 (kg parse, CPU, worktree at commit ac9a23149, VESUM dict_corp_vis.txt.bz2, an empty list of names):

The last two rows show the limit: the extractor has no sentence splitter of its own (the Tatoeba input is one sentence per line) and no anaphora; a pronoun is a word with its own lemma.

The coref crate (coref — SLM coreference: a sieve of rules with a reason for every link) is Mova's deterministic coreference sieve. Its mention detection works from UPOS and UD relations, and its agreement features come from UD FEATS, so on Ukrainian gold trees it would find mentions and read their gender. But its pronoun table (pron_class in coref/src/lexicon.rs) maps English forms (he, she, it …), its speech verbs are English (SAY: say, tell, reply …), its gendered nouns are English, and it is evaluated on English GUM. It has not been run on Ukrainian, and it has no notion of a zero subject, of той as a topic-shifting anaphor, or of свій. The pronoun resolver inside the frame reader (prag/src/frames.rs) reads English trees only. Mova has no Ukrainian UD parser (no model in mova-parse; langseg is Latin-script only), so no Ukrainian text reaches either resolver.

12.1 Not yet

• No resolution of implicit subjects or pronouns to antecedents; no cross-sentence links in kg.
• No sentence splitting in kg parse; capitalised pronouns after a full stop may be read as names.
• No binding: свій vs його, себе, той vs він are not distinguished in meaning.
• No honorific plural, no feminine agreement with masculine titles: both are rejected as non-agreement.
• No Ukrainian coreference data in Mova; Kuchmiichuk's silver sets and the Winograd translation are not loaded or measured.
• coref and the frame reader resolve English only.

13. Sources

• Smerechynskyi, Serhii. 1932. Нариси з української синтакси (у зв'язку з фразеологією та стилістикою). Kharkiv: Radianska shkola. Ch. V (the neuter for the unknown, familiar or disparaged person; воно, щось), ch. XVII (folk indirect speech in the forms of direct speech), ch. XXIII and XXIX, II (paratactic relative links with той and the personal pronoun). smerechynskyi-1932-narysy-z-ukrainskoi-syntaksy (public domain in UA/EU)
• Simovych, Vasyl. 1918. Практична граматика української мови. Section on possessive and demonstrative pronouns (свій in all three persons; сей, ся, се). simovych-1918-praktychna-hramatyka-ukrainskoi-movy (public domain)
• Simovych, Vasyl. 1919. Граматика української мови для самонавчання та в допомогу шкільній науці. § 347 (parenthetical words: каже, кажуть, мовляв). simovych-1919-hramatyka-ukrainskoi-movy-dlia (public domain)
• Український правопис. 2019. § 32.4 (feminine suffixes), § 158 (the intensifying і той, no comma). ukrainskyi-pravopys-2019 (official text, not under copyright)
• Fellerer, Jan. 2022. Subject pro-drop and past-tense auxiliary clitics in South-Western Ukrainian. Russian Linguistics 46(3): 201–216. fellerer-2022-pro-drop-sw (CC BY 4.0)
• Kuchmiichuk, Pavlo. 2023. Silver Data for Coreference Resolution in Ukrainian: Translation, Alignment, and Projection. UNLP 2023. kuchmiichuk-2023-silver-data-coreference-resolution-ukrainian (CC BY 4.0)
• Synchak, Olena and Vasyl Starko. 2022. Ukrainian Feminine Personal Nouns in Online Dictionaries and Corpora. COLINS 2022. synchak-2022-feminine-dictionaries (CC BY 4.0)
• Hentschel, Gerd. 2023. Ukrainian and Russian in the lexicon of Ukrainian Suržyk: reduced variation and stabilisation in central Ukraine and on the Black Sea coast. Russian Linguistics 48(1). hentschel-2023-surzhyk-lexicon (CC BY 4.0)
• Universal Dependencies 2.18: UD_Ukrainian-IU (CC BY-NC-SA 4.0; counts only), UD_Ukrainian-ParlaMint (CC BY-SA 4.0; quoted by sent_id); counted 2026-11-06, all splits, with awk.
• Leipzig Corpora Collection, ukr_news_2024 (1 M sentences; Goldhahn, Eckart & Quasthoff 2012; CC BY); counted with grep 2026-11-06.
• Mova: kg/src/uk.rs (fn verbal, fn agrees, fn adj_agrees), kg/src/gram.rs (Gram::implicit_subject, IMPLICIT), kg — the Mova knowledge graph, coref — SLM coreference: a sieve of rules with a reason for every link, coref/src/lexicon.rs, prag/src/frames.rs, mova-parse.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/uk/35-reference-and-coreference/