home edit page issue tracker

UD for Latvian

Tokenization and Word Segmentation

In general, words are delimited by whitespace characters and punctuation is separated. Description of exceptions follows:

• A whitespace separating digits in a large number is not treated as a word separator. For example, 1 000 000 (“1,000,000” by English rules) is one token.
• Abbreviations without spaces are treated as single words and may contain punctuation (utt. “etc.”). In following cases we treat abbreviation as a single token even if whitespace is used between part of abbreviation and punctuation mark: u.t.jpr., u.c., u.tml., v.tml., u.t.t., N.B., P.S. and P.P.S.
• Double surnames such as Vīķe-Freiberga and words abbreviated with dashes such as e-pasts “e-mail”, k-dze “Ms.” are tokenized as a single token.
• In Latvian ordinal numerals are written with punctuation mark without whitespace like abbreviations (1.), so we tokenize ordinal numeral together with punctuation mark as one token.
• Multiple dots (… and .. ) are considered as one token. Multiple ?! are considered one token, ?!… is considered to be two tokens (?! and …).

Paragraph borders from the original text is indicated by comment line # newpar in cases when paragraph borders aligns sentence borders and MISC value NewPar=Yes for the token following mid-sentence paragraph break. MISC value SpaceAfter=No is used to note tokens lacking any whitespace after.

Morphology

Tags

Latvian uses all 17 universal POS categories.

Particles

PART tag is used for following function words: acīmredzot, ak, ar, arī, arīdzan, da, diemžēl, diez, diezin, droši, gan, i, ij, ik, ir, it, itin, ja, jau, jā, jel, jo, kaut, kā, lai, laikam, mjā, ne, nea, nebūt, nez, nezin, nē, nu, nudien, nujā, nū, nūja, nūjā, pat, patiesi, patiešām, protams, proti, taču, tad, tak, tā, tāpat, tātad, tiešām, tik, tikai, tikpat, tipa, tomēr, turklāt, vai, varbūt, vēl, vien, vienīgi, vis.

Particles can be homonymous with other POS, most notably, conjunctions CCONJ and SCONJ, interjections INTJ, and adverbs ADV, correct POS is assigned based on sentence context.

Pronouns and Determiners

Effectively distinguishing PRON and DET categories in Latvian is very hard as words used as DET can also be used as PRON, and, thus, traditional Latvian grammar does not define determiners as a distinct POS. Since version 2.15 pronoun (PRON) vs. determiner (DET) distinction is done by lemma (similarly as is done with PDT). In earlyer versions distinction was made based on tree structure.

Currently DET are: abas, abi, cikais, cikas, ciki, cita, cits, daudzi, daža, dažs, ikkatra, ikkatrs, ikkura, ikkurš, ikviena, ikviens, jebkāda, jebkāds, jebkura, jebkurš, jelkāda, jelkāds, jūsējs, kāda, kādā, kādais, kāds, katra, katrs, kura, kurā, kurais, kurs, kurš, manējs, mana, mans, mūsējs, nekāda, nekādā, nekādais, nekāds, neviena, neviens, pate, pati, pats, savējs, sava, savs, šāda, šāds, šī, šis, šitāda, šitāds, šitaids, šitejāda, šitejāds, šitā, šitais, šitas, šitentāda, šitentāds, šitentas, štā, štas, štis, tāda, tāds, tā, tas, taste, tāte, tavējs, tava, tavs, vairāki, vēlviena, vēlviens, vienotra, vienotrs, viņējs, viņā, viņais, visa, viss.

PRON are: daudzkas, es (“I”), jebkas, jelkas, jis, jūs (“you”, plural), kas, mēs (“we”), nekas, nezinkas, sevis, tu (“you”, singular), viņa (“she”), viņš (“he”), viš (“they/he/she”).

Syntax role det is used for Latvian pronoun category, which modify nouns in the sentence and agree with this noun in gender, number and case. Pronominal quantifiers daudzi “many” and vairāki “several” , and personal possessives manējais, tavējais, mūsējais, jūsējais, viņējais are DET, however in Latvian grammar they are described as adjectives.

Auxiliary Verbs

Latvian has three auxiliary verbs AUX: būt “to be”, tikt “to get”, and tapt “to become” (obsolete). The auxiliary verb is used in several types of constructions: * Analytic word forms of verbs (būt, tikt). * The copula in non-verbal predicates (būt). * The copula in infinitive predicates (būt).

Būt, tikt and tapt may still occur as normal VERB if they are used in purely existential sentences or indicate location. Verbs with modal meaning are not considered auxiliary in Latvian.

Deverbal Nouns, Participles, Coverbs

Deverbal nouns with endings -šana, -šanās (skriešana “running”) are tagged as NOUN. Most converbs with endings -ot, -oties, -am, -ām, -amies, -āmies, -dams, dama, -damies, -damās are tagged as VERB or AUX. Most adjectival participles (redzams, aizgājis, negaidīts, velkošs) are tagged as VERB. Exceptions are lexicalized uses with separate meaning, like protams “of course”, acīmredzot “obvious”, which are tagged as PART, and iespējams “possible”, which is tagged as ADJ.

Features

Nominal Features

• Nominal words (NOUN, PROPN and PRON) have an inherent Gender feature with one of two values: Masc or Fem.
• The following parts of speech inflect for Gender as they must agree with nouns: ADJ, DET, NUM, VERB, AUX. For verbs (including auxiliaries), only participles inflect for Gender. Finite verbs don’t.
• The two main values of the Number feature are Sing and Plur. The following parts of speech inflect for number: NOUN, PROPN, PRON, ADJ, DET, VERBand AUX (finite, participles and verbal nouns), marginally NUM. Selected nouns are plurale tantum Ptan or singulare tantum Coll.
• Case has 6 possible values: Nom, Gen, Dat, Acc, Loc, Voc. It occurs with the nominal words, i.e., NOUN, PROPN, PRON, ADJ, DET, NUM, VERB and AUX (participles and verbal nouns).
• Definite has 2 possible values: Ind and Def. The following parts of speech inflect for definitnes: ADJ, VERB and AUX (participles).

Degree and Polarity

• Degree applies to adjectives (ADJ), adverbs (ADV), and some participles (VERB, AUX), and has one of three possible values: Pos, Cmp, Sup.
• Polarity has two values, Pos and Neg, and applies to verbs (VERB, AUX).
  • Words ne, nē “no” occurs as independent negation particles (PART) and are marked with Polarity=Neg.
  • Occasionaly ne occurs as a part of correlative conjunction and is marked with Polarity=Neg.
  • Word jā occurs as an independent affirmation particle (PART) and is marked with Polarity=Pos.
  • The Polarity feature is not used with pronouns and determiners, although there is a subset of pronouns and determiners which are considered to be negated traditionally. The PronType=Neg feature is used there instead.

Verbal Features

• There are five main (de)verbal form types, distinguished by the UPOS tag and the value of the VerbForm feature:
  • Infinitive Inf, tagged VERB or AUX.
  • Finite verb Fin, tagged VERB or AUX.
  • Participle Part, tagged VERB or AUX.
  • Converb Conv, tagged VERB or AUX.
  • Verbal noun Vnoun, tagged NOUN.
• Aspect applies only to part of participles (VERB, AUX) and is either imperfective Imp or perfective Perf.
• Finite verbs always have one of five values of Mood: Ind, Imp, Cnd, Qot or Nec.
• Tense is used for verbs and participles:
  • Verbs in the indicative mood always have one of three Tense values: Past, Pres or Fut.
  • Infinitive, imperative, conditional, quotative, and necessitative forms do not have the Tense feature.
  • The Tense feature is also used to distinguish declinable participles (tagged VERB or AUX) into two groups: present participles (ziedošs “[it is] flowering” and lasāms “[it is] readable”) and past participles (darījis “[he has] been doing” and pateikts “[it has] been said”).
• There are two values used for the [Voice() feature: Act and Pass:
  • Passive participles (lasāms “[it is] readable” and pateikts “[it has] been said”) has Voice=Pass.
  • Finite verb forms and active participles (ziedošs “[it is] flowering” and darījis “[he has] been doing”) have Voice=Act.
• Evident applies to finite verb forms (VERB, AUX) and depends on value of Mood: quotatives have value Nfh, but indicative have value Fh.

Pronouns, Determiners, Quantifiers

• PronType is used with pronouns PRON, determiners DET and pronominal adverbs ADV with 8 permissible values: Prs, Rcp, Int, Rel, Dem, Tot, Neg, Ind.
• NumType is used with numerals (also cardinal numbers) NUM, ordinal numbers ADJ, and some adverbs ADV:
  • Numerals and ordinal numbers has one of three possible values: Card, Ord or Frac.
  • Adverbs vienreiz “once”, divreiz “twice”, trīsreiz “thrice”, četrreiz, piecreiz, sešreiz, septiņreiz, astoņreiz, deviņreiz, desmitreiz “ten times”, pusotrreiz “one and a half times” has NumType=Mult.
• The Poss feature marks possessive personal pronouns and determiners (e.g., mans “my”) and possessive adjectives (e.g., tavējais “yours”) with value Yes.
• The Reflex feature marks reflexive pronoun sevis.
  • Reflexivity is also marked on reflexive verbs and participles (VERB, e.g., mazgāties, pusapģērbusies).
• Person is marked for pronouns and finite verbs and has three values: 1, 2 and 3.
  • It is a lexical feature of personal pronouns PRON like es “I”, tu “you” (singular), viņš “he”, viņa “she”, mēs “we”, jūs “you” (plural), viņi “they” (plural, masculine), viņas “they” (plural, feminine).
  • It is a lexical feature of personal possessives DET/PRON mans, manējais “my/mine”, tavs, tavējais “your/yours” (singular), mūsējais “our/ours”, jūsējais “your/yours” (plural), viņējais “his/hers/theirs”.
  • Person is also marked on some demonstrative pronouns with value 3.
  • As a cross-reference to subject, person is also marked on finite verbs (VERB, AUX).
• Foreign is annotated Yes for foreign words X.
• Abbr is annotated Yes for abbreviations, which can be nouns NOUN (DJ), PROPN (NATO), ADJ (god. “honored”), VERB (skat. “see”), ADV (v.j.l. “above sea level”), SYM (utt. “etc.”).

ExtPos

ExtPos is currently used for annotating fixed constructions. See ExtPos for Latvian for currently used values and examples.

Unused Features

Features not applicable for Latvian:

• Animacy,
• Clusivity,
• NounClass,
• Polite.

Syntax

Core Arguments

• Nominal subject (nsubj) is a noun phrase usually in the nominative case. However:
  • If the noun phrase is quantified, it may be in the genitive, which is required by the quantifier.
  • With predicates nebūt, trūkt, pietikt, netrūkt, nepietikt noun phrase can be in genitive.
  • A finite subordinate clause may serve as the subject and is labeled csubj.
  • The noun phrase may be in the dative, if the predicate is in the necessitative mood (man jāskatās spēle “I have to watch the game”) or if the predicate is with modal meaning and has subordinated infinitive (viņam vajadzētu pasteigties “he should hurry”).
• Objects as defined in the Latvian grammar may be either bare noun phrases in accusative, dative, or genitive, or prepositional phrases in accusative, dative, genitive. All objects are labeled as obj or iobj.
  • However, if the predicate is in the necessitative mood, object may be in nominative (zēnam jāuzraksta mājasdarbs “the boy has to write a homework.”), and it is labeled as obj.
  • Accusative objects are considered obj.
  • Objects in dative and genitive cases and prepositional objects are considered iobj.

Non-verbal Clauses

The copula verb būt “be” is used in equational and attributional nonverbal clauses. Purely existential clauses (also indicating location) use būt as well, but it is treated as the head of the clause and tagged VERB.

Relations Overview

The following relation subtypes are used in Latvian:

• nsubj:pass for nominal subjects of passive verbs
• csubj:pass for clausal subjects of passive verbs
• aux:pass for passive auxiliaries
• flat:foreign for non-first words in quoted foreign phrases
• flat:name for exocentric complex name
• advmod:neg for negative particles
• advmod:emph for emphasizing particles

The following relation types are not used for Latvian: clf, dislocated, list, reparandum. However, reparandum should be introduced in future, as appropriate speech texts are annotated.

Annotating Textual Errors

Following MISC values can be used to annotate errors in the source text interfering with treebank annotation:

• CorrectionType=Spelling for typos (FORM is given as in text, while LEMMA, UPOS, XPOS and FEATS as for word without the error)
• CorrectionType=Spacing for missing or unnecessary whitespaces
• CorrectionType=InsertedPunctAfter for cases when there is missing punctuation mark (usually comma) after this token
• CorrectionType=RemovedPunctuation for unnecessary punctuation (usually comma)
• In case of CorrectionType=Spelling additional feature CorrectedForm=… gives the corrected form.

Treebanks

There are 2 Latvian UD treebanks:

• Latvian-LVTB
• Latvian-Cairo

© 2014–2024 Universal Dependencies contributors. Site powered by Annodoc and brat

.