UD for Slovenian 
Tokenization and Word Segmentation
In Slovenian UD treebanks, tokens are generally delimited by whitespace or punctuation, with the following exceptions being treated as single tokens:
- words with an apostrophe, e.g. O’Brian
- numerical expressions, e.g. 30:00, 200.000,000
- words with hyphens before suffixes, e.g. “OZN-ovski* (“UN-like”), a-ju (“to the letter a”), 15-i (“the 15th”)
- abbreviations without whitespace, e.g. dr.
Multiword tokens are not used in Slovenian UD treebanks. This means that fused words, such as bound pronouns (e.g. name “on me”), multi-word abbreviations (e.g. npr. “for example”) and colloquial contractions (e.g. nauš “won’t”), are currently treated as single tokens.
For more details, see the Slovenian tokenization description page.
Morphology
Tags
- Slovenian uses all 17 universal part-of-speech tags.
- With a few exceptions, the UD POS categorization principles are similar to the original JOS annotation scheme for Slovenian.
- As a JOS legacy, the PART tag is used for more than 60 different words annotated as JOS particles in the original corpora. Therefore, words, such as ne “no”, are currently always tagged as PART regardless whether they are used as modal particles (Ne motiš se. “You are not wrong”) or as syntactically independent feedback particles (Ne, motiš se. “No, you are wrong.”).
- The AUX tag is only used for verb biti (be), when used as an auxiliary or a copula verb.
- The DET tag is used for JOS pronouns and adverbs that predominantly occurr as nominal premodifiers, regardless of their actual function in context, e.g. moj “my”, tak “such”, ta “this”; dosti “enough”, manj “less”, veliko “many”, etc.
- Most (de)verbal forms are annotated as VERB or AUX, with the exception of adverbial participles (ADV, e.g. rekoč, hote, upoštevaje), adjectival participles (ADJ, e.g. prepričan, sprejet, ostal), and gerunds (NOUN, e.g. sedenje, igranje, plavanje).
For more details, see the language-specific guidelines for individual tags with a detailed explanation of the JOS-to-UD conversion rules, and the Slovenian UD treebanks overview with statistical details on the tagset distribution.
Features
- The Slovenian UD tagset includes all features from the universal tagset except for
Voice,Typo,NounClass,Evident,PoliteandClusivity. In addition to that, the set of universal features has been extended with four additional features to either describe language-specific features (such as Gender[psor], Number[psor] and Variant) or preserve some finer-grained morphological information encoded in the original ssj500k treebank (such as NumForm). - Nouns have inherent Gender (feminine, masculine and neutral) and inflect for Number (singular, dual or plural), Case (nominative, genitive, dative, accusative, locative, instrumental), Animacy and Definite (indefinite or definite).
- Verbs have inherent Aspect. Non-finite forms include infinitives, supine forms and l-participles. Finite forms inflect for Gender (feminine, masculine and neutral), Number (singular, dual or plural), Mood (indicative and imperative) and Person (first, second and third). The verb biti (be) also inflects for conditional Mood and future Tense, while the verbs biti (be), imeti (have) and hoteti (want) also inflect for Polarity (positive and negative).
For more details on all other morphological features see the language-specific guidelines for individual features with a detailed explanation of the JOS-to-UD conversion rules, and the Slovenian UD treebanks overview with statistical details on the tagset distribution.
Syntax
- Nominal subjects (nsubj) are nominal phrases typically occurring in nominative or (negated) genitive case.
- Nominal objects (obj) are non-adpositional predicate arguments in all other cases, regardless of semantic role.
- In case of two objects in a sentence, one is considered to be indirect object (iobj), typically the phrase in dative case.
- All prepositional phrases are annotated as oblique arguments (obl), regardless of semantic role.
- The copula (cop) label is used for verb biti (be) in all non-existential uses. The only exception are biti-constructions with prepositional phrases, which are currently always labelled as obliques, even in attributive role (e.g. biti v škripcih “to be in trouble”).
- The following subtypes are used in Slovenian treebanks:
- cc:preconj for preconjuncts (e.g. tako in tako X kot Y “both X and Y”)
- discourse:filler for filled pauses (e.g. eee “uhm”)
- flat:foreign for non-first words in quoted foreign phrases (e.g. Chamber of torture)
- flat:name for exocentric complex names (e.g. Novak in Janez Novak)
- parataxis:discourse for clausal discourse markers (e.g. a veš “you know”)
- parataxis:restart for repaired sentence beginnings (e.g. sits in the image is- … this man sits)
- The Slovenian treebanks do not use the
clf]andcompound]relations.
For more details on the syntactic annotation, see the Slovenian UD treebanks overview with statistical details on the dependency relation distribution.
Treebanks
There are two Slovenian UD treebanks:
- Slovenian-SSJ written language treebank
- Slovenian-SST spoken language treebank