home edit page issue tracker

This page pertains to UD version 2.

UD for Abaza

Tokenization and Word Segmentation



Abaza uses all universal POS tags except PUNCT and SYM.

The POS tag VERB is applied to all word-forms which have at least some verbal affixes.

Two predicates in the treebank are marked as auxiliaries: акIвызлара ‘be’ and чпара ‘do’. The verb чпара ‘do’ is marked as an auxililiary when it is used in constructions with lexical verbs borrowed from Russian, e.g., вязать й-с-чпа-тI (knit.INF (Rus) 3PL.ABS-1SG.ERG-do-DCL) ‘I knitted them’.


Morphological features are added automatically based on glosses and then manually checked. Not all glosses are converted to morphological features. The selected morphological features available for most nouns and verbs are presented below.

Nominal features

Verbal features

Negative verbal forms have the feature Polarity=Neg. Verbal forms with the causative prefix have the feature Caus=Yes. Interrogative verbal forms have the feature Int=Yes. Imperative verbal forms have the feature Mood=Imp. For relative verbal forms, the type of the relative clause is specified: RelType=Abs, RelType=Mnr, RelType=Tmp, etc.



Abaza is a morphologically ergative but syntactically nominative-accusative language. Subjects are usually cross-referenced on the verb as absolutive (in intransitive clauses) or ergative (in transitive clauses) arguments, direct objects are usually cross-referenced on the verb as absolutive arguments (in transitive clauses), indirect objects are cross-referenced on the verb as indirect objects, oblique arguments are not marked on the verb.

Relations Overview

Abaza uses all universal UD relations except appos, clf, compound, dep, expl, goeswith, list, orphan and punct.

One language-specific relation is introduced for the Abaza treebank: dep:repeat. It is used when a word is repeated for no particular reason.


There is just one Abaza UD treebank: