The Annotation of Multiword Predicates in Kyrgyz Universal Dependencies
Year: 2026
Type: conference-paper
Abstract: This paper examines the annotation of multiword and complex predicates in Kyrgyz within the Universal Dependencies (UD) framework. It complements prior work on copular constructions and non-verbal negation in Turkic languages. Three classes of multiword constructions are considered: 1. Serial and compound verb sequences (V+V) that function as a single predicate, such as alip kel- 'bring' and basip çıgar- 'print (out)'; 2. Noun–verb light-verb constructions (N+V) that include experiencer idioms whose nominal element carries possessive morphology (baş aylan- 'be/get dizzy') and predicates with a non-referential object (kol koy- 'sign'); 3. Superficially similar but fully compositional sequences that should instead receive ordinary argument or modifier relations. The two existing Kyrgyz UD treebanks (Kyrgyz-KTMU and Kyrgyz-TueCL) annotate these constructions differently. In one of them, the compound:svc annotation ranges from genuine serial verbs to auxiliaries, copular particles, and the non-verbal negator. The compound:lvc relation is present in one treebank and absent from the other. We propose a morphology-based decision procedure that allows us to separate genuine multiword predicates from compositional sequences. Diagnostics such as the availability of possessive and case marking on the nominal element, separability by an intervening adverb, and the lexical and categorial status of the non-verbal element are used for these purposes. This improves the consistency, linguistic motivation, and cross-linguistic comparability of Kyrgyz UD annotation.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7224810702