DOI: 10.18485/judig.2026.2.ch25 ISSN:

Modeling Expression of Eventuality Types in Bulgarian with Large Language Models

Svetla Koeva
Show PDF Cite
The paper examines the relationship between Bulgarian verbal aspect and eventuality types. It adopts a semantic framework based on two privative oppositions, intervality and terminativity, to distinguish eventuality types, and applies these features to Bulgarian perfective and imperfective verbs. The proposed features provide a systematic account of how Bulgarian imperfective and perfective verbs are distributed across eventuality types, while also capturing contextual coercion. A dataset of 254 Bulgarian verbs (the Aspect:BG dataset) was annotated with appropriate derivational and semantic features for each verb. Multiple experimental runs investigated the use of Large Language Models (LLMs) to evaluate human classification and to further support semi-automated classification. The analysis shows that different LLMs can be used individually or in combination to generate annotations for an expanded version of the dataset, with different models applied to different semantic or derivational features; however, expert verification is still recommended.

More from our Archive