SEMANTIC ROLE LABELING FOR KAZAKH: MODELS AND DATASETS

Abstract

A fundamental component of natural language understanding, semantic role labeling (SRL) clarifies the relationship between predicates and their arguments, therefore enabling activities including information extraction, machine translation, and question answering. Though much study has been done on SRL for high-resource languages, low-resource languages like Kazakh still relatively underexplored. This work fills the gap by offering both unique datasets and model architectures tailored specifically for KazakhSRL. Starting with annotated SRL datasets that reflect Kazakh's rich morphological characteristics, including agglutinative suffixes and case-marking patterns, we build. Building on these data sources, we create and contrast many SRL models, from feature-driven traditional machine learning techniques to neural architectures improved by morphological embeddings. Our findings show how using Kazakh's unique language traits improves performance and draw attention to ongoing issues caused by data sparsity and complex morphology. We also address pragmatic issues for dataset generation, annotation consistency, and generalization to other Turkic languages. The findings highlight the possibility of high-quality SRL in low-resource environments and open new paths forKazakh-language NLP study.

More by Aigerim Aitim

• Fast Neuro-Symbolic Constraint Solving for Knowledge Extraction Via NP-Complete Subset Sum ModelsLecture notes in mechanical engineering · 2026
• QazNLP: A Morphology-Aware Kazakh NLP Toolkit with Structured Decoding for POS and NERLecture notes in mechanical engineering · 2026
• MPNet-Based Semantic Retrieval for Legal Evidence Selection: A Set Cover and Subset Selection PerspectiveEngineering Technology & Applied Science Research · 2026
• MORPHOLOGICAL DISAMBIGUATION FOR THE KAZAKH LANGUAGE USING TRANSFORMER-BASED MODELSHerald of Kazakh-British technical university · 2026