ParlMix-UA-RU - Ukrainian Parliamentary Code-Mixing Dataset

ParlMix-UA-RU: Ukrainian Parliamentary Code-Mixing Dataset

Overview

ParlMix-UA-RU is a specialized, manually annotated linguistic resource designed to facilitate research on code-mixing (CM) and language identification (LID) between Ukrainian and Russian. The dataset is derived from official transcripts of Ukrainian parliamentary sessions (Verkhovna Rada), capturing language dynamics in a formal, high-register political discourse.

Dataset Composition & Selection

The corpus was developed through a multi-stage process to ensure its utility for complex Natural Language Processing (NLP) tasks:

• Code-Mixed Sentences: Initially, sentences were extracted based on a threshold of more than two out-of-vocabulary (OOV) tokens relative to monolingual Ukrainian dictionary. This selection typically indicates intra-sentential code-switching or mixed language use.

• Monolingual Balancing: To improve the robustness of Language Identification models and prevent classification bias, the dataset was subsequently enriched with monolingual Russian sentences from the parlamentary scripts. This ensures a balanced distribution of tokens across the target languages.

Annotation Methodology

The dataset follows a rigorous "human-in-the-loop" approach to establish a Gold Standard for linguistic research:

• Manual Tagging: Every token has been manually labeled with its specific language tag by expert linguists.

• Zero AI Involvement: No automated labeling or AI systems were used for the final annotations, ensuring maximum precision and the preservation of subtle linguistic nuances.

• Volume: The final dataset contains approximately 150,000 tokens with token-level manual annotations.

Research Applications

ParlMix-UA-RU is a valuable resource for:

1. Training and benchmarking Language Identification (LID) systems, especially for closely related languages.

2. Computational analysis of intra-sentential code-mixing patterns.

3. Sociolinguistic studies of language use in official governmental settings.

Technical Details

• Language(s): Ukrainian (UA), Russian (RU)

• Source: Verkhovna Rada transcripts

• Format: json

• Annotation level: Token-level language identification

Citation & Attribution

If you utilize this resource in your academic work, please use the following citation:

Kanishcheva, O., Shvedova, M., Dyka, L., & Husenko, K. (2026). ParlMix-UA-RU: Ukrainian Parliamentary Code-Mixing Dataset (Version 1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.14724542