esimijoq/Kazakh-Russian-Child-Directed-Speech-Corpus

Kazakh-Russian Child-Directed Language Corpus

Dataset Description

The corpus combines narrative texts and child-adult dialogue transcripts in Kazakh and Russian. It was created to support research on low-resource NLP, language acquisition, language modeling, and morphologically aware tokenization.

The corpus includes narrative materials such as fairy tales, children’s literature, cartoons/subtitles, translated stories, and educational texts.

This dataset is a work in progress and will be updated periodically as additional Kazakh and Russian child-directed materials are collected, cleaned, and organized.

This dataset accompanies the accepted paper:

A Kazakh-Russian Corpus of Child-Directed Language for Low-Resource Languages Albina Mukusheva, Achille Fusco, Cristiano Chesi. Accepted at the International Conference LAnguage TEchnologies for Low-resource Languages, Fes, Morocco. Forthcoming.

Languages

- Kazakh (`kk`)
- Russian (`ru`)

Dataset Structure

The dataset contains organized text folders and Russian dialogue data:

kazakh/
russian/

Each organized .txt file follows this format:

Title


Sentence 1.
Sentence 2.
Sentence 3.

The Russian dialogue data is stored in spreadsheet format.

Dialogue Data

The Russian dialogue files contain child-adult interaction data. The main columns are:

Column       Description
━━━━━━━━━━━  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
AGES         Child age at the time of the utterance
───────────  ───────────────────────────────────────────────────
SPEAKERS     Speaker identity, e.g. child, mother, interviewer
───────────  ───────────────────────────────────────────────────
SENTENCES    Dialogue utterance text

Dataset Statistics

The current release contains approximately:

Language    Documents    Tokens
━━━━━━━━━━  ━━━━━━━━━━━  ━━━━━━━━
Kazakh          1,655     1.07M
──────────  ───────────  ────────
Russian         1,988     2.27M

The Russian portion includes both narrative texts and child-adult dialogue transcripts.

Data Sources

The corpus was compiled from multiple child-directed sources, including fairy tales, children’s literature, cartoons/subtitles, educational materials, translated stories, and Russian child-adult dialogue transcripts.

Because the dataset is being expanded, the source list will be updated over time. Detailed source-level information is provided in SOURCES.md.

Users should check the original source licenses and terms of use before commercial use or redistribution.

Intended Use

This dataset is intended for:

• —tokenizer training and evaluation
• —low-resource language modeling
• —Kazakh and Russian corpus analysis
• —child-directed language research
• —morphology-aware NLP experiments

Limitations

- Some narrative TXT files were segmented into sentences automatically, so sentence boundaries may occasionally be imperfect.
- The Russian dialogue data is provided in spreadsheet form and may require separate preprocessing depending on the intended experiment.

Contact

If you find any mistakes or issues in the dataset, please contact Albina Mukusheva at albina.mukusheva@iusspavia.it.

Citation

If you use this dataset, please cite:

@inproceedings{mukusheva2026kazakhrussian,
title = {A Kazakh-Russian Corpus of Child-Directed Language for Low-Resource Languages},
author = {Mukusheva, Albina and Fusco, Achille and Chesi, Cristiano},
booktitle = {Proceedings of the International Conference LAnguage TEchnologies for Low-resource Languages},
year = {2026},
note = {Forthcoming}
}