You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Kazakh Sentence Similarity Dataset

Overview: This dataset is designed for training and evaluating models on semantic textual similarity in the Kazakh language. Each example consists of a pair of sentences and a label that defines the semantic relationship between them. The dataset supports multi-class classification across various levels of semantic similarity.

Format csv:

text text_pair label
Мысық үйде отыр. Мысық үйде отыр. exact
Мен алма жедім. Мен алма жедім. exact
Бүгін ауа райы жылы. Күн жайма-шуақ. paraphrase
Ол университетке түсті. Ол жоғары оқу орнына қабылданды. paraphrase
Мен дүкенге бардым. Мен нан сатып алдым. partial
Ол футбол ойнады. Ол далада жүгірді. partial
Экологияны сақтау маңызды. Қоқысты қайта өңдеу пайдалы. contextual
Оқушылар сабақ оқып жатыр. Мұғалімдер жоспар құруда. contextual
Мен кітап оқыдым. Аспанда бұлттар жүр. unrelated
Телефонды қуаттадым. Бүгін мен балық ауладым. unrelated

Label Definitions:

  • exact — Sentences are identical or nearly identical with equivalent meaning.
  • paraphrase — Sentences convey the same meaning using different wording.
  • partial — Sentences partially overlap in meaning but aren't fully aligned.
  • contextual — Sentences are on a related topic but do not express the same idea directly.
  • unrelated — Sentences have no meaningful connection.

Use Cases:

  • Sentence similarity modeling in low-resource languages
  • Paraphrase detection for Kazakh
  • Fine-tuning multilingual models like KazBERT, mBERT, XLM-R
  • Multilingual and cross-lingual NLP tasks
Downloads last month
45