DatasetsEval/RusLang-Edu-100

Russian Linguistics & Grammar Instruction Dataset (v2.0)

A curated and academically verified instruction dataset for evaluation (Eval/Benchmarking), Supervised Fine-Tuning (SFT), and Alignment (RLHF / DPO) of Large Language Models (LLMs) on Russian grammar, orthography, punctuation, morphology, syntax, stylistics, and linguistic analysis.

📌 Key Highlights

• —Language: Standard Russian (ru)
• —Volume: 100 expert-curated and linguistically verified instruction cards
• —Granular Taxonomy: 12 linguistic categories, 16 distinct task types
• —Calibrated Complexity: CEFR scale (A1–C2) aligned with linguistic difficulty
• —Annotation Depth: Dual-target outputs (output with full didactic reasoning + concise answer_short), typed error_spans, and theoretical explanation
• —License: Creative Commons Attribution 4.0 International (CC-BY-4.0)

Dataset Structure & Schema

Data Fields

Task Types (task_type)

The dataset spans 16 distinct operational NLP and linguistic evaluation tasks:

• —error_correction — correcting orthographic, grammatical, or punctuation errors.
• —fill_gap — filling in missing letters, affixes, or punctuation marks.
• —choose_correct — selecting the normative option from multiple choices.
• —rule_apply — applying a specific orthographic/syntactic algorithm to raw inputs.
• —rule_explain — generating comprehensive explanations of grammatical rules.
• —qa_linguistics — answering theoretical questions across general and Slavic linguistics.
• —style_transfer — rewriting text across registers (e.g., colloquial $\to$ official-business, informal $\to$ scientific).
• —normalize_text — eliminating dialectisms, internet slang, or bureaucratese (канцелярит).
• —morph_analysis — full morphological parsing (parts of speech, case, gender, aspect, inflection).
• —syntax_analysis — parsing clause structures, grammatical centers, and sentence schemes.
• —error_detection — identifying and classifying linguistic violations without prompting the fix.
• —instruction_follow — executing multi-step textual editing constraints.
• —paraphrase — generating semantically invariant structural paraphrases.
• —compare_variants — comparative analysis of phonetic/dialectal systems (e.g., оканье vs. аканье).
• —dictation_like — phoneme-to-grapheme decoding from phonetic transcriptions.
• —generate_example — producing didactic examples and mnemonics for exceptions.

Linguistic Categories (category)

1. 1.`orthography` — Spelling (single/double Н/НН, НЕ/НИ, ПРЕ/ПРИ, root vowel alternation, hyphens, uppercase/lowercase).
2. 2.`punctuation` — Punctuation (compound/complex sentences, participial/adverbial clauses, appositions, dashes, colons).
3. 3.`morphology` — Morphological paradigms (noun gender/case, numeral declension, verb mood/aspect, degrees of comparison).
4. 4.`syntax` — Syntactic relations (verb government, clause schemes, direct/indirect speech conversion, word order).
5. 5.`lexicology` — Lexical semantics (paronyms, homonyms, synonymic series, idioms, archaisms).
6. 6.`stylistics` — Culture of speech (pleonasms, tautology, officialese, speech etiquette).
7. 7.`phonetics_orthoepy` — Orthoepy and phonology (normative accentuation, reduction, devoicing).
8. 8.`dialects_variants` — Dialectology and contact linguistics (isoglosses, regionalisms, linguistic interference).
9. 9.`grammar_rules` — Deductive formulations and step-by-step decision algorithms.
10. 10.`applied_writing` — Applied textual editing (official letters, resumes, paragraph compression).
11. 11.`error_taxonomy` — Classification of multi-error sentences and logical fallacies.
12. 12.`pedagogy` — Didactic formulations, children's explanations, and associative mnemonics.

Usage Example

Loading via Hugging Face datasets

from datasets import load_dataset

# Load train and validation splits
dataset = load_dataset("DatasetsEval/RusLang-Edu-100")

# Inspect a sample record
sample = dataset["train"][0]

print(f"ID: {sample['id']}")
print(f"Task: {sample['task_type']} | CEFR: {sample['level_cefr']}")
print(f"Instruction: {sample['instruction']}")
print(f"Input: {sample['input']}")
print(f"Short Answer: {sample['answer_short']}")
print(f"Explanation: {sample['explanation']}")

Sample Record (JSON)

{
  "id": "rl-000001",
  "task_type": "error_correction",
  "category": "orthography",
  "subcategory": "n-nn",
  "difficulty": "intermediate",
  "level_cefr": "B1",
  "language": "ru",
  "register": "neutral",
  "dialect_region": "standard",
  "instruction": "Исправь орфографическую ошибку и кратко объясни правило.",
  "input": "На улице было очень ветренно.",
  "output": "На улице было очень ветрено.\n\nВ наречиях на -о/-е пишется столько же «н», сколько в производящем прилагательном. Прилагательное «ветреный» пишется с одной «н» (исключение), следовательно, наречие «ветрено» пишется с одной «н» (ср. приставочные образования: «безветренный» — «безветренно»).",
  "answer_short": "ветрено",
  "explanation": "В наречиях сохраняется количество «н» производящей основы. Слово «ветреный» — исключение с одной «н».",
  "rule_tags": ["н-нн", "исключения", "наречие"],
  "skills": ["spelling", "explanation"],
  "certainty": "high",
  "split": "train",
  "source": "expert_curated",
  "license": "cc-by-4.0",
  "gold_unique": true,
  "meta": {
    "batch_id": "B001-HQ-V2",
    "cleaned": true,
    "review_status": "verified",
    "schema_version": "2.0"
  },
  "error_spans": [
    {
      "wrong": "ветренно",
      "right": "ветрено",
      "type": "orthography"
    }
  ]
}

📜 Citation & License

Distributed under the Creative Commons Attribution 4.0 International ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) license.

@misc{ruslang_instruction_2026,
  title={Russian Linguistics & Grammar Instruction Dataset (v2.0)},
  author={DatasetsEval Team},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/DatasetsEval/RusLang-Edu-100}}
}