Fine-tuning DataText

Multilingual Legal QA Text Dataset

Type
QA Dataset
Domain
Law and Public
Language
KoreanHindiIndonesianArabicThaiBengaliArabic (Egypt)Japanese

Overview

A parallel multilingual question-answering dataset covering Korean public, civil, and criminal law, reviewed and corrected by law majors and working legal professionals. Answers do not stop at a conclusion: they name the governing statute and article, work through the conditions and exceptions that qualify the rule, and where the matter is administrative rather than purely legal, close by pointing the asker to the competent authority.

The same question and answer are carried across eight languages, and the legal substance stays Korean law in every one of them rather than being re-localized to another jurisdiction. Statute names and article references are rendered into each target language, so the dataset also serves as a parallel corpus of Korean legal terminology. Questions arrive in two registers — first-person practical consultation from someone in the situation, and doctrinal questions phrased the way a practitioner or examinee would frame them — so the set trains both advisory and analytical responses.

Applicable Areas

  • Supervised fine-tuning of legal-domain LLMs and assistants
  • Building retrieval-augmented legal question-answering systems
  • Cross-lingual evaluation of legal reasoning and statutory citation accuracy
  • Multilingual legal terminology alignment and translation quality benchmarking

Dataset Preview

indexcategoryq_idquestionanswerquestion_token_countanswer_token_count
2LawPublicPUB_00002사회복지센터에서 정규직이 아닌 계약직 근로자로 근무하고 있습니다. 근로기간이 1년 이상인 경우, 출산전후휴가와 육아휴직을 사용할 수 있나요?계약직 근로자도 근로기준법과 남녀고용평등과 일-가정 양립 지원에 관한 법률에 따라 산전후휴가와 육아휴직을 사용할 수 있습니다. (…) 출산 전과 출산 후를 통틀어 90일(한 번에 둘 이상의 자녀를 임신한 경우 120일)의 휴가를 주어야 하며 (…) (「근로기준법」 제7...48326
41LawCivilCIV_00041양도담보를 설정해 준 부동산을 20년간 소유의 의사로 평온, 공연하게 점유했습니다. 저는 양도담보권자를 상대로 점유취득시효를 원인으로 하여 소유권이전등기의 말소를 구할 수 있나요?아니요, 이 사안의 경유 소유권이전등기의 말소를 구할 수 없습니다. (…)68541
62LawCriminalCRI_00062재물손괴죄의 객체에 대한 법리오해, 채증법칙위반, 심리미진을 이유로 원심판결을 파기할 수 있는가?네, 질문하신 내용은 원심판결을 파기할 수 있는 사유에 해당합니다. (…)40619
  • Please review the data specifications and actual samples in advance.
  • The complete data shown in this preview is available in the sample download.

How to create

Collection

Questions are sourced across public, civil, and criminal law and assigned a category-prefixed identifier so that domain balance is auditable at a glance. Both registers are collected deliberately: practical consultation questions of the kind actually brought to a legal help desk, and doctrinal questions of the kind that turn on a point of interpretation.

Answers are drafted to a fixed shape — state the applicable rule, cite the governing statute and article, set out the conditions and the exceptions that limit it, and refer the asker onward when the next step is administrative rather than legal. Each answer is then localized into the eight target languages, with statute names, article numbers, and legal terms carried across intact.

Validation

Law majors and practicing legal professionals verify each answer against the cited statute, correcting answers that are directionally right but point to the wrong article, and rejecting answers that state a conclusion without the reasoning or without the exception that would change it. Reviewers also check that the answer addresses the question actually asked, since practical consultation questions often contain a detail — a length of service, a headcount, a property status — that determines the outcome.

Localized versions are then checked back against the Korean source so that no language version softens a legal requirement, drops an exception, or renumbers an article. Token counts are recorded per locale after review, which surfaces cases where one language version is materially shorter than its source and therefore likely to have lost content.

Flitto Curation Data Flitto Curation Data