Datasets:
metadata
license: apache-2.0
language:
- uk
task_categories:
- text-classification
tags:
- legal
- legislation
- annotation
- ukrainian
- defectradar
- legdefqa
pretty_name: DefectRadar Human Annotations
size_categories:
- n<1K
DefectRadar Human Annotations
Human expert annotations of definitional defects in Ukrainian legislation. Gold-standard labels for the LegDefQA benchmark.
Annotation Task
Given a legal definition extracted from Ukrainian legislation, the annotator classifies it along two independent dimensions.
Dimension 1: Circulus in definiendo
Question: Does the definition add at least one distinguishing criterion that is not already present in the term itself?
| Label | Description | Example |
|---|---|---|
TRUE_DEFECT |
Definition is normatively empty -- it restates the term without adding content | "програмно-апаратна помилка" defined as "помилка у програмному забезпеченні" |
DOMAIN_REUSE |
Root word repeats but the genus provides substantive narrowing | "Державний реєстр лікарських засобів" defined as "єдина державна інформаційна система, що містить відомості про зареєстровані лікарські засоби" |
FALSE_POSITIVE |
Pipeline error -- no actual morphological or semantic overlap | Rare; indicates extraction or lemmatization error |
Dimension 2: Ignotum per ignotum
Question: Can a jurist apply this norm without consulting other laws for the meaning of terms used in the definition?
| Label | Description | Example |
|---|---|---|
CRITICAL |
Term undefined anywhere in the legislative corpus; norm is inapplicable without subjective interpretation | "креативне вираження" -- not defined in any Ukrainian law |
CROSS_REF |
Term is defined in another law; an explicit cross-reference would resolve the gap | Financial terms in payment services law defined only in Banking Law |
COMMON |
Well-known term; no statutory definition needed | General terms like "особа", "майно" |
FALSE_POSITIVE |
Pipeline error -- term is actually defined in the same law | Extraction missed the definition |
Sampling Strategy
300 definitions sampled from 5,799 extracted by DefectRadar from 44,021 active Ukrainian legislative acts:
| Stratum | Count | Description |
|---|---|---|
circulus |
50 | Pipeline flagged circulus only |
ignotum |
100 | Pipeline flagged ignotum only |
both |
50 | Pipeline flagged both defects |
clean |
100 | Pipeline found no defects |
Data Schema
Each annotation record contains:
{
"annotator_id": "string",
"definition_id": 1234,
"rada_id": "1089-IX",
"definiendum": "term being defined",
"pipeline_circulus": true,
"pipeline_ignotum": false,
"human_circulus": "TRUE_DEFECT|DOMAIN_REUSE|FALSE_POSITIVE|SKIPPED",
"human_circulus_comment": "optional free text",
"human_ignotum": "CRITICAL|CROSS_REF|COMMON|FALSE_POSITIVE|SKIPPED",
"human_ignotum_comment": "optional free text",
"timestamp": "2026-06-02T...",
"stratum": "circulus|ignotum|both|clean"
}
Annotation Tool
Interactive tool: overthelex/defectradar-annotator
Usage
from datasets import load_dataset
# After consolidation:
ds = load_dataset("overthelex/defectradar-annotations")
Related Resources
- Pipeline output: overthelex/ua-defectradar (5,799 definitions with pipeline labels)
- Paper: DefectRadar: Automated Detection of Definitional Defects in Legislation (Ovcharov & Kyrychenko, 2026)
Citation
@article{ovcharov2026defectradar,
title={DefectRadar: Automated Detection of Definitional Defects in Legislation via Morpho-Semantic NLP Pipeline over 24,000 Ukrainian Laws},
author={Ovcharov, Volodymyr and Kyrychenko, Ihor},
year={2026}
}