| --- |
| title: DefectRadar Annotation Tool |
| emoji: ⚖️ |
| colorFrom: blue |
| colorTo: indigo |
| sdk: gradio |
| sdk_version: "5.33.0" |
| app_file: app.py |
| pinned: false |
| license: apache-2.0 |
| --- |
| |
| # DefectRadar -- Annotation Tool |
|
|
| Human annotation tool for definitional defects in Ukrainian legislation. |
| Part of the **LegDefQA** benchmark construction for the DefectRadar paper. |
|
|
| ## Task |
|
|
| Annotators classify pipeline-flagged definitions along two dimensions: |
|
|
| ### Circulus in definiendo |
| - **TRUE_DEFECT**: Definition is normatively empty, adds no distinguishing criterion |
| - **DOMAIN_REUSE**: Root word repeats but definition is substantively meaningful |
| - **FALSE_POSITIVE**: Pipeline error, no actual overlap |
| |
| ### Ignotum per ignotum |
| - **CRITICAL**: Term undefined anywhere in the corpus, norm is inapplicable |
| - **CROSS_REF**: Term defined in another law, needs explicit reference |
| - **COMMON**: Well-known term, no definition needed |
| - **FALSE_POSITIVE**: Pipeline error, term is defined in the same law |
| |
| ## Data |
| |
| - **Input**: 300 stratified definitions from 5,799 extracted from 44,021 active Ukrainian laws |
| - **Output**: Saved to [overthelex/defectradar-annotations](https://huggingface.co/datasets/overthelex/defectradar-annotations) |
| |
| ## Citation |
| |
| ```bibtex |
| @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} |
| } |
| ``` |
| |
