README.md

---
license: cc-by-4.0
task_categories:
  - text-classification
  - token-classification
language:
  - uk
tags:
  - legal
  - legal-nlp
  - legislative-quality
  - definitional-defects
  - circulus-in-definiendo
  - ignotum-per-ignotum
  - ukrainian-law
size_categories:
  - 1K<n<10K
---

# DefectRadar: Definitional Defects in Ukrainian Legislation

Corpus of 5,799 legal definitions extracted from 879 Ukrainian laws (out of 44,021 active legislative acts), annotated for two types of definitional defects:

- **Circulus in definiendo** (tautology): the definition uses the term being defined (or its morphological derivative) in the *differentia specifica*
- **Ignotum per ignotum** (unknown through unknown): the definition relies on terms that are not defined in the same law

## Key Statistics

| Metric | Value |
|--------|-------|
| Legislative acts scanned | 44,021 |
| Laws with definitions | 879 (2.0%) |
| Unique definitions | 5,799 |
| Circulus in definiendo | 1,880 (32.4%) |
| Cross-reference ignotum | 1,885 (32.5%) |
| Both defects | 749 (12.9%) |
| Any defect | 3,016 (52.0%) |

## Error Analysis (LLM-as-Judge Validation)

Pipeline-flagged circulus cases were independently classified by a RAG-augmented GPT-4o judge:

| Classification | % |
|---------------|---|
| Domain terminology reuse (morphologically circular but substantively meaningful) | 84.8% |
| True defect (normatively empty) | 9.1% |
| False positive | 6.1% |

Ignotum severity stratification:

| Severity | % |
|----------|---|
| Cross-reference gap (defined in another law) | 70.8% |
| Critical (undefined anywhere in corpus) | 20.8% |
| Common (well-known, no definition needed) | 8.3% |

## Data Fields

| Field | Type | Description |
|-------|------|-------------|
| `rada_id` | string | Verkhovna Rada law identifier |
| `law_title` | string | Official title of the law |
| `definiendum` | string | The term being defined |
| `genus_proximum` | string | Superordinate category (genus) |
| `differentia_specifica` | string | Distinguishing characteristics |
| `full_definiens` | string | Complete right-hand side of the definition |
| `is_circulus` | bool | Pipeline circulus detection result |
| `circulus_tier` | string | Detection tier (morphological / semantic) |
| `overlapping_lemmas` | list[str] | Lemmas shared between definiendum and differentia |
| `circulus_explanation` | string | Pipeline explanation |
| `has_ignotum` | bool | Whether undefined terms were found |
| `undefined_terms` | list[str] | Terms used but not defined in this law |
| `ignotum_severity` | string | Severity classification |
| `ignotum_explanation` | string | Pipeline explanation |

## Source

Definitions extracted from the Verkhovna Rada Open Data portal (`data.rada.gov.ua`) using the LEX AI platform's RadaLegislationAdapter. Corpus snapshot: April 2026.

## Pipeline

The DefectRadar pipeline operates in three stages:
1. **Definition extraction**: rule-based (em-dash copula patterns) + fine-tuned XLM-RoBERTa for inline definitions
2. **Circulus detection**: Ukrainian lemmatization (pymorphy2-uk) + genus exclusion + discriminating-adjective overlap
3. **Ignotum detection**: cross-legislative definition graph across 879 laws

Pipeline code: [github.com/overthelex/secondlayer/tree/main/scripts/defectradar](https://github.com/overthelex/secondlayer/tree/main/scripts/defectradar)

## 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},
  journal={arXiv preprint},
  year={2026}
}
```

## License

CC BY 4.0