German Legal Sentences with Weak Supervision Labels

German Legal Sentences (GLS) is a training dataset for semantic sentence matching in German legal documents, created by lavis-nlp and last updated in October 2022. It was automatically generated using weak supervision, where imperfect labels are created through heuristics like legal citation matching and BM25 similarity. The sentences and citations are parsed from real judicial decisions provided by Open Legal Data.

Use Cases

• Train semantic similarity models on German legal sentences using weak supervision labels for text retrieval tasks.
• Fine-tune models for semantic similarity scoring on monolingual German text in the legal domain.
• Analyze the relationship between legal sentences and their citations parsed from judicial decisions.
• Benchmark text retrieval algorithms on machine-generated annotations for German legal documents.

Strengths

• Dataset is derived from real judicial decisions provided by Open Legal Data, grounding it in authentic legal text.
• Employs a combination of legal citation matching and BM25 similarity for label generation, providing a structured weak supervision approach.
• Specifically designed for semantic sentence matching in the German legal domain, offering task-focused data.

Limitations

• Labels are generated via heuristics and described as 'imperfect', indicating potential label noise.
• Key dataset metrics like row count, column details, and file size are unknown, limiting assessment of scale and structure.
• The dataset's last update was in October 2022, which may affect its relevance for current legal language models.

Provenance

Source

Open Legal Data (http://openlegaldata.io/)

Collection Method

Sentences and citations were parsed from judicial decisions. Labels were automatically generated using weak supervision with heuristics (legal citation matching and BM25 similarity).

Freshness

Last updated on 2022-10-20.

Geography

Region is indicated as 'us' in tags, but the data consists of German legal sentences; spatial coverage is unclear.