With a large DeBERTa model, annotating 500 to 1000 sentences per legal concept matches full annotation, and LLM-based annotation (Qwen 2.5) achieves NDCG scores close to or better than human-annotation-trained models on the statutory interpretation retrieval task.
Can gpt alleviate the burden of annotation? In Legal Knowledge and Information Systems
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Are manual annotations necessary for statutory interpretations retrieval?
With a large DeBERTa model, annotating 500 to 1000 sentences per legal concept matches full annotation, and LLM-based annotation (Qwen 2.5) achieves NDCG scores close to or better than human-annotation-trained models on the statutory interpretation retrieval task.