README.md

metadata

task_categories:
  - text-generation
language:
  - en
tags:
  - report-generation
  - information retrieval
  - retrieval
  - generation
  - legal
  - law
size_categories:
  - 1M<n<10M

Usage

The dataset is in folder according to the task and type (e.g. generation or collection for IR).

You can load a specific file (say the test set of the generation task) with this command:

from datasets import load_dataset
dataset = load_dataset("jhu-clsp/CLERC", data_files={"data": f"generation/test.jsonl"})["data"]

Dataset Description:

CLERC is a large-scale dataset of federal case law from the paper (https://arxiv.org/abs/2406.17186). It has a few subsets:

1. collection: records 1.84M original federal case law, both in doc and passage forms. This can be used for doing retrieval or alternative tasks.
2. generation: includes data for the RAG task as described in the paper. Also see the github (https://github.com/bohanhou14/CLERC) for more details.
3. qrels and queries: qrels and queries used for the retrieval task, all following MSMARCO formats.
4. triples, teva_train_triples: triples in different formats for evaluation and training (with tevatron)