metadata

configs:
  - config_name: zero_billsum
    data_files:
      - split: train
        path: zero_billsum/train.jsonl
  - config_name: zero_eurlexsum
    data_files:
      - split: train
        path: zero_eurlexsum/train.jsonl
  - config_name: zero_govreport
    data_files:
      - split: train
        path: zero_govreport/train.jsonl
  - config_name: zero_multilong
    data_files:
      - split: train
        path: zero_multilong/train.jsonl
  - config_name: zero_multishort
    data_files:
      - split: train
        path: zero_multishort/train.jsonl
  - config_name: zero_multitiny
    data_files:
      - split: train
        path: zero_multitiny/train.jsonl
  - config_name: zero_inabs
    data_files:
      - split: train
        path: zero_inabs/train.jsonl
  - config_name: zero_ukabs
    data_files:
      - split: train
        path: zero_ukabs/train.jsonl

ZeroLexSumm Benchmark

The ZeroLexSumm Benchmark contains 8 datasets from various juresdictions such as the US, UK, EU, and India. The datasets are curated out of the LexSum datasets, each conatining 50 samples. All datasets adheres to the same format with columns: input, output, id, cluster, old_id and length.

cluster: which inout bracket 0-8k, 8k-16k, 16k+ and longest
old_id: ID in the LexSumm dataset
length: length of the input

Below you can find the links to the seperate datasets from ZeroLexSum.

BillSum: US Congressional bills with summaries by the Congressional Research Service.
InAbs: Indian Supreme Court cases with headnotes as summaries.
UKAbs: UK Supreme Court judgments with official press summaries.
EurLexSum: Summaries of enforced EU legislation from the EUR-Lex platform.
GovReport: U.S. Government Accountability Office reports with expert-written summaries.
MultiLexSum-Long: Multi-paragraph summaries of U.S. civil rights lawsuits.
MultiLexSum-Short: Single-paragraph summaries of civil rights lawsuits.
MultiLexSum-Tiny: One-sentence summaries of civil rights lawsuits in Twitter-like format.