README.md

metadata

license: cc-by-nc-nd-4.0
task_categories:
  - question-answering
tags:
  - reasoning
  - linguistics
  - benchmark
pretty_name: L2
size_categories:
  - 1K<n<10K
source_datasets:
  - https://huggingface.co/datasets/ambean/lingOly
configs:
  - config_name: default
    data_files:
      - split: test
        path: test_small.zip
extra_gated_prompt: >-
  ### LingOly-TOO LICENSE AGREEMENT

  This dataset is governed by a CC-BY-NC-ND-4.0 license. 

  In addition to this license, we ask that uses of the dataset are in line with
  the Acceptable Use policy described below.

  ### Acceptable Use Policy

  This dataset is intended as a benchmark for reasoning in large language
  models. For the integrity of the benchmark, users should not:
      * Re-distribute the questions or answers of the benchmark in formats (such as plain text) which leak the benchmark to web-scraping.
      * Train language models directly using the content of this benchmark.
extra_gated_fields:
  By clicking Submit below I accept the terms of the license and Acceptable Use policy: checkbox
extra_gated_button_content: Submit

Links

• Website
• Paper
• Leaderboard
• Repo

Summary

Dataset format

The data was created by generating up to 6 obfuscations per problem for 82 problems source from original LingOly benchmark.

{'question_n':  # The question number in the problem
 'prompt':      # The main text of the question 
 'subprompts': [
  {'questionpart_n':         # For questions with several sub-parts (e.g. translating several sentences or matching paris of sentences)
   'question':               # The text of the question part
   'answer':                 # The correct answer
   'matches_human_scoring':  # Annotation metadata
   'fuzzy_matching':         # Annotation metadata
   'manual_edit':            # Annotation metadata
  }],
 'metadata.preamble':        # General text from the overall problem that is shared across all questions for that problem
 'metadata.context':         # Context text that includes important information, you should prepend your prompt with context for solvability
 'metadata.obfuscated':      # If this example was obfuscated or not
 'metadata.obf_num':         # Obfuscation counter (unique per problem). This is 0 if unobfuscated 
 'metadata.overall_question_n':    # The problem number
 'metadata.obfuscated_question_n': # Concatenation of problem number and obfuscation number
}

Citation

@article{khouja2025lingolytoo,
      title={LINGOLY-TOO: Disentangling Memorisation from Reasoning with Linguistic Templatisation and Orthographic Obfuscation}, 
      author={Khouja, Jude and Korgul, Karolina and Hellsten, Simeon and Yang, Lingyi and Neacșu, Vlad A. and Mayne, Harry and Kearns, Ryan O. and Bean, Andrew M. and Mahdi, Adam},
      year={2025},
      eprint=,
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url=, 
}