jbdatascience/cryptonite

The Official Repository of the Cryptonite Dataset

Cryptonite: A Cryptic Crossword Benchmark for Extreme Ambiguity in Language

tl;dr

A dataset of cryptic crosswords ⬛⬜🕵️

Paper

https://arxiv.org/pdf/2103.01242.pdf

Abstract

Current NLP datasets targeting ambiguity can be solved by a native speaker with relative ease. We present Cryptonite, a large-scale dataset based on cryptic crosswords, which is both linguistically complex and naturally sourced. Each example in Cryptonite is a cryptic clue, a short phrase or sentence with a misleading surface reading, whose solving requires disambiguating semantic, syntactic, and phonetic wordplays, as well as world knowledge. Cryptic clues pose a challenge even for experienced solvers, though top-tier experts can solve them with almost 100% accuracy. Cryptonite is a challenging task for current models; fine-tuning T5-Large on 470k cryptic clues achieves only 7.6% accuracy, on par with the accuracy of a rule-based clue solver (8.6%).

Available on HuggingFace (🤗) Datasets:

Explore

https://huggingface.co/datasets/viewer/?dataset=cryptonite

Use

!pip install datasets
from datasets import load_dataset
dataset = load_dataset('cryptonite')

Alternative Title

Cryptonite: How I Stopped Worrying and Learned(?) to Love Ambiguity ⬛⬜🕵️