Cryptonite: A Cryptic Crossword Benchmark for Extreme Ambiguity in Language

TL;DR

Cryptonite introduces a large-scale, linguistically complex dataset of cryptic crossword clues designed to challenge NLP models with extreme ambiguity, requiring advanced disambiguation skills beyond current model capabilities.

Contribution

The paper presents Cryptonite, a novel dataset based on cryptic crosswords that captures complex linguistic ambiguity and challenges existing NLP models.

Findings

Fine-tuned T5-Large achieves only 7.6% accuracy on Cryptonite.

Cryptonite's clues are solvable by experts with nearly 100% accuracy.

Current models perform at chance level, highlighting the dataset's difficulty.

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%).