We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Cryptonite: A Cryptic Crossword Benchmark for Extreme Ambiguity in Language

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%).
Comments: EMNLP 2021
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2103.01242 [cs.CL]
  (or arXiv:2103.01242v2 [cs.CL] for this version)

Submission history

From: Avia Efrat [view email]
[v1] Mon, 1 Mar 2021 19:01:01 GMT (6986kb,D)
[v2] Mon, 1 Nov 2021 18:44:36 GMT (7168kb,D)

Link back to: arXiv, form interface, contact.