Research map / Large language models / Agents, code and math
Decrypting Cryptic Crosswords: Semantically Complex Wordplay Puzzles as a Target for NLP
Joshua Rozner, Christopher Potts, Kyle Mahowald
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ACW + ACW-descramble curriculum before fine-tuning | T5-base · naive split test | top-correct (%): 16.3 -> 21.8 | Improved |
| add ACW + ACW-descramble curriculum | T5-base · word-initial disjoint split test | top-correct (%): 1.1 -> 6.5 | Improved |
| curriculum with ACW only (ablation) | T5-base · naive split full dev | top-output correct (%): 15.7 -> 18.3 | Improved |
| curriculum with ACW-descramble only (ablation) | T5-base · naive split full dev | top-output correct (%): 15.7 -> 13.1 | Hurt |
| replace ACW-descramble with bare ACW-descramble-word in the curriculum (ablation) | T5-base · naive split full dev | top-output correct (%): 20.2 -> 17.8 | Hurt |
| add Anagrams dataset to ACW + ACW-descramble curriculum (ablation) | T5-base · naive split full dev | top-output correct (%): 20.2 -> 20.1 | No clear effect |
| add Anagrams dataset to curriculum (ablation) | T5-base · anagram subset of dev | top-output correct (%): 24.0 -> 27.1 | Improved |
| append answer-length enumeration to input and filter outputs by length (ablation) | T5-base · naive split test | top correct (%): 15.6 -> 16.3 | Improved |
| append answer-length enumeration to input (ablation) | T5-base · naive split test, no filter | outputs with correct length (%): 16.2 -> 48.5 | Improved |
| switch from naive random split to word-initial disjoint split | T5-base · test | top-correct (%): 16.3 -> 1.1 | Hurt |
| restrict dev clues to answers not seen in training | T5-base · naive split dev | top-correct (%): 16.0 -> 3.0 | Hurt |
| add phrasal definition to the scrambled letters input (ablation) | T5-base · descrambling, word-initial disjoint | top correct (%): 3.7 -> 12.4 | Improved |
Closest papers by meaning
- A Reasoning-Based Approach to Cryptic Crossword Clue Solving (ICML 2025 · similarity 0.88)
- LINGOLY: A Benchmark of Olympiad-Level Linguistic Reasoning Puzzles in Low Resource and Extinct Languages (NeurIPS 2024 · similarity 0.85)
- Enigmata: Scaling Logical Reasoning in Large Language Models with Synthetic Verifiable Puzzles (NeurIPS 2025 · similarity 0.84)
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 2025 · similarity 0.84)
- MathAttack: Attacking Large Language Models towards Math Solving Ability (AAAI 2024 · similarity 0.84)
- From Tokens to Words: On the Inner Lexicon of LLMs (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
