Reproduce “From Natural Language to Exact Cover: A Neuro-Symbolic Approach to Zebra Puzzles” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Paulius Skaisgiris, Thomas Pammer, Veronika Semmelrock, Mykyta Ielanskyi, Maximilian Heisinger, Erich Kobler
NeuralExactCover combines an LLM that translates Zebra puzzle clues into structured tool calls with MiniExact, which searches an exact-cover encoding and checks whether it has a unique solution. On ZebraExtreme, the feedback-enabled system achieves 84.75% puzzle accuracy, versus 40.5% for Logic-LM. This benchmark shows performance on generated Zebra puzzles, not that clue translations faithfully capture prose.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.