Research map / Large language models / Agents, code and math
From Hypothesis to Premises: LLM-based Backward Logical Reasoning with Selective Symbolic Translation
Qingchuan Li, Mingyue Cheng, Zirui Liu, Daoyu Wang, Yuting Zeng, Tongxuan Liu
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 |
|---|---|---|---|
| use backward hypothesis-driven reasoning instead of forward CoT reasoning (ablation) | HBLR · ProofWriter | accuracy: 81.24 -> 89.41 | Improved |
| use backward hypothesis-driven reasoning instead of forward CoT reasoning (ablation) | HBLR · AR-LSAT | accuracy: 38.86 -> 44.07 | Improved |
| use selective translation instead of all-FOL translation (ablation) | HBLR · ProntoQA | accuracy: -1.72% vs All-FOL | Mixed |
| use selective translation instead of all-NL translation (ablation) | HBLR · five benchmarks | accuracy: +7.2% average | Improved |
| use selective translation instead of all-FOL translation (ablation) | HBLR · five benchmarks | accuracy: +4.7% average | Improved |
| use selective translation instead of all-FOL translation (ablation) | HBLR · ProofWriter | accuracy: +12.83% | Improved |
| HBLR vs SymbCoT on GPT-4 | HBLR · FOLIO | accuracy: 78.19 -> 84.22 | Improved |
| HBLR vs Logic-LM on GPT-3.5-Turbo | HBLR · LogicalDeduction | accuracy: 69.06 -> 49.77 | Hurt |
| HBLR vs CoT on DeepSeek-R1 | HBLR · AR-LSAT | accuracy: 76.91 -> 86.47 | Improved |
| HBLR vs CoT on DeepSeek-R1 | HBLR · ProntoQA | accuracy: 99.57 -> 99.72 | No clear effect |
| backward reasoning raises effective reasoning rate vs SymbCoT | HBLR · ProntoQA | effective reasoning rate: 73.65 -> 95.58 | Improved |
| upgrade DeepSeek-V3 to DeepSeek-R1 under HBLR | HBLR · average of benchmarks | accuracy: 82.31 -> 94.34 | Improved |
Closest papers by meaning
- Divide and Translate: Compositional First-Order Logic Translation and Verification for Complex Logical Reasoning (ICLR 2025 · similarity 0.89)
- Reasoning-as-Logic-Units: Scaling Test-Time Reasoning in Large Language Models Through Logic Unit Alignment (ICML 2025 · similarity 0.89)
- Large Language Models Meet Symbolic Provers for Logical Reasoning Evaluation (ICLR 2025 · similarity 0.89)
- RATT: A Thought Structure for Coherent and Correct LLM Reasoning (AAAI 2025 · similarity 0.88)
- Efficient Reasoning for Large Reasoning Language Models via Certainty-Guided Reflection Suppression (AAAI 2026 · similarity 0.88)
- Phenomenal Yet Puzzling: Testing Inductive Reasoning Capabilities of Language Models with Hypothesis Refinement (ICLR 2024 · similarity 0.88)
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.
