Research map / Large language models / Agents, code and math
A Balanced Neuro-Symbolic Approach for Commonsense Abductive Logic
Joseph Cotnareanu, Didier Chételat, Yingxue Zhang, Mark Coates
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 SC solver alongside the symbolic solver (ablation) | ARGOS-Symbolic (Llama3-8B) · FOLIO | accuracy: 59% -> 81% | Improved |
| add SC solver alongside the symbolic solver (ablation) | ARGOS-Symbolic (Llama3-8B) · CLUTRR | accuracy: 72% -> 76% | Improved |
| add commonsense score threshold (ablation) | ARGOS (Llama3-8B) · FOLIO | accuracy: 79% -> 81% | Improved |
| add relevance score threshold (ablation) | ARGOS (Llama3-8B) · FOLIO | accuracy: 80% -> 81% | Improved |
| add abduced commonsense clauses guided by SAT backbone to self-consistency | ARGOS (Llama3-8B) · CLUTRR | accuracy: 69% -> 76% | Improved |
| add abduced commonsense clauses to self-consistency | ARGOS (Llama3-70B) · QUAIL | accuracy: 75% -> 80% | Improved |
| add abduced clauses on datasets with strong neural baselines | ARGOS (Llama3-8B) · ProntoQA, ESNLI, CosmosQA | accuracy: 95% -> 97% | No clear effect |
| use imperfect logical translation including failures | ARGOS (Llama3-8B) · FOLIO | accuracy: 80% -> 78% | Hurt |
| use imperfect logical translation including failures | ARGOS (Llama3-8B) · QUAIL | accuracy: 82% -> 73% | Hurt |
| restrict ARGOS to 20 chain-of-thought calls per problem on hardest CLUTRR problems | ARGOS (Llama3-8B) · hardest CLUTRR problems | accuracy: 65% -> 54% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAT-LM — SatLM: Satisfiability-Aided Language Models Using Declarative Prompting (NeurIPS 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Divide and Translate: Compositional First-Order Logic Translation and Verification for Complex Logical Reasoning (ICLR 2025 · similarity 0.88)
- Learning to Reason via Mixture-of-Thought for Logical Reasoning (ICLR 2026 · similarity 0.88)
- TypedThinker: Diversify Large Language Model Reasoning with Typed Thinking (ICLR 2025 · similarity 0.87)
- RULEBREAKERS: Challenging LLMs at the Crossroads between Formal Logic and Human-like Reasoning (ICML 2025 · similarity 0.87)
- Logically Consistent Language Models via Neuro-Symbolic Integration (ICLR 2025 · similarity 0.87)
- NeSTR: A Neuro-Symbolic Abductive Framework for Temporal Reasoning in Large Language Models (AAAI 2026 · similarity 0.87)
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.
