Research map / Architectures and efficiency / RNNs and normalization
Can Neural Networks Understand Logical Entailment?
Richard Evans, David Saxton, David Amos, Pushmeet Kohli, Edward Grefenstette
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 PossibleWorldNet, evaluating formulas across learned representations of multiple possible worlds | Propositional Logic entailment · test (easy) | accuracy: 77.8 -> 98.6 | Improved |
| Use PossibleWorldNet, evaluating formulas across learned representations of multiple possible worlds | Propositional Logic entailment · test (hard) | accuracy: 74.2 -> 96.7 | Improved |
| Use PossibleWorldNet, evaluating formulas across learned representations of multiple possible worlds | Propositional Logic entailment · test (big) | accuracy: 74.2 -> 93.9 | Improved |
| Use PossibleWorldNet, evaluating formulas across learned representations of multiple possible worlds | Propositional Logic entailment · test (massive) | accuracy: 59.3 -> 73.4 | Improved |
| Use PossibleWorldNet, evaluating formulas across learned representations of multiple possible worlds | Propositional Logic entailment · test (exam) | accuracy: 85.0 -> 96.0 | Improved |
| Use parse-tree structure rather than sequence-only encoding | Propositional Logic entailment · test (easy) | accuracy: 68.3 -> 77.8 | Improved |
| Use convolutional sequence encoders rather than LSTM encoders | Propositional Logic entailment · test (easy) | accuracy: 68.3 -> 59.7 | Hurt |
| Add symbolic vocabulary permutation during training (ablation) | Weaker models, including BiDirLSTM encoders and traversal models and convolutional networks · Propositional Logic | accuracy: 2--3 point increase | Improved |
| Add symbolic vocabulary permutation during training (ablation) | Tree-based models · Propositional Logic | accuracy: 7--15 point increases | Improved |
| Increase the number of imagined possible worlds (ablation) | PossibleWorldNet · Propositional Logic validation | error rate | Improved |
| Replace tree-structured formula encoding with LSTM traversal of the joined sequent | LSTM Traversal versus LSTM Encoders · Propositional Logic test (easy) | accuracy: 68.3 -> 61.8 | Hurt |
Closest papers by meaning
- Infusing Knowledge into the Textual Entailment Task Using Graph Convolutional Networks (AAAI 2020 · similarity 0.87)
- Proposition Entailment in Educational Applications using Deep Neural Networks (AAAI 2018 · similarity 0.87)
- SciTaiL: A Textual Entailment Dataset from Science Question Answering (AAAI 2018 · similarity 0.87)
- Recognizing and Justifying Text Entailment Through Distributional Navigation on Definition Graphs (AAAI 2018 · similarity 0.86)
- Defeasible Visual Entailment: Benchmark, Evaluator, and Reward-Driven Optimization (AAAI 2025 · similarity 0.86)
- Boule or Baguette? A Study on Task Topology, Length Generalization, and the Benefit of Reasoning Trace (NeurIPS 2026 · 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.
