Research map / Vision-language and multimodal / Multimodal LLMs
ReaSCAN: Compositional Reasoning in Language Grounding
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 random distractors instead of challenging distractors (ablation) | M-LSTM and GCN-LSTM · 2-relative-clauses test | exact match: 52.77 (14.6) -> 89.81 (0.60); 96.85 (0.75) -> 97.97 (0.48) | Helped |
| Increase command complexity from Simple to 2-relative-clauses | M-LSTM · ReaSCAN random split | exact match: 93.64 (2.52) -> 52.77 (14.6) | Hurt |
| Increase command complexity from Simple to 2-relative-clauses | GCN-LSTM · ReaSCAN random split | exact match: 97.86 (1.27) -> 96.85 (0.75) | Hurt |
| Use GCN-LSTM instead of M-LSTM | ReaSCAN · ReaSCAN random split | exact match: M-LSTM: 93.64 (2.52), 61.28 (1.81), 52.77 (14.6); GCN-LSTM: 97.86 (1.27), 97.19 (0.79), 96.85 (0.75) | Helped |
| Test on novel color and shape target combinations | M-LSTM and GCN-LSTM · A2: novel color attribute | exact match: 14.65 (0.55); 42.05 (4.55) | Hurt |
| Test on novel co-occurrence of relations rather than objects | M-LSTM and GCN-LSTM · B2: novel co-occurrence of relations | exact match: B1: 52.17 (1.63), 69.74 (0.30); B2: 39.41 (1.53), 52.80 (2.75) | Hurt |
| Test on a longer conjunctive clause | M-LSTM and GCN-LSTM · C1: novel conjunctive clause length | exact match: 49.68 (2.73); 57.01 (7.99) | Hurt |
| Test on novel recursive relative clauses | M-LSTM and GCN-LSTM · C2: novel relative clauses | exact match: 25.74 (1.36); 22.07 (2.66) | Hurt |
Closest papers by meaning
- Investigating Compositional Challenges in Vision-Language Models for Visual Grounding (CVPR 2024 · similarity 0.86)
- CraftFactory: A Conditioned Control Policy Benchmark for Compositional Generalization (AAAI 2025 · similarity 0.86)
- Compositional Reasoning in Language Models under Reinforcement Learning Post-Training (NeurIPS 2026 · similarity 0.86)
- Neural-Symbolic Recursive Machine for Systematic Generalization (ICLR 2024 · similarity 0.86)
- ConMe: Rethinking Evaluation of Compositional Reasoning for Modern VLMs (NeurIPS 2024 · similarity 0.86)
- NeSyCoCo: A Neuro-Symbolic Concept Composer for Compositional Generalization (AAAI 2025 · similarity 0.86)
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.