Research map / Vision-language and multimodal / Visual question answering
DecAF: Joint Decoding of Answers and Logical Forms for Question Answering over Knowledge Bases
Donghan Yu, Sheng Zhang, Patrick Ng, Henghui Zhu, Alexander Hanbo Li, Jun Wang, Yiqun Hu, William Yang Wang, Zhiguo Wang, Bing Xiang
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 |
|---|---|---|---|
| combine direct answers with answers from executed logical forms (ablation) | DecAF · GrailQA dev | F1 overall: 54.7 -> 78.7; 72.4 -> 78.7 | Improved |
| use separate readers instead of one shared reader (ablation) | DecAF · GrailQA dev | F1 overall: 78.7 -> 80.3 | Mixed |
| use separate readers instead of one shared reader (ablation) | DecAF · CWQ | Hits@1: 68.1 -> 66.5 | Hurt |
| increase the logical-form answer weight from 0.49 to 0.51 (ablation) | DecAF answer combination · GrailQA dev | F1: 56.5 -> 78.7 | Improved |
| prefer the first executable logical-form answer, falling back to the first generated answer (ablation) | DecAF answer combination · GrailQA dev | F1: 54.7 -> 78.7 | Improved |
| generate multiple direct answers instead of one (ablation) | DecAF · GrailQA dev | F1 overall: 78.7 -> 78.1 | Hurt |
| increase the number of retrieved passages from 5 to 100 (ablation) | DecAF · all four benchmarks | QA performance | Improved |
| increase reader size from FiD-large to FiD-3B (ablation) | DecAF · GrailQA test | F1 overall: 76.0 -> 78.7 | Improved |
| use DPR instead of BM25 retrieval (ablation) | DecAF · CWQ | Hits@1: 68.1 -> 67.0 | Hurt |
| use DPR instead of BM25 retrieval (ablation) | DecAF · WebQSP | Hits@1: 79.0 -> 80.7 | Improved |
| use DPR instead of BM25 retrieval (ablation) | DecAF · FreebaseQA | Hits@1: 78.8 -> 79.0 | Improved |
| increase inference beam size from 1 to 20 (ablation) | DecAF · GrailQA dev | F1 | Improved |
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.
- FILM — FILM: Following Instructions in Language with Modular Methods (ICLR 2022) · 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
- TrustUQA: A Trustful Framework for Unified Structured Data Question Answering (AAAI 2025 · similarity 0.87)
- KBQA-o1: Agentic Knowledge Base Question Answering with Monte Carlo Tree Search (ICML 2025 · similarity 0.86)
- Promoting Knowledge Base Question Answering by Directing LLMs to Generate Task-relevant Logical Forms (AAAI 2025 · similarity 0.86)
- FlexKBQA: A Flexible LLM-Powered Framework for Few-Shot Knowledge Base Question Answering (AAAI 2024 · similarity 0.85)
- What Does a Query Answer Tell You? Informativeness of Query Answers for Knowledge Bases (AAAI 2024 · similarity 0.85)
- KG-Guard: Graph-Based Hallucination Detection for Knowledge Base Question Answering (NeurIPS 2026 · similarity 0.85)
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.
