Research map / Architectures and efficiency / RNNs and normalization
GAAMA 2.0: An Integrated System That Answers Boolean and Extractive Questions
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 |
|---|---|---|---|
| Add question type classification, boolean answer classification, and score normalization | GAAMA · TyDi QA dev (second half) | Minimal Answer F1: 68.6 -> 72.6 | Helped |
| Use the integrated boolean-question pipeline | GAAMA · TyDi QA test | Minimal Answer F1: 68.0 -> 72.3 | Helped |
| Replace separate transformer stacks with a shared transformer and adapters (ablation) | GAAMA 2.0 · TyDi QA dev | Minimal Answer F1: 72.6 -> 73.0 | No clear effect |
| Replace separate transformer stacks with a shared transformer and adapters (ablation) | GAAMA 2.0 · TyDi QA dev | Parameter count (× 10^6): 1680 -> 563 | Helped |
| Replace separate transformer stacks with a shared transformer and adapters (ablation) | GAAMA 2.0 · TyDi QA dev | Model size (MiB): 3204 -> 1074 | Helped |
| Use the question-type classifier and span score to normalize answerability scores conditioned on question type (ablation) | GAAMA · TyDi QA dev boolean questions | Gold boolean questions above the answerability threshold: 259 -> 781 | Helped |
| Train the boolean answer classifier on expanded passage context rather than only GAAMA's minimal evidence span (ablation) | Boolean answer classifier · TyDi QA | YES/NO classification performance | Helped |
| Use adapters for the boolean answer classifier (ablation) | Boolean answer classifier · TyDi QA dev | YES/NO F1: 91.0/44.5 -> 90.9/41.5 | Mixed |
| Use adapters for the boolean answer classifier (ablation) | Boolean answer classifier · BoolQ-X dev | YES/NO F1: 81.3/65.7 -> 83.0/68.8 | Helped |
Closest papers by meaning
- S-Net: From Answer Extraction to Answer Synthesis for Machine Reading Comprehension (AAAI 2018 · similarity 0.87)
- SuRe: Summarizing Retrievals using Answer Candidates for Open-domain QA of LLMs (ICLR 2024 · similarity 0.86)
- NovelQA: Benchmarking Question Answering on Documents Exceeding 200K Tokens (ICLR 2025 · similarity 0.86)
- Rethinking Generative Large Language Model Evaluation for Semantic Comprehension (ICML 2024 · similarity 0.86)
- TACT: Advancing Complex Aggregative Reasoning with Information Extraction Tools (NeurIPS 2024 · similarity 0.85)
- TableBench: A Comprehensive and Complex Benchmark for Table Question Answering (AAAI 2025 · 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.