Research map / Large language models / Retrieval and knowledge
MMM: Multi-Stage Multi-Task Learning for Multi-Choice Reading Comprehension
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 the full MMM framework | RoBERTa-Large · DREAM test | accuracy: 85.0 -> 88.9 | Helped |
| add second-stage multi-task learning with RACE (ablation) | MMM · DREAM dev | accuracy: 68.5 -> 72.6 | Helped |
| add second-stage multi-task learning with RACE (ablation) | MMM · MC160 dev | accuracy: 72.5 -> 86.7 | Helped |
| add first-stage NLI coarse-tuning (ablation) | MMM · DREAM dev | accuracy: 69.5 -> 72.6 | Helped |
| add first-stage NLI coarse-tuning (ablation) | MMM · MC160 dev | accuracy: 80.8 -> 86.7 | Helped |
| use MAN instead of FCNN (ablation) | MMM · DREAM dev | accuracy: 71.2 -> 72.6 | Helped |
| use MAN instead of FCNN (ablation) | MMM · MC500 dev | accuracy: 81.5 -> 83.5 | Helped |
| add speaker normalization (ablation) | MMM · DREAM dev | accuracy: 71.4 -> 72.6 | Helped |
| use sliding windows with overlap 256 (ablation) | BERT-Base · TOEFL | accuracy: 50.0 -> 53.2 | Helped |
| use NLI rather than sentiment-analysis source-task tuning | BERT-Base · DREAM dev | accuracy: SST-2: 62.7; MultiNLI: 67.0 | Mixed |
| use MRQA as the span-based QA source task | BERT-Base transfer learning · DREAM dev | accuracy: 63.2 -> 61.2 | Hurt |
| train RACE and target jointly rather than sequentially | BERT-Base transfer learning · DREAM dev | accuracy: 70.2 -> 70.7 | Helped |
Closest papers by meaning
- Multi-Matching Network for Multiple Choice Reading Comprehension (AAAI 2019 · similarity 0.90)
- Multi-Task Learning with Generative Adversarial Training for Multi-Passage Machine Reading Comprehension (AAAI 2020 · similarity 0.89)
- Assessing the Benchmarking Capacity of Machine Reading Comprehension Datasets (AAAI 2020 · similarity 0.88)
- Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions (ICLR 2025 · similarity 0.86)
- Rethinking Generative Large Language Model Evaluation for Semantic Comprehension (ICML 2024 · similarity 0.86)
- PMRC: Prompt-Based Machine Reading Comprehension for Few-Shot Named Entity Recognition (AAAI 2024 · 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.