Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
ALP: Data Augmentation Using Lexicalized PCFGs for Few-Shot Text Classification
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 ALP instead of no augmentation | Self-Training · Few-shot classification across AG News, SST-2, IMDB, and Yahoo! | accuracy: 77.73 { $\pm$ 4.91} -> 82.30 { $\pm$ 3.34}; 82.13 { $\pm$ 3.99} -> 86.18 { $\pm$ 1.27}; 54.38 { $\pm$ 3.79} -> 63.40 { $ | Helped |
| Use ALP augmentation with UST | UST · AG News, SST-2, IMDB, and Yahoo!, 5- and 10-shot | accuracy: 79.65 -> 84.72; 83.85 -> 87.41; 57.13 -> 73.22; 62.71 -> 78.01; 63.60 -> 71.32; 73.63 -> 76.33; 55.49 -> 61.20; 63.54 -> | Helped |
| Use ALP augmentation with MixText | MixText · AG News, SST-2, IMDB, and Yahoo!, 5- and 10-shot | accuracy: 81.14 -> 83.50; 87.11 -> 87.72; 51.46 -> 52.44; 50.91 -> 57.06; 68.09 -> 84.35; 72.87 -> 84.16; 66.60 -> 67.31; 67.40 -> | Helped |
| Replace ALP with EDA | Self-Training · SST-2, 10-shot | accuracy: 61.82 -> 53.96 | Hurt |
| Replace ALP with back-translation | MixText · IMDB, 5-shot | accuracy: 68.09 -> 62.44 | Hurt |
| Replace ALP with SSMBA | MixText · IMDB, 5-shot | accuracy: 68.09 -> 54.11 | Hurt |
| Use AugTrain-Train instead of AugTrain-Val | ALP · Average across four datasets, 5-shot | accuracy: 66.45 -> 72.40 | Helped |
| Use AugTrain-Train instead of AugTrain-AugTrain | ALP · Average across four datasets, 5-shot | accuracy: 67.21 -> 72.40 | Helped |
| Use Augment-and-Split with 80:20 ratio instead of AugTrain-Val | ALP · Average across four datasets, 5-shot | accuracy: 66.45 -> 72.13 | Helped |
| Use AugTrain-Train instead of Augment-and-Split (80:20) | ALP · Average across four datasets, 5-shot | accuracy: 72.13 -> 72.40 | Helped |
| Use AugTrain-Train instead of AugTrain-Val | EDA, back-translation, and SSMBA · Few-shot classification, k in {5,10} | average accuracy: 62.73 -> 64.42; 63.59 -> 67.40; 63.65 -> 68.09; 66.90 -> 67.31; 68.90 -> 73.15; 67.15 -> 68.91 | Helped |
| Use ALP rather than back-translation or SSMBA | Text augmentation · Augmented text across four benchmarks | SelfBLEU-2 / SelfBLEU-5: 0.75 / 0.49; 0.54 / 0.32 -> 0.33 / 0.08 | Helped |
Closest papers by meaning
- Tuning Language Models as Training Data Generators for Augmentation-Enhanced Few-Shot Learning (ICML 2023 · similarity 0.89)
- DEALT: LLM-driven Diversity-Enhanced Data Augmentation for Long-Tail Text Classification (AAAI 2026 · similarity 0.87)
- Compositional Generalization for Multi-Label Text Classification: A Data-Augmentation Approach (AAAI 2024 · similarity 0.86)
- Liberating Seen Classes: Boosting Few-Shot and Zero-Shot Text Classification via Anchor Generation and Classification Reframing (AAAI 2024 · similarity 0.85)
- Improving Unsupervised Constituency Parsing via Maximizing Semantic Information (ICLR 2025 · similarity 0.85)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 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.