Research map / Large language models / Instruction and fine-tuning
Learning to Learn Morphological Inflection for Resource-Poor Languages
Katharina Kann, Samuel R. Bowman, Kyunghyun Cho
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 |
|---|---|---|---|
| Use MAML initialization and fine-tuning instead of multi-task training and fine-tuning | Pointer-generator network · 29 target languages (average) | accuracy: +1.21% | Improved |
| Use MAML initialization and fine-tuning instead of multi-task training and fine-tuning | MED · 29 target languages (average) | accuracy: +3.16% | Improved |
| Add fine-tuning after multi-task training (ablation) | Pointer-generator network · 29 target languages (average) | accuracy: +10.41% | Improved |
| Add fine-tuning after multi-task training (ablation) | MED · 29 target languages (average) | accuracy: +19.06% | Improved |
| Use MAML-based cross-lingual fine-tuning instead of the UZH system | Pointer-generator network · 29 target languages (average) | accuracy: +1.7% absolute accuracy | Improved |
| Add source languages from other families while retaining related-family sources (ablation) | MAML-PG · Development languages (average) | accuracy: 61.85 -> 60.83 | No clear effect |
| Add source languages from other families while retaining related-family sources (ablation) | MulPG+FT · Development languages (average) | accuracy: 61.51 -> 60.43 | No clear effect |
| Exclude source languages from the target language family (ablation) | MAML-PG · Development languages (average) | accuracy: 60.83 -> 43.47 | Hurt |
| Exclude source languages from the target language family (ablation) | MulPG+FT · Development languages (average) | accuracy: 60.43 -> 42.41 | Hurt |
| Increase the number of MAML inner update steps from 2 to 5 (ablation) | MAML · Development languages (average) | accuracy: 61.99 -> 62.27 | No clear effect |
| Increase the number of MAML inner update steps from 5 to 8 (ablation) | MAML · Development languages (average) | accuracy: 62.27 -> 61.62 | No clear effect |
Closest papers by meaning
- Linear Decoding of Morphology Relations in Language Models (Student Abstract) (AAAI 2025 · similarity 0.85)
- Can LLMs Really Learn to Translate a Low-Resource Language from One Grammar Book? (ICLR 2025 · similarity 0.84)
- Language Ranker: A Metric for Quantifying LLM Performance Across High and Low-Resource Languages (AAAI 2025 · similarity 0.83)
- The Same but Different: Structural Similarities and Differences in Multilingual Language Modeling (ICLR 2025 · similarity 0.83)
- In-Context Unlearning: Language Models as Few-Shot Unlearners (ICML 2024 · similarity 0.83)
- SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability (ICML 2025 · similarity 0.82)
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.
