Research map / Architectures and efficiency / RNNs and normalization
Two Birds with One Stone: Investigating Invertible Neural Networks for Inverse Problems in Morphology
Gözde Gül Şahin, Iryna Gurevych
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 one INN trained in both directions with Ly + Lx | Morphological inflection · Finnish, French, Irish, Romanian, Tagalog, Turkish | surface exact match (%): 85.15 -> 92.23; 88.94 -> 97.55; 61.11 -> 79.63; 89.45 -> 97.67; 30.05 -> 33.33; 88.89 -> 97.45 | Improved |
| Add reverse loss Lx to forward-only training (ablation) | INN morphological inflection · Finnish | lemma exact match (%): 0.01 -> 95.56 | Improved |
| Add reverse loss Lx to forward-only training (ablation) | INN morphological inflection · French | lemma exact match (%): 0.0 -> 98.56 | Improved |
| Add reverse loss Lx to forward-only training (ablation) | INN morphological inflection · Irish | lemma exact match (%): 0.0 -> 97.22 | Improved |
| Add reverse loss Lx to forward-only training (ablation) | INN morphological inflection · Romanian | lemma exact match (%): 0.0 -> 99.26 | Improved |
| Add reverse loss Lx to forward-only training (ablation) | INN morphological inflection · Tagalog | lemma exact match (%): 0.0 -> 85.41 | Improved |
| Add reverse loss Lx to forward-only training (ablation) | INN morphological inflection · Turkish | lemma exact match (%): 0.0 -> 99.62 | Improved |
| Add categorical latent variables with Lz (ablation) | INN lemmatization · All six languages | lemma exact match (%): The paper reports improvements in all languages; see Table 2 | Improved |
| Increase latent dimension and category count from z dim 2/categories 3 to z dim 6/categories 4 (ablation) | INN lemmatization · Finnish, French, Irish, Romanian, Tagalog, Turkish | lemma exact match (%): 97.54 -> 98.18; 96.03 -> 96.14; 94.2 -> 95.65; 88.85 -> 90.95; 90.03 -> 91.52; 99.01 -> 99.54 | Improved |
| Add tag loss Lt (ablation) | INN morphological inflection with Ly + Lx · Finnish and Tagalog | lemma exact match (%): 95.56 -> 93.33; 85.41 -> 69.73 | Hurt |
| Add tag loss Lt (ablation) | INN morphological inflection with Ly + Lx · Irish and Turkish | lemma exact match (%): 97.22 -> 100; 99.62 -> 99.74 | Improved |
| Add tag loss Lt (ablation) | INN morphological inflection with Ly + Lx · Finnish, French, Irish, Romanian, Tagalog, Turkish | tag F1 (%): 11.84 -> 12.42 -> 49.99; 23.5 -> 13.32 -> 26.01; 25.98 -> 20.55 -> 67.33; 16.5 -> 14.24 -> 0.02; 15.24 -> 10.13 -> 43.14 | Mixed |
Closest papers by meaning
- Labeling Neural Representations with Inverse Recognition (NeurIPS 2023 · similarity 0.84)
- Linear Decoding of Morphology Relations in Language Models (Student Abstract) (AAAI 2025 · similarity 0.84)
- Learning to Learn Morphological Inflection for Resource-Poor Languages (AAAI 2020 · similarity 0.84)
- A Latent Morphology Model for Open-Vocabulary Neural Machine Translation (ICLR 2020 · similarity 0.84)
- Training Neural Networks as Recognizers of Formal Languages (ICLR 2025 · similarity 0.81)
- Linguistic Collapse: Neural Collapse in (Large) Language Models (NeurIPS 2024 · similarity 0.81)
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.
