Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Generalizing and Improving Bilingual Word Embedding Mappings with a Multi-Step Framework of Linear Transformations
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 whitening and de-whitening w.r.t. own language (ablation) | orthogonal mapping · EN-IT/DE/ES | accuracy: 39.27% -> 39.47% (EN-IT); 30.62% -> 29.71% (EN-FI) | Mixed |
| whiten both languages without de-whitening (CCA) (ablation) | orthogonal mapping · 4 language pairs | accuracy: 39.27% -> 32.27% (EN-IT) | Hurt |
| de-whiten w.r.t. one language (OLS) vs own language (ablation) | framework · 4 language pairs | accuracy: 2-4 points | Hurt |
| add target-language re-weighting (ablation) | framework · 4 language pairs | accuracy: 39.47% -> 43.80% (EN-IT) | Helped |
| add source-language re-weighting (ablation) | framework · 4 language pairs | accuracy: 39.47% -> 38.53% (EN-IT) | Hurt |
| add dimensionality reduction without re-weighting (ablation) | framework · 4 language pairs | accuracy: 39.47% -> 42.53% (EN-IT) | Helped |
| add dimensionality reduction on top of target re-weighting (ablation) | framework · 4 language pairs | accuracy: 43.80% -> 44.00% (EN-IT) | Helped |
| use inverted softmax retrieval instead of nearest neighbor (ablation) | proposed mapping · 4 language pairs | accuracy: 44.00% -> 45.27% (EN-IT); 44.27% -> 44.13% (EN-DE) | Mixed |
| use inverted nearest neighbor retrieval (ablation) | proposed mapping · 4 language pairs | accuracy: 44.27% -> 42.20% (EN-DE) | Hurt |
Closest papers by meaning
- Bilingual Lexicon Induction from Non-Parallel Data with Minimal Supervision (AAAI 2017 · similarity 0.90)
- Word translation without parallel data (ICLR 2018 · similarity 0.88)
- Unsupervised Hyper-alignment for Multilingual Word Embeddings (ICLR 2019 · similarity 0.88)
- Normalization of Language Embeddings for Cross-Lingual Alignment (ICLR 2022 · similarity 0.87)
- Enhancing Bilingual Lexicon Induction via Bi-directional Translation Pair Retrieving (AAAI 2024 · similarity 0.87)
- MMTEB: Massive Multilingual Text Embedding Benchmark (ICLR 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.