Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Hyper-alignment for Multilingual Word Embeddings
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 |
|---|---|---|---|
| jointly constrain alignments across a language triplet | UMH · Triplet alignment, indirect translation | accuracy: 61.7 -> 68.3; 72.3 -> 77.9; 31.9 -> 42.4 | Helped |
| jointly align 11 languages instead of aligning pairs independently | UMH · MUSE indirect translation (All) | accuracy: 51.7 -> 55.3 | Helped |
| jointly align 11 languages instead of aligning pairs independently | UMH · MUSE direct translation (Avg.) | accuracy: 74.9 -> 75.2 | No clear effect |
| use uniform weights rather than larger weights for pivot alignments (ablation) | UMH · MUSE translation with NN (Direct) | accuracy: 69.4 -> 65.5 | Hurt |
| use uniform weights rather than larger weights for pivot alignments (ablation) | UMH · MUSE translation with NN (Indirect) | accuracy: 55.3 -> 56.9 | Helped |
| initialize with Gromov-Wasserstein instead of convex relaxation (ablation) | UMH · MUSE translation with NN (Direct) | accuracy: 77.8 -> 78.6 | Helped |
| initialize with Gromov-Wasserstein instead of convex relaxation (ablation) | UMH · MUSE translation with NN (Indirect) | accuracy: 71.5 -> 69.8 | Hurt |
| add five languages to training (ablation) | UMH trained on six languages · MUSE direct translation (Avg., six common languages) | accuracy: 78.6 -> 78.0 | Hurt |
| add five languages to training (ablation) | UMH trained on six languages · MUSE training on six common languages | training time: 2h -> 5h | Hurt |
Closest papers by meaning
- Normalization of Language Embeddings for Cross-Lingual Alignment (ICLR 2022 · similarity 0.91)
- Word translation without parallel data (ICLR 2018 · similarity 0.90)
- Unsupervised Cross-Modal Alignment of Speech and Text Embedding Spaces (NeurIPS 2018 · similarity 0.89)
- Multilingual Alignment of Contextual Word Representations (ICLR 2020 · similarity 0.89)
- MMTEB: Massive Multilingual Text Embedding Benchmark (ICLR 2025 · similarity 0.86)
- HyperET: Efficient Training in Hyperbolic Space for Multi-modal Large Language Models (NeurIPS 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.