Research map / Label-efficient and robust learning / Recognition and long-tail
Offline bilingual word vectors, orthogonal transformations and the inverted softmax
Samuel L. Smith, David H. P. Turban, Steven Hamblin, Nils Y. Hammerla
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 the SVD alignment with inverted softmax and dimensionality reduction instead of Mikolov's linear mapping | Bilingual word translation with the expert word dictionary · English to Italian | precision@1: 0.338 -> 0.431 | Improved |
| Add inverted softmax retrieval to SVD alignment (ablation) | Bilingual word translation with the expert word dictionary · English to Italian | precision@1: 0.369 -> 0.417 | Improved |
| Add dimensionality reduction after inverted-softmax retrieval (ablation) | Bilingual word translation with the expert word dictionary · English to Italian | precision@1: 0.417 -> 0.431 | Improved |
| Add inverted softmax retrieval to SVD alignment (ablation) | Bilingual word translation with the expert word dictionary · Italian to English | precision@1: 0.322 -> 0.373 | Improved |
| Add dimensionality reduction after inverted-softmax retrieval (ablation) | Bilingual word translation with the expert word dictionary · Italian to English | precision@1: 0.373 -> 0.380 | Improved |
| Replace the expert word dictionary with an identical-character-string pseudo-dictionary | SVD bilingual word alignment · English to Italian | precision@1: 0.010 -> 0.399 | Improved |
| Train alignment using paired Europarl sentence vectors instead of a word dictionary | SVD bilingual word alignment · English to Italian | precision@1: 0.313 -> 0.428 | Improved |
| Use inverted softmax instead of nearest-neighbor retrieval (ablation) | SVD sentence-vector alignment with a phrase dictionary · Italian to English, 200k candidate sentences | precision@1: 0.656 -> 0.486 | Hurt |
| Use paired sentence vectors as the training dictionary | SVD bilingual word alignment · English to Italian, 200k candidate sentences | precision@1: 0.268 -> 0.431 | Improved |
| Use paired sentence vectors as the training dictionary | SVD bilingual word alignment · Italian to English, 200k candidate sentences | precision@1: 0.473 -> 0.656 | Improved |
Closest papers by meaning
- Word translation without parallel data (ICLR 2018 · similarity 0.86)
- Generalizing and Improving Bilingual Word Embedding Mappings with a Multi-Step Framework of Linear Transformations (AAAI 2018 · similarity 0.86)
- Accessing Higher Dimensions for Unsupervised Word Translation (NeurIPS 2023 · similarity 0.85)
- Normalization of Language Embeddings for Cross-Lingual Alignment (ICLR 2022 · similarity 0.84)
- Enhancing Bilingual Lexicon Induction via Bi-directional Translation Pair Retrieving (AAAI 2024 · similarity 0.83)
- Calibrating Translation Decoding with Quality Estimation on LLMs (NeurIPS 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.
