Research map / Label-efficient and robust learning / Self-supervised and contrastive
Cross-lingual Retrieval for Iterative Self-Supervised Training
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 |
|---|---|---|---|
| Iteratively mine pseudo-parallel sentences and train a multilingual seq2seq model | mBART · Unsupervised machine translation, 9 of 10 directions | BLEU: +2.4 BLEU average improvement | Helped |
| Iteratively mine pseudo-parallel sentences and train a multilingual seq2seq model | mBART · Tatoeba, 16 languages | accuracy: 21.5% in absolute accuracy | Helped |
| Initialize supervised MT from CRISS rather than mBART | mBART · Supervised MT, 34 directions | BLEU: +1.8 BLEU average | Helped |
| Continue iterative mining and training from Iter 1 to Iter 2 (ablation) | CRISS · WMT'16 English-Romanian unsupervised MT | BLEU: 30.1 -> 33.9; 32.2 -> 35 | Helped |
| Continue iterative mining and training from Iter 1 to Iter 2 (ablation) | CRISS · WMT'16 English-Romanian retrieval | accuracy: 98.3 -> 98.6 | Helped |
| Use bilingual rather than multilingual training (ablation) | CRISS pseudo-parallel data from 48 directions · English-centric translation | BLEU | Hurt |
| Increase the number of pivot languages from 1 toward 4 (ablation) | CRISS Iter 1 · Translation and sentence retrieval | translation quality and retrieval accuracy | Helped |
| Use more than 2 pivot languages (ablation) | CRISS Iter 1 · Translation and sentence retrieval | translation quality and retrieval accuracy | No clear effect |
| Finetune mBART using one Japanese-English parallel corpus | mBART · TED58 sentence retrieval across language directions | retrieval accuracy: 27% (absolute) on average | Helped |
| Use English-Romanian bilingual specialization instead of the multilingual CRISS25 model (ablation) | CRISS · WMT'16 English-Romanian retrieval | accuracy: 98.6 vs 98.5 | Mixed |
Closest papers by meaning
- Modeling Sequential Sentence Relation to Improve Cross-lingual Dense Retrieval (ICLR 2023 · similarity 0.88)
- XTREME: A Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalisation (ICML 2020 · similarity 0.88)
- FILTER: An Enhanced Fusion Method for Cross-lingual Language Understanding (AAAI 2021 · similarity 0.88)
- Improving Semantic Proximity in English-Centric Information Retrieval through Cross-Lingual Alignment (ICLR 2026 · similarity 0.88)
- CL2CM: Improving Cross-Lingual Cross-Modal Retrieval via Cross-Lingual Knowledge Transfer (AAAI 2024 · similarity 0.86)
- Constrained Decoding for Cross-lingual Label Projection (ICLR 2024 · similarity 0.84)
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.