Research map / Vision-language and multimodal / CLIP and zero-shot
Multilingual Transfer Learning for QA using Translation as Data Augmentation
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 |
|---|---|---|---|
| Translate questions while keeping contexts in English | mBERT QA · MLQA | G-XLT F1: 51.7 -> 60.9 | Helped |
| Translate contexts while keeping questions in English | mBERT QA · MLQA | XLT F1: 61.7 -> 58.5 | Hurt |
| Combine question-only, context-only, and question-plus-context translation data | mBERT QA · MLQA | XLT F1: 61.7 -> 64.2 | Hurt |
| Add language-adversarial training across all languages | mBERT QA · MLQA | G-XLT F1: 51.7 -> 61.2 | Helped |
| Add language-arbitration answer-agreement and question-similarity objectives across all languages | mBERT QA · MLQA | G-XLT F1: 51.7 -> 61.9 | Helped |
| Use all translated languages instead of only Chinese for adversarial training (ablation) | AT · MLQA | G-XLT F1: 56.5 -> 61.2 | Helped |
| Use all translated languages instead of only Chinese for language arbitration (ablation) | LAF PSA+QS · MLQA | G-XLT F1: 56.8 -> 61.9 | Helped |
| Add question-similarity loss to PSA with all languages (ablation) | LAF · MLQA | G-XLT F1: 61.8 -> 61.9 | Helped |
| Add question-similarity loss to PSA with one translated language (ablation) | LAF · MLQA | G-XLT F1: 56.9 -> 56.8 | Hurt |
| Apply MLQA-trained question translation to TyDiQA without translating into TyDiQA languages | mBERT QA · TyDiQA GoldP | XLT F1: 60.8 -> 60.8 | No clear effect |
| Translate questions into TyDiQA languages | mBERT QA · TyDiQA GoldP | XLT F1: 60.8 -> 64.3 | Helped |
| Add language-adversarial training with TyDiQA-language translations | mBERT QA · TyDiQA GoldP | XLT F1: 64.3 -> 65.2 | Helped |
Closest papers by meaning
- Improving the Cross-Lingual Generalisation in Visual Question Answering (AAAI 2023 · similarity 0.89)
- Evaluating the Cross-Lingual Effectiveness of Massively Multilingual Neural Machine Translation (AAAI 2020 · similarity 0.88)
- Cross-Lingual Natural Language Generation via Pre-Training (AAAI 2020 · similarity 0.87)
- Enhancing Non-English Capabilities of English-Centric Large Language Models Through Deep Supervision Fine-Tuning (AAAI 2025 · similarity 0.86)
- Constrained Decoding for Cross-lingual Label Projection (ICLR 2024 · similarity 0.86)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (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.