Research map / Label-efficient and robust learning / Recognition and long-tail
Translation-Based Matching Adversarial Network for Cross-Lingual Natural Language Inference
Kunxun Qi, Jianfeng Du
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 |
|---|---|---|---|
| add TMAN framework (ablation) | ESIM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 67.3 -> 69.0 | Improved |
| add domain adversarial network (ablation) | ESIM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 67.9 -> 69.0 | Improved |
| add TMAN framework | multilingual BERT · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 72.0 -> 73.7 | Improved |
| add TMAN framework | XLM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 76.7 -> 77.2 | Improved |
| add adversarial network (ablation) | XLM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 76.1 -> 77.2 | Improved |
| mix 2 languages in a mini-batch (ablation) | TMAN-XLM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 76.9 -> 77.2 | Improved |
| add transformer encoder (ablation) | TMAN-XLM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 76.9 -> 77.2 | Improved |
| fixed cross-lingual word embeddings instead of ones used by TMAN-ESIM (ablation) | TMAN-ESIM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 69.0 -> 68.7 | Hurt |
| updatable monolingual word embeddings (ablation) | TMAN-ESIM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 69.0 -> 68.0 | Hurt |
| fixed monolingual word embeddings (ablation) | TMAN-ESIM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 69.0 -> 67.7 | Hurt |
| random word embeddings (ablation) | TMAN-ESIM · XNLI test, TRANSLATE-TRAIN, 15-language average | accuracy (%): 69.0 -> 62.0 | Hurt |
| TMAN-XLM vs XLM on German | TMAN-XLM · XNLI test, TRANSLATE-TRAIN, de | accuracy (%): 80.3 -> 80.2 | Hurt |
Closest papers by meaning
- Improving Neural Language Modeling via Adversarial Training (ICML 2019 · similarity 0.88)
- ABSent: Cross-Lingual Sentence Representation Mapping with Bidirectional GANs (AAAI 2020 · similarity 0.88)
- Cross-lingual Language Model Pretraining (NeurIPS 2019 · similarity 0.88)
- Comparing Unsupervised Word Translation Methods Step by Step (NeurIPS 2019 · similarity 0.88)
- Unsupervised Interlingual Semantic Representations from Sentence Embeddings for Zero-Shot Cross-Lingual Transfer (AAAI 2020 · similarity 0.87)
- Constrained Decoding for Cross-lingual Label Projection (ICLR 2024 · similarity 0.86)
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.
