Research map / Vision-language and multimodal / Visual question answering
Improving the Cross-Lingual Generalisation in Visual Question Answering
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 |
|---|---|---|---|
| Add a similarity-based linguistic-prior loss using WordNet relations (ablation) | UC2 and M3P fine-tuning · xGQA zero-shot, Avg. across target languages | accuracy: +7.99 UC2; +6.9 M3P | Helped |
| Use GloVe embedding distances instead of WordNet relations for the linguistic prior (ablation) | UC2 and M3P fine-tuning · xGQA zero-shot, Avg. across target languages | accuracy: +8.66 UC2; +7.6 M3P | Helped |
| Add task-specific sparse fine-tuning to the embedding-prior method (ablation) | UC2 · xGQA zero-shot, Avg. across target languages | accuracy: +0.55 | Helped |
| Add task-specific sparse fine-tuning to the embedding-prior method (ablation) | M3P · xGQA zero-shot, Avg. across target languages | accuracy: 34.97 -> 34.02 | Hurt |
| Add task-specific sparse fine-tuning to the embedding-prior method (ablation) | UC2 and M3P · xGQA zero-shot, across target languages and random seeds | variance | Helped |
| Add code-mixing on top of embedding-prior loss and sparse fine-tuning (ablation) | UC2 · xGQA zero-shot, Avg. across target languages | accuracy: 38.28 -> 42.19 | Helped |
| Add code-mixing on top of embedding-prior loss and sparse fine-tuning (ablation) | M3P · xGQA zero-shot, Avg. across target languages | accuracy: 34.02 -> 40.00 | Helped |
| Apply code-mixing without sparse fine-tuning (ablation) | UC2 · xGQA zero-shot, Avg. across target languages | accuracy: 37.73 -> 35.16 | Hurt |
| Apply code-mixing without sparse fine-tuning (ablation) | M3P · xGQA zero-shot, Avg. across target languages | accuracy: 34.97 -> 38.80 | Helped |
| Include code-mixing during the pruning stage rather than only during sparse fine-tuning (ablation) | UC2 and M3P · xGQA zero-shot | accuracy | Hurt |
| Count synonym predictions as correct in evaluation (ablation) | UC2 models · xGQA zero-shot, Avg. across target languages | accuracy: +0.89 to +1.85 | Helped |
| Count synonym predictions as correct in evaluation (ablation) | M3P models · xGQA zero-shot, Avg. across target languages | accuracy: +3.42 to +4.56 | Helped |
Closest papers by meaning
- Cross-Lingual Text-Rich Visual Comprehension: An Information Theory Perspective (AAAI 2025 · similarity 0.92)
- Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models (ICLR 2024 · similarity 0.91)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 2024 · similarity 0.91)
- BOK-VQA: Bilingual outside Knowledge-Based Visual Question Answering via Graph Representation Pretraining (AAAI 2024 · similarity 0.90)
- Cross-modal Information Flow in Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- LRM-LLaVA: Overcoming the Modality Gap of Multilingual Large Language-Vision Model for Low-Resource Languages (AAAI 2025 · similarity 0.90)
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.