Multilingual Wikipedia Image-Text Associations

Comprising over 37 million image-text associations extracted from Wikipedia articles. It is a multilingual dataset covering 108 languages, curated by Google Research and released by Wikimedia.

Use Cases

• Train multimodal models on image-text associations for cross-modal retrieval tasks.
• Develop multilingual image captioning systems using text from 108 different languages.
• Analyze the distribution of visual content across Wikipedia articles in various languages.

Strengths

• Over 37 million image-text pairs provide a large-scale resource.
• Multilingual coverage spans 108 languages.
• Curated from Wikipedia, a well-known and structured source.

Limitations

• Specific column structure and data format are not detailed in the input.
• The sample data is unavailable for preliminary inspection.
• The dataset's last known update was in November 2022, which may limit recency.

Provenance

Source

Google Research, released by Wikimedia

Collection Method

Extracted from Wikipedia articles

Freshness

2022-11-04