WMT24++: Source URLs and Document Screenshots for 55 Languages

WMT24++ Images provides source URLs and full-page document screenshots for the translation data used in the WMT24++ project. The dataset, created by Google and last updated on 2025-02 24, preserves original document structure with embedded images. It is intended to support multimodal translation and language understanding research.

Use Cases

• Train multimodal translation models based on the pairing of document screenshots and source URLs.
• Analyze document structure and layout for language understanding tasks using the full-page screenshots.
• Benchmark OCR or visual document processing systems on diverse, real-world document formats.
• Study the impact of embedded images on translation quality and context preservation.

Strengths

• Data preserves original document structure and embedded images, as stated in the description.
• Associated with the WMT24++ project, which expanded language coverage to 55 languages and dialects.
• Source URLs are provided, allowing for potential verification or data expansion.

Limitations

• Row count, file formats, and column-level documentation are unknown.
• Description metadata is limited; actual data quality requires manual inspection after download.
• The license is unspecified, which may restrict commercial or derivative use.

Provenance

Source

Google

Collection Method

Likely gathered as part of the WMT24++ project for machine translation.

Time Range

null

Freshness

Last updated 2025-02-24 23:53:05.

Geography

null