CAPRI: Cultural and Pragmatic Response Inference Dataset

CAPRI is a multimodal dataset for studying whether large language models act as pragmatic speakers by tailoring answers to a user's perceived cultural background. It was created by yisongmiao and last updated on June 17, 2026. Each item is a short conversation with varying cultural cues, followed by a visual question about an image.

Use Cases

• Benchmarking LLM cultural sensitivity based on conversational cues described in the dataset.
• Training models for pragmatic response inference based on multimodal context.
• Studying the interaction between visual question answering and cultural background inference.
• Analyzing how conversational history influences model outputs in a multimodal setting.

Strengths

• Dataset is explicitly designed for a specific research problem: cultural and pragmatic response inference.
• Data items combine conversation and visual questions, providing a multimodal structure.
• Last updated on 2026-06-17, suggesting recent maintenance.

Limitations

• Row count, file formats, and column-level documentation are unknown, which limits suitability assessment.
• License information is absent, requiring verification before use.
• The dataset's scale and specific cultural backgrounds covered are not detailed in the provided metadata.

Provenance

Source

huggingface

Collection Method

Likely constructed for research purposes, involving the creation of conversational items with cultural cues and visual questions.

Freshness

Last updated 2026-06-17 14:03:59