MITW-KYM: A Validated Multimodal Meme Interpretation Dataset

A dataset of 105 selected meme images validated for complex interpretation. Each item was selected by a human researcher and validated using two frontier multimodal LLMs. The dataset focuses on cases where meaning emerges through image-text interaction, pragmatic inference, cultural context, ambiguity, incongruity, or potential false-positive moderation risk.

Use Cases

• Benchmarking multimodal LLMs on meme interpretation tasks based on the described focus on image-text interaction.
• Training content moderation systems to identify potential false-positive risks based on the dataset's stated focus.
• Studying cultural context and pragmatic inference in multimodal communication based on the dataset's selection criteria.
• Analyzing humor mechanisms like ambiguity and incongruity in internet memes based on the described features.

Strengths

• Contains 105 validated meme images, providing a focused collection.
• Each item was human-selected and validated using two frontier multimodal LLMs, suggesting a quality control process.
• Focuses on specific, complex interpretation challenges like cultural context and ambiguity.

Limitations

• Dataset is described as 'small' with only 105 items, which may limit statistical power for training models.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.

Provenance

Source

huggingface user 'sovereigndeveloper'.

Collection Method

Human researcher selection and validation using multimodal LLMs.

Freshness

Last updated 2026-05-11 10:51:31; freshness should be verified.