LLM-Generated Norms for 300 English Metaphors Across Five Psycholinguistic Dimensions

Laura Pissani's study investigates the use of large language models to generate psycholinguistic norms for 300 English two-word metaphor combinations. The dataset, last updated in June 2026, contains model-generated ratings for familiarity, aptness, concreteness, metaphoricity, and constituency. It compares norms generated by eight LLMs under different presentation and response formats against human ratings.

Use Cases

• Benchmarking LLM performance on psycholinguistic tasks based on the five normed dimensions.
• Analyzing the relationship between word co-occurrence patterns and metaphor ratings based on the described findings.
• Comparing the effect of stimulus presentation (in-context vs. isolation) on model-generated norms.
• Replicating analyses of metaphor comprehension using automated norm generation.

Strengths

• Focuses on 300 specific English metaphor combinations, providing a defined scope.
• Evaluates norms across five distinct psycholinguistic dimensions: familiarity, aptness, concreteness, metaphoricity, and constituency.
• Compares outputs from eight different large language models.
• Uses a Creative Commons CC-BY-4.0 license for open sharing.

Limitations

• Dataset scale is limited to 145.4 KB, suggesting a small number of records.
• Column-level documentation is absent; field semantics must be inferred from the description.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Laura Pissani

Collection Method

Generated by prompting eight large language models to norm metaphor combinations.

Freshness

Last updated 2026-06-04 04:27:29; freshness should be verified.