How effective are Quantum Entropy Metrics for Detecting Humor?

Anonymous

Abstract: Having ways to measure humor in a text can help provide systems with real-time feedback to enhance their outputs. Thus, we evaluated two Quantum Entropy-based scores as humor quality metrics by comparing their behavior across various corpora in different languages. Results showed significant differences between humorous and non-humorous instances in certain corpora, but an analysis of effect sizes implied minimal practical significance. The metrics' behavior also varies depending on the dataset and language being evaluated. Experiments using a shared multilingual vector space led to more consistent scoring, but also reduced the metric’s ability to differentiate humor and non-humor. Results suggest that despite the potential, the effectiveness of these metrics was inconsistent, indicating a need for further research on humor quality measurements.

Paper Type: short

Research Area: Resources and Evaluation

Contribution Types: Reproduction study, Data analysis

Languages Studied: English, French, Spanish, Portuguese