HUMOR: A Crowd-Annotated Spanish Corpus for Humor Analysis

Computational Humor, as its name indicates, studies humor from a computational perspective, and it fosters several tasks, such as humor recognition, humor generation and humor scoring. The area has been little explored, making it attractive to be tackled by novel Natural Language Processing and Machine Learning techniques. For this to be possible, human-curated data is necessary. In this work we present a corpus of almost 40,000 tweets written in Spanish and crowd-annotated by their humor and funniness value by several people on the Internet. It is equally divided between tweets coming from humorous and non-humorous accounts. There is certain humor value agreement between the raters, with a Krippendorff’s alpha value of 0.3654. However, it does not show a clear agreement for the funniness value. The dataset is available for general usage and has already been used successfully for humor recognition. Additionally, other aspects of the dataset are analyzed, such as the distribution by the number of annotations and the categories.