Selection process for the "30 candidate words" used in LAION labeling

[hyeongtakji]

Hello,

I apologize if this question seems out of context for GitHub issues.

I am writing with a question regarding the process described in your paper for assigning labels to LAION dataset vectors. Specifically, the paper states:

We assign each image embedding its keyword list by taking the 3 words with highest text-to-image CLIP scores from a candidate list of 30 common adjectives and nouns (e.g., 'animal', 'scary').

However, it is not quite clear how these 30 candidate words were selected. Can you provide additional insight into the criteria or process used to choose this candidate list?