University of Amsterdam at the CLEF 2025 JOKER Track

This paper reports on the University of Amsterdam’s participation in the CLEF 2025 JOKER track. Our overall goal is to investigate the non-literal use of language, such as in humor and wordplay, that remains challenging for current information retrieval and natural language processing technologies. Our specific focus is on a post hoc approach where we exploit wordplay or humorous text detection as a means to filter out humorous translations or search results. Our main findings are the following. First, we successfully developed effective humor detection classifiers for both English and French. Second, for humor-aware information retrieval, we could increase retrieval effectiveness by filtering for humorous content in search results ranked solely on topical relevance. Third, for wordplay translation, we manage to generate multiple translation candidates and select the one with the highest pun score based on the detector. This approach performs well, with limited gain on the automatic evaluation measures, but qualitative analysis confirms that this is an encouraging strategy.