Filling the Blanks (hint: plural noun) for Mad Libs Humor

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 638–647 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Filling the Blanks (hint: plural noun) for Mad Libs R⃝ Humor Nabil Hossain∗, John Krumm†, Lucy Vanderwende†, Eric Horvitz† and Henry Kautz∗ ∗Department of Computer Science †Microsoft Research AI University of Rochester Redmond, Washington {nhossain,kautz}@cs.rochester.edu {jckrumm,lucyv,horvitz}@microsoft.com Abstract Computerized generation of humor is a notoriously difficult AI problem. We de- velop an algorithm called Libitum that helps humans generate humor in a Mad Lib R⃝, which is a popular fill-in-the-blank game. The algorithm is based on a ma- chine learned classifier that determines whether a potential fill-in word is funny in the context of the Mad Lib story. We use Amazon Mechanical Turk to create ground truth data and to judge humor for our classifier to mimic, and we make this data freely available. Our testing shows that Libitum successfully aids humans in filling in Mad Libs that are usually judged funnier than those filled in by humans with no computerized help. We go on to analyze why some words are better than others at making a Mad Lib funny. 1 Introduction As technologists attempt to build more natu- ral human-computer interfaces, the inclusion of computer-generated humor becomes more impor- tant for creating personable interactions. However, computational humor remains a long-standing challenge in AI. Despite decades devoted to the- ories and algorithms for humor, the best comput- erized humor is still mediocre compared to hu- mans. Humor requires creativity, sophistication of language, world knowledge, empathy and cogni- tive mechanisms, which are extremely difficult to model theoretically. A more modest goal for com- putational humor is to build machines that help hu- mans create humor rather than replace them. We develop and test an algorithm, called Li- Figure 1: Example of a Mad Lib sentence. The original words for the blanks, in order, were “apartment”, “impatient” and “gallery”. bitum1, for a computer-aided approach to humor generation. Its aim is to help a human player fill in the blanks of a Mad Lib R⃝ story to make it funny. The algorithm generates candidate words, and its core component is a machine-trained clas- sifier that can assess whether a potential fill-in- the-blank word is funny, based on several fea- tures, including the blank’s surrounding context. We trained the classifier on Mad Lib stories that were filled in and judged by humans. We note that in our work, we give players, both human and computer, access to the full Mad Lib story, includ- ing the sentences surrounding the blanks. In regu- lar Mad Libs, players do not see the surrounding sentences. Figure 1 shows a sentence from a typi- cal Mad Lib, completed by a human player. The work presented here makes three contri- butions. The first is the creation of a challenging benchmark for humor generation, a vital aspect of human communication which has received rela- tively little attention in the NLP community. This benchmark, based on Mad Libs, (i) …

Content truncated to 500 words