“President Vows to Cut <Taxes> Hair”: Dataset and Analysis of Creative Text Editing for Humorous Headlines

Abstract

We introduce, release, and analyze a new dataset, called Humicroedit, for research in computational humor. Our publicly available data consists of regular English news headlines paired with versions of the same headlines that contain simple replacement edits designed to make them funny. We carefully curated crowdsourced editors to create funny headlines and judges to score a to a total of 15,095 edited headlines, with five judges per headline. The simple edits, usually just a single word replacement, mean we can apply straightforward analysis techniques to determine what makes our edited headlines humorous. We show how the data support classic theories of humor, such as incongruity, superiority, and setup/punchline. Finally, we develop baseline classifiers that can predict whether or not an edited headline is funny, which is a first step toward automatically generating humorous headlines as an approach to creating topical humor.

Context-Driven Satirical News Generation

“…West and Horvitz (2019) built a corpus of unfunny headlines via a game that asks crowdworkers to make minimal edits that render satirical headlines unfunny and then analyzed structural differences between matched pairs of serious and satirical headlines. Taking an alternative approach, Hossain et al (2019) introduced a corpus of news headlines with one-word edits While both of the aforementioned research efforts make inroads into understanding the rules underlying satire, both of the collected datasets are relatively small and curated. More importantly, both datasets do not consider the broader context that forms the basis of the joke.…”

“…Rather than generation, the emphasis thus far has been on humor classification and ranking. Work by Shahaf et al (2015) built classifiers to rank the funniness of submissions to the New Yorker Magazine caption contest, and Hossain et al (2019) have introduced a headline-editing evaluation task. Raskin (2012) notes that both humor detection and generation research have been hindered by "the difficulty of accessing a context sensitive, computationally based world model," but that "such difficulties are eliminated when the humor analysis is done with a system capable of capturing the semantics of text."…”

Stimulating Creativity with FunLines: A Case Study of Humor Generation in Headlines

“…In fact, FunLines uses one of our humor detectors to give instant feedback when players edit a headline. We compare the FunLines data to Humicroedit, which is a humorous headlines dataset we collected previously from crowd workers (Hossain et al, 2019). FunLines data is less expensive, higher quality, and leads to improved automated humor detection.…”

Can Humor Prediction Datasets be used for Humor Generation? Humorous Headline Generation via Style Transfer

“…This rise in popularity has even translated to a SemEval task of predicting the level of humor in text (Hossain et al, 2019). In their work, as well as others (Mihalcea and Strapparava, 2005;Weller and Seppi, 2019) that seek to understand various aspects of humor, the authors note that their work may be influential in helping create systems that can automatically generate humor.…”

“…Instead current systems rely on retrieveand-edit models (He et al, 2019) or models based on word senses (Luo et al, 2019). The recent works of Hossain et al (2019Hossain et al ( , 2020b) have created pairs of minimal changes that turn a regular news sentence into a humorous news sentence, by only changing one phrase. Because of its popularity and impact, as well as the clear insight that can be gained from minimal pair datasets (Kaushik et al, 2019;Gardner et al, 2020), we choose to examine the former as an initial exploration of what can be done.…”