Decomposition and Distribution of Humorous Effect in Interactive Systems

We aim to identify and control unintentional humor occurring in human-computer interaction, and recreate it intentionally. In this research we focus on text prediction systems, a type of interactive programs employed in mobile phones, search engines, and word processors. More specifically, we identified two design principles, inspired by humor and emotion theories, and implemented them in a proof-of-concept tool simulating a specific type of text prediction. Computer programs do not always work as well as we would like. Even though they are generally designed to respond efficiently to users’ needs, they sometimes make foolish mistakes that make them look less smart than expected. In some cases these events are unexpectedly funny. Our aim is to identify and control unintentional humor occurring in human-computer interaction, and recreate it intentionally. One major problem in computational humor generation is the large amount of knowledge required to model people’s expectations and violate them for achieving humorous effects. We address this issue in the context of humancomputer interaction, where the required knowledge can be limited to the concepts and actions on which the user is currently focused. There are several widely used tools, such as text editors or Web browsers, familiar enough that users have typical expectations about their behavior. In this research we are specifically focused on text prediction systems, a type of interactive programs employed in mobile phones, search engines, and word processors. Two commonly used text prediction systems are autocomplete and autocorrect. These systems are particularly interesting for exploring forms of humor generation through the controlled induction of surprise effects. As their name suggests, they are characterized by the capability to predict user’s intentions about actions or information. For example, the autocomplete feature allows the user to input a text (e.g. a query to a search engine) typing only the first letters. If more completions are possible for the same prefix, the system predicts the most probable one, thus performing a form of expectation modeling. The humorous counterpart of a normal text prediction Copyright c © 2012, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. system should be able to produce outputs violating the user’s expectations in a funny way (see example in Figure 1). We propose a way for designing humor in text prediction systems, as a first step for the study of more general classes of interactive programs. This approach is based on two assumptions inspired by humor and emotion theories: • Decomposition of humorous effect. The humorous effect can be achieved through the separate treatment of surprise and funniness. In other words, different strategies and resources can be employed separately for inducing surprise and for making it funny. • Distribution of humorous effect. The repetition of surprising and semantically related stimuli can be potentially humorous. What can make surprise funny is not the meaning of the single event but its repeated occurrence. We assume the semantically related repetition could be used as a way to simulate a sort of idiosyncratic attachment to the prefixed topic and, thus, make the program’s behavior recognizable as ridiculous. The above claims were adopted as design principles and implemented in a proof-of-concept text predictor system. At this stage of research, we focus on the control of the repetition of semantically related stimuli as a preliminary step for the investigation of more complex forms of interactive humor based on surprise effect. The rest of the paper is organized as follows. The next section provides a brief introduction to the proposed approach, discussing some of the background ideas inspiring it. Next, the proposed design principles are introduced. Their application in a proof-of-concept prototype are then described. Finally, some of the next steps of our research are discussed.