Game of Tropes: Exploring the Placebo Effect in Computational Creativity Tony Veale School of Computer Science and Informatics University College Dublin, Belfield D4, Ireland. Tony.Veale@UCD.ie Abstract Twitter has proven itself a rich and varied source of language data for linguistic analysis. For Twitter is more than a popular new channel for social interaction in language; in many ways it constitutes a whole new genre of text, as users adapt to its new limitations (140 character messages) and to its novel conventions such as retweeting and hash-tagging. But Twitter presents an opportunity of another kind to computationally-minded researchers of language, a generative opportunity to study how algorithmic systems might exploit linguistic tropes to compose novel, concise and re-tweetable texts of their own. This paper evaluates one such system, a Twitterbot named @MetaphorMagnet that packages its own metaphors and ironic observations as pithy tweets. Moreover, we use @MetaphorMagnet, and the idea of Twitterbots more generally, to explore the relationship of linguistic containers to their contents, to understand the extent to which human readers fill these containers with their own meanings, to see meaning in the outputs of generative systems where none was ever intended. We evaluate this placebo effect by asking human raters to judge the comprehensibility, novelty and aptness of texts tweeted by simple and sophisticated Twitterbots. Tropes: Containers of Meaning A mismatch between a container and its contents can often tell us much more than the content itself, as when a person places the ashes of a deceased relative in a coffee can, or sends a brutal death threat in a Hallmark greeting card. The communicative effectiveness of mismatched containers is just one more reason to be skeptical of the Conduit metaphor (Reddy, 1979) – which views linguistic constructs as containers of propositional content to be faithfully shuttled between speaker and hearer – as a realistic model of human communication. Language involves more than the faithful transmission of logical propositions between information-hungry agents, and more effective communication – of attitude, expectation and creative intent – can often be achieved by abusing our linguistic containers of meaning than by treating them with the sincerity that the Conduit metaphor assumes. Consider the case of verbal irony, in which a speaker deliberately chooses containers that are pragmatically ill- suited to the conveyance of their contents. For instance, the advertising container “If you only see one [X] this year, make it this one” assumes that [X] denotes a category of event – such as “romantic comedy” or “movie about superheroes” – with a surfeit of available members for a listener to choose from. When [X] is bound to the phrase “comedy about Anne Frank” or “musical about Nazis”, the container proves too hollow for its content, and the reader is signaled to the presence of playful irony. Though such a film may well be one-of-a-kind, the ill- fitting container suggests there are good reasons for this singularity that do not speak to X’s quality as an artistic event. Yet if carefully chosen, an apparently inappropriate container can communicate a great deal about a speaker’s relationship to the content conveyed within, and as much again about the speaker’s relationship to their audience. As more practical limitations are placed on the form of linguistic containers, the more incentive one has to exploit or abuse containers for creative ends. Consider the use of Twitter as a communicative medium: writers are limited to micro-texts of no more than 140 characters to convey both their meaning and their attitude to this meaning. So each micro-text, or tweet, becomes more than a container of propositional content: each is a brick in a larger edifice that comprises the writer’s online personae and textual aesthetic. Many Twitter users employ irony and metaphor to build this aesthetic and thus build up a loyal audience of followers for their world view. Yet Twitter challenges many of our assumptions about irony and metaphor. Such devices must be carefully modulated if an audience is to perceive a speaker’s meaning in the playful (mis)match of a linguistic container to its contents. Failure to do so can have serious repercussions when one is communicating to thousands of followers at once, with tweets that demand concision and leave little room for nuance. It is thus not unusual for even creative tweets to come packaged with an explicit tag such as #irony, #sarcasm or #metaphor. Metaphor and irony are much-analysed phenomena in social media, but this paper takes a generative approach, to consider the production rather than the analysis of creative linguistic phenomena in the context of a fully- Proceedings of the Sixth International Conference on Computational Creativity June 2015 78