An Algorithm for Irony

by kiminogomi

Writing on the violent global reaction to the post on YouTube of an anti-Islamic video, New York Timestechnology correspondent Somini Gupta comments:

Can the companies that run those speech platforms [YouTube, Twitter, etc.] predict what words and images might set off carnage elsewhere? Whoever builds that algorithm may end up saving lives. (9/23, Sunday Review 4)

As the events surrounding the appearance of this mysterious video demonstrate, how a piece of information is tagged for indexing and retrieval is more than simply a question of research methodology. The video represents itself as a trailer for a film, and as such it sidesteps YouTube’s definition of hate speech, which as Gupta points out is very difficult to define to begin with. In hindsight, it is easy to speculate that the intent of the video was precisely to cause violence and chaos, at a time when political transitions make for increased volatility. We are in an age of cyberwarfare, and propaganda is easily produced and distributed on a global scale. Given this risk, how can technological “speech platforms” identify potentially dangerous materials before they drop their terrible payload? And how can they distinguish creative and critical content from toxic waste?

The problem of classification here is not merely one of locating offensive words, patterns of words, or even phrases; it is rather one of deriving meaning from a complex context. Human mechanisms for producing the effects of intentionality through distance media (print, video, etc.) have developed over the course of the modern era, but the only algorithm so far for detecting these devices is the human mind. And not just any mind — as Engelbart points out, it has to be a trained one. One way to see this process in action is to look at a very sophisticated verbal coding system, one with which many readers have difficulty: irony. I say readers because irony in speech is much more easily conveyed, by tone, facial expression, etc. But how do writers convey irony into the expressively depleted medium of text?

For the purposes of simplicity let’s say that irony derives primarily from a kind of disconnect: something in the text is totally at odds with a deeper level of meaning. In dramatic irony, for example, the reader knows something that the characters don’t, which adds a layer of meaning. So when the hero of Graham Greene’s Ministry of Fear loses his memory, and with it the knowledge that he killed his terminally ill wife, his determination to find out the truth about himself becomes tinged with irony. We know that he wanted to forget his past, and that he will be devastated when he finds out. Another form, which we might call narrative irony, occurs when Dickens refers to a villain as a kindly old gentleman and an innocent victim as a young rogue. Our overall knowledge of the story tells us that Dickens is reversing the ethical polarity of the text, presumably to engage our passions on behalf of the victim and those he or she represents.

In both of the above cases, decoding the ironic mode requires both rhetorical awareness concerning the tone of specific passages and a global comprehension of the text’s modalities. Both Greene and Dickens employ what is best thought of as a satiric mode, that is, a systematic set of tonalities that forces the reader to take a critical perspective on content. Readers trained merely to “relate to” fictional or journalistic narrators and characters, and those who are familiar only with argumentative modes, will often struggle to detect such sophisticated devices, and as a result will equate mentioning something with endorsing it. But readers who learn to read “against the grain,” to think through dissonance and ambiguity, will conceivable be more able to endure the existence of even serious dissent from their own views (thus the observed liberalizing effect of a liberal arts education). Irony and satire require a trained reader, and one whose training includes fictions rich in these modes. Recent changes to US standards for public education, which now emphasize non-fiction literacy, should be reconsidered in the light of this problem: literal-minded readers have no tolerance for irony, or for any representation that does not reflect their version of truth.