Political Speeches & AI

Διαβάστε το στα ελληνικά εδώ.

At a time when the USA moves full speed ahead towards the presidential election, and the candidates fight and make speeches, everyone else, all around the world, battle against each other on Facebook, on Twitter and most of them argue (sic) as well.

Political speeches from Antiquity till today have been, for me, one of the most exciting, tough and demanding, provocative and balancing acts in the world. Because in order to deliver them, apart from a vision and a plan (I consider them to be parts of the content, but sadly they are optional) one needs tools, like the proper command of the language, the knowledge base, the arguments constructed and presented in the past conserving the subject they are referring to.

In 2016 Valentin Kassarnig (College of Information & Computer Science, University of Massachusetts Amherst) presented a “Political Speech Generator”, a system of Artificial Intelligence able to create partisan political speeches. Based on data derived from 3857 speeches made in Congress since 2005 the creators trained the system initially to detect whether a speech is pro or against a proposed bill. The speeches were analyzed individually, but, more importantly, also in comparison to the argumentation of the rest of the speakers. This way the system allows you to determine whether any given speech should present a supportive or not opinion.

This system is based on a combination of different advanced methods. The lexical sequences are created based on the probabilities derived from two underlying models: a linguistic model that ensures the grammatical accuracy and a subject model that aims at the cohesiveness of the text.

In his report (which you can find here, in case you’re interested) Kassarning describes this method and other parameters used in the project at length, the results of which are high quality speeches, both grammatically and in terms of transitioning from one concept to another.

To be honest, we, the people, have already long ago discovered our Political Speech Generator, which is none other than: “Unmoving Faith to Personal Dogma”. It mainly works by starting directly from the conclusion and working all the way backwards towards it, depending on what or whom we want to support. And now Technology comes to prove that it can do even that, only better!

But, political speech is not (just) partisan speech. From the times of ancient rhetoricians until the emblematic speeches of the last century, it has kept for itself a role much more important and valuable: as catalyst for the social contract’s constant renewal.