In my last post, I showed how Inspirobot, an AI designed to create creepy "inspirational" posters, is able to create an uncountably large number of unique phrases by having a relatively small number of multi-part inspirational-seeming phrase structures and being competent at replacing each of the parts with a huge number of (relatively) viable alternatives. Today, I’m going to go into the details of how this can be done.
The main steps are just data collection and enhancement. Quite simply, we want to collect “training” data of as many inspirational phrases as we can easily harvest from the internet and “enhance” each phrase with both part-of-speech tagging and semantic broadening. Fortunately, there are a wide variety of tools that have been developed over the past few years that will do these tasks for us.
My favorite tool for part-of-speech tagging is the Link Grammar Parser originally developed at CMU. For generalization, I would recommend ConceptNet over the less-well-supported WordNet. A Rube Goldberg method would involve automatically translating each phrase into Basic English – but this would probably vastly expand linking similar phrases with different sentence structures in the original.
Even at this point, many of the semantic broadenings will have generated the numerous identical multi-part inspirational-seeming phrase structures necessary -- including the <strive-to><result><time/occasion> or the <causing><entity1><to><verb><entity2><as-relationship><description><entity2><as it is to><entity1> structures from the last post. Making lists of all the expansions of each phrase-part enables the simple creation of new inspirational phrases simply by selecting a new expansion for each phrase-part.
There are many ways in which this can be improved and enhanced – for example, by favoring semantically-related items when choosing from the lists – but the fundamental functionality of Inspirobot has already been generated (and can be automatically added to). If anyone would actually like to do this, please contact me and I would love to help out!
Interesting stuff. Guess I might have to check that out.
This post has received a sweet gift of Dank Amps in the flavor of 25.00 % upvote from @lovejuice thanks to: @ethical-ai. Vote for Aggroed!
@ethical-ai
Great content!
Thanks for sharing!