A word replacement game that utilizes the Google Natural Language API and Princeton’s Wordnet database to analyze user-submitted base stories for certain keywords (e.g. adjectives, verbs, and specific types of nouns) and then allows users to replace each keyword to create funny original stories. Users can then personalize each story with a title and a matching animated gif.
Here’s a quick breakdown of how the game works:
In order to identify which words of a given text input we needed to replace, we first needed to understand the word’s role and importance in the sentence. Thus I needed to devise an algorithm that would identify not just a word’s part of speech i.e. noun, verb, adverb, direct object, etc., and salience (i.e. the word entity’s significance to the entire text snippet) using Google’s NLP APIs and also its top-level category (aka synset family) using a JSON file generated from a database compiled by wordnet to generate the replacement prompts.
Combining these two systems allowed us to identify optimal words to replace in order to preserve some semblance of meaning and create more accurate word prompts (i.e. instead of noun, the prompt could be living_thing). This was a decision derived from the hypothesis that generated stories, while mostly nonsensical due to random replacement, are a lot more amusing and enjoyable when they make a little bit of grammatical and logical sense.
Most stories still have a few sentences that don’t quite make sense, but the algorithm tweaks in general gave the user more agency to make relevant word choices which resulted in improved storytelling while maintaining the delight of surprising turns of phrase and irreverent bits of random humor. Try it yourself at copypastereplace.com!