Consider these words you are reading, arbitrary collections of sound whose connection to the actual experiences they evoke is tenuous at best. Yet working from that huge, interconnected jumble of experiences juxtaposed with bits of language that you have had over many years of life, you have been able to learn distinct words, rules of grammar, and systems for effectively communicating with language in endless novel situations. Trying to reproduce this feat of richness and flexibility, many artificial intelligence researchers today are considering neural networks, modeled after the human mind. But neural networks struggle to generalize grammatical rules and structure out of the webs of linguistic information they are trained with. I propose that artificial neural networks can face this challenge and better learn and process the systematic grammar of natural languages by drawing on the rich structure of the real world in perceptual information, because the structure we discover in our perceptual experiences is what we use during our human language learning process to build up systematic concepts around the new words we encounter.
That’s the beginning of this final thesis defense paper that I wrote for an Intro to Philosophy class in 2017. It explores some ways that computer scientists and linguists are trying to model the human mind in a computer so it can understand language, and the challenges these models have in capturing the full flexibility and structure of language. Then it outlines a new philosophical approach, where I try to combine a naturalistic view of language as useful labels for perception with a Christian dualistic view, where language can access absolute truth. Maybe computers can discover some of the structure of language by exploring the structure of the real world around us.
It’s 14 pages long and I’m trying to mash in lots of concepts in that suddenly small space, but it’s also meant to be something that any curious educated person could read and learn a few things about language, philosophy, and computer science from. If you’re up for my philosophizing, you can read a PDF of it here.