MIT EECS Undergraduate Research and Innovation Scholar
A Natural Language Processing System Based On Sequence- Seeking
Patrick H. Winston
In the course of everyday conversation, humans frequently understand concepts expressed in sentences that are grammatically incorrect and conceptually ambiguous. This suggests that our language faculty is rooted in something more than just grammar. In this paper, I propose that human language operates by aligning low-quality sensory information with expectations of perceived messages. I will demonstrate a natural language processing system based on the process of Sequence- Seeking which will align bottom-up perceptual streams with top-down information from stories that the system has read over the course of its lifetime. I will then demonstrate how context can be implemented as a top-down information stream and how grammar can be implemented as a bottom-up information stream.
I became interested in AI after taking 6.034 with Patrick Winston my sophomore fall. By the end of that class, I decided to join Winston’s Genesis group as a UROP to explore what AI research was about, and I’ve been working in AI ever since. In my SuperUROP project, I hope to learn more about how humans understand language and I’m excited by the idea of an NLP system which functions more similarly to the human brain.