James Reynolds
December 12, 2011

A program written to process words, undertand them, and produce responses based on that understanding. A user can enter sentences about objects and relationships between them and the program will represent that in a graph structure. The program will then attempt to produce output related to previous input, built on the concepts created in the graph over n inputs.

Concepts Demonstrated

  • ''Reinforcement Learning - to develop a positve or negitive relationship between words.
  • ''Natural Language Processing - to break down and build sentences.


The method for storing the information to create a concept of thought and understanding was an innovative part of this project.

Evaluation of Results

The entire project did not work as expected, mainly because a very sophisticated sentence constructor is needed to build coherent sentences. Input sentences will be broken down and added to the graph appropriately based where positive and negative words are placed within the sentence structure. So the AI will successfully create positive and negative relationships between words as well as updates and builds a more prominent graph with each user input sentence. The current sentence builder replies with incoherent but occasionally logical short sentences at the beginning of a conversation. However, as the graph expands far too many words get related and the output is too focused on one given noun and is garbled.