Understanding English with lattice-learning

Abstract

A computer program that can understand the meaning of written English must be tremendously complex. It would break the spirit of any programmer to try to code such a program by hand; the range of meaning we can express in natural language is far too broad, too nuanced, too filled with exception. So I present UNDERSTAND, a program you can teach by example. Learning by example is an engineering expedient: it is much easier for us to come up with specific examples of a concept than some sort of perfect Platonic model. UNDERSTAND uses a technique I call Lattice-Learning to generalize accurately from just a few examples: "Robins, bees and helicopters can fly, but cats, worms and boats cannot," is enough for UNDERSTAND to narrow in on our concept of flying things: birds, insects and aircraft. It takes only 8 positive and 4 negative examples to teach UNDERSTAND how to interpret sentences as complicated as "The cat ran from the yard because a dog appeared." UNDERSTAND is implemented in 2300 lines of Java.