A trainable system for the extraction of meaning from text

This project is developing a trainable system that can extract meaning from texts in different domains (example: various Internet news-groups). The system does partial parsing based on a large dictionary containing approximately 150,000 words. The system assists the user in extracting a semantic network representation for each member of a set of training articles contained in some large database. Based on the user's training, the system forms statistical tables, a knowledge base, and a set of rules mirroring the user's actions. The system then generalizes these rules. Using statistically based semantic classification, the system applies these rules to new articles from the database for automatically building semantic networks.