How we are solving the problems - work in progress

We are currently supported by two NSF grants (National Science Foundation), one for text (natural language understanding) and the other for diagrams.

• Download the 300 million words and store them in Oracle in a novel sequence format using "extreme tokenization". Example: "1.2" is "1", ".", "2". Everything based on UIDs, no strings involved in most of the work. Take full advantage of DB indexing, Java hash tables in 8GB of RAM.
• Look for "standard containers" in text. Example: "Tomorrow's high temperature will be 31 degrees." The "31" has high information content (ala Shannon). The rest is a "container". Map queries to containers to produce answers for users. Example: "What will the high be tomorrow?"
• Use the rare PDF files with vector diagrams to investigate vector issues. Example: Render vector to spatial index pyramid (SPAS) and use object statistics to train a Support Vector Machine classifier (95% accuracy by leave-on-out measure).
• Build a vectorization system that turns raster diagrams into vector diagrams. Current systems ignore Moore's Law. Example: We represent a megapixel image by an array of one million Java objects, not integers. High-quality, model-based approach, the Strategic Vectorization Project. Uses Swing visualization.
• Build a parser generator to solve the constraint-based diagram parsing problem. Design semantic interpreters that walk the parse tree and build meaning compositionally.