Benjy's Research Interests

Dissertation Work

My dissertation research was in the area of natural language generation. I developed a system called Kalos that generated parts of a draft users' guide for a microprocessor from a knowledge base describing the microprocessor. I focused on two areas: First, I developed techniques to generate text iteratively using revision processing. Revision processing works similarly to what we as humans do when we produce draft text and iteratively revise it. The application of revision processing to automated text generation helps to reduce the complexity of individual generation modules by postponing some decisions for consideration by revision modules.

Second, I worked on developing a uniform knowledge base for natural language generation with revision. For revision to be applied efficiently, the revision software must be able to make decisions based on all sorts of knowledge. In traditional systems, knowledge for different phases of generation are segregated for efficiency reasons, but in Kalos, I encoded all the system knowledge using a uniform representation language to which a single inference technique could be applied. I also worked on techniques for dealing with the inefficiency of a large, uniform knowledge base.

PostScript copies of my dissertation and an AAAI-94 article are available via anonymous ftp:

Kalos Architecture Overview

Other Research Interests

I'm interested in automating processes that are traditionally difficult to automate, e.g. the automatic generation of machine simulators from abstract descriptions.

I'm interested in AI information retrieval apprentices. One area of particular interest is in the tracking of a user's changing areas of interest over time.

Back to Benjy's home page.