My research can be easily understood by telling you what the goals of the research are; then explaining what we've already done to meet these goals; then what we're doing now; then what we hope to do in the future.

Goals

• Build representations of text content. Example: Represent the content of "A is larger than B." and distinguish it from "B is larger than A." They both have the same words, but different (opposite) meanings.
• Build representations of diagram content. Example: Extract the data values from a figure containing an x,y data graph.
• Build systems that allow users to ask questions about content and get the right items returned. Example: "Is A larger than B?"
• Do the above efficiently and accurately, applied very large collections of documents. Example: We have licensed access to 50,000 articles from the journals of the American Society for Microbiology -- 300 million words of text and 500,000 diagrams.

What we have done so far

We have published many papers on text and diagram analysis. The work described there includes statistical analysis of word meanings, interactive hypertext systems (before HTML was developed), diagram parsing. Here are some examples and comments on the students involved.

• ISMB 94 Knowledge from papers (text -> knowledge representation)
• Digital Libraries 94 - OODBs for text databases
• Digital Libraries Workshop 94 - Corpus linguistics
• ICDAR 95 - Diagram parsing
• The diagram demo site, 1998
• Summarizing diagrams (1999)
• Ambiguity in diagrams (1999)
• Text-diagram relations (2001)
• Machine learning to classify diagrams (submitted, ICDAR 2003)