A Domain Independent Event Extraction Toolset

In this task, we seek to assess the technical feasibility of developing tools for automatic information extraction from electronic documents. Our focus is on the development of a domain independent, calendar event extraction tool. Regardless of a document's format, or the domain it represents (e.g., business news, community newspaper), it is possible to extract key data such as name, location/address, date, organization name, etc. and represent it in a searchable database. Although there is a significant body of research in this area, existing technology needs to be made more reliable and portable to different domains in order to be used in commercial applications. Performance evaluation of current technology and the target technology will be researched. The principal research areas include (i) domain independent techniques for grouping words into names, date/time, and subsequent classification as person name, location, organization name, etc., (ii) domain dependent techniques for resolving ambiguities in the previous step, extracting event type information, and (iii) an automatic/semi automatic learning procedure for domain dependent patterns such that a non expert user can specify these patterns easily. The final deliverable is a system design, complete with algorithmic specification for all modules, user interfaces, performance specifications, etc. Preliminary testing results will be included.