The subtitle was "Theory for systems; application for theories", and the aim was to find common ground between those who work on defining algorithms and methods and building systems for the acquisition of knowledge from text and those who use knowledge for information-access-related tasks, primarily for understanding text and user queries. The presentations ranged from new theories for computational modeling of brain information processing and applications of various types of multivariate statistical models to text categorization, to name identification across documents and ontology-based search systems.
The discussion centered on representational issues. Representation is central to projects aiming at modeling the human brain, important to projects aiming at information reuse, less important to those who aim for behavioral verisimilitude or computational efficiency, and ad-hoc but preferably handy and inspectable for those aiming at solving some concrete task. The question of how to compare and unify representations was identified as the focus point for future work.
The hopeful aim of the symposium was to find a common task to test methods and representations systematically, in the spirit of the Text Retrieval Conferences organized annually by the US National Institute of Standards and Technology; the consensus at the workshop was that this was too tall an order at this time and that further separate experimentation would be necessary for systems and frameworks to mature to the point where bridging the gap between knowledge acquisition and application would be possible, and where common tasks would provide a fruitful research motivation. A tentative aim is for the participants to meet again in some two years' time, after performing new sets of experiments, to discuss the same or similar questions.