Towards a semantic extraction of named entities

In this paper, we discuss the new challenges posed by the progression from information extraction to content extraction, as demonstrated by the ACE program. We explore whether traditional IE approaches are sufficient, and describe the adaptation of a generic IE system to this kind of application. Results suggest that a deeper level of processing is necessary to achieve excellent results in all areas, although rule-based systems can still produce results of a reasonable quality with a small amount of adaptation. In particular, the task of entity detection and tracking on texts of varying genre and quality is one of the most challenging.