A reference memory (RM) has been designed and implemented, that allows for storing the results of a shallow parser and finding anaphoric relations between analyzed expressions. In addition, unrecognized expressions can be recognized on the basis of the analyzed ones. The approach aims at being flexible: it does not implement a specific theory, nor does it specialize on a language or a fixed range of anaphoric phenomena. A set of user-specified rules determines which of the entered items are anaphoric, and how antecedents for the anaphors are found. The analyses must be entered as typed feature structures; the rules use conditions on (parts of) those feature structures. No requirements are imposed on the information that is present in the input. The rules are written depending on the kind of input, and should make best use of the information available. The project shows how the RM is used to extend the coverage of a grammar for named entities in German newspaper texts, by combining it with DFKI's shallow processing system SProUT.
|Place of Publication||Enschede|
|Publisher||University of Twente|
|Publication status||Published - 30 Aug 2005|