Rule-based reference resolution for unrestricted text using part-of- speech tagging and noun phrase parsing
Sandra Williams, Mark H. Harvey · 2003
This paper describes an experimental syntactic rule-based method for reference resolution in unrestricted texts. References can be resolved automatically and this overcomes a major hurdle in text analysis and provides a key advantage in text `understanding' and information extraction. A shortcoming of systems that locate and extract sentences from unrestricted text to help people assimilate information, and cope with `information overload', arises from references to sentences not present in the extract. Reference resolution is a technology that can be used in such systems and one of the motivations for the current work was to investigate ways of identifying and resolving such references, thus allowing us to enhance the performance and usability of these systems. The reference resolution system accomplishes two tasks: firstly it identifies noun phrases, either as references or as non-references, and secondly it resolves references. The identification of reference/non-reference noun phrases is achieved in two stages: 1) part-of-speech tagging; 2) noun-phrase parsing. The reference resolution is achieved with rules and knowledge-bases of names, titles, and General Knowledge. Tests results show that up to 97% of reference and non-reference noun phrases are correctly identified; and up to 76% of all references that the system attempts are resolved correctly within the first three estimates. Of these, up to 61% are resolved correctly on the first estimate.