Using Multiple Adaptively-Weighted Strategies for the Resolution of Demonstratives
Abstract
The resolution of demonstratives is an interesting topic which nonetheless has received relatively little attention. As with pronominal anaphora, there is currently no comprehensive theory of demonstrative references, although there are numerous partial or microtheories. This dissertation describes a multistrategy approach to resolving demonstrative pronouns and noun phrases, using constraint, preference, and recovery strategies implementing various microtheories and heuristics. The multistrategy approach, which has proven itself for resolving pro anaphora, allows easy integration of new microtheories. This provides incremental improvements in performance, permitting a decreasing reliance on user intervention as the system is able to resolve a larger percentage of demonstratives. A major weakness of the multistrategy approach, the necessity to carefully choose the weights of the strategy is overcome by automatically optimizing the weights as demonstratives are resolved. Such an adaptively-weighted multistrategy approach has been implemented in the modular and extensible MASTER-D system which is expected to be applicable to other linguistic phenomena in addition to the resolution of demonstratives.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 10, 1993
- Accession Number
- ADA266463
Entities
People
- Ralf D. Brown
Organizations
- Carnegie Mellon University