| |
Abstract:
A model of predication is described that is based on a
combination of knowledge representation via Latent Semantic
Analysis and the Construction-Integration model of comprehension.
The predication algorithm selects features of a predicate that are
appropriate for the argument of predication and inhibits
inappropriate features. This model accounts for the emergence of
different senses of a predicate, e.g., for two senses of the verb
"run" in "The horse runs" and "The color runs". In the present
paper the model is extended to homonym disambiguation by exploring
whether the same predication algorithm yields correct
interpretations of homonyms or whether additional disambiguation
processes are involved. E.g., if we predicate" are leaves of a
plant" or "is a flavored candy" about "mint," does the predication
algorithm by itself yield correct interpretations.
|
