Make use of producer-for-product, etc. patterns where applicable. Prefer interpretations that fit a known pattern to ones that would require on-the-fly creation of a new pattern
Aptness conditions
Which characteristics of an entity will a speaker most likely choose as metonyms? Visible over unseen, context relevant over context irrelevant, short over long, etc.
Components of Markert and Hahn’s metonymy resolution system
KL-ONE knowledge representation system with information about the hierarchical and lateral relations between entities in an information technology domain. Defines relationship duals such as “has-part”, “part-of”, etc.
“Path Classifier” component and set of interpretation conditions for obtaining and ranking literal and metonymic interpretations
Path Finder
Finds all well-formed paths between concepts that correspond to conceptual interpretations of the syntactic relationship between the referents of the concepts. Uses these definitions:
conceptual relatedness (of two concepts)
connectivity (of a series of relations)
cyclicity (of a connective path)
A path is cyclic iff it contains two non-identical relations that are inverses of each other or have an inverse super-relation.
Ex. (Printer, Printer-of, Computer-system, Has-Central-Unit Central-Unit) is cyclic because Printer-ofis a sub-relation of Part-of and Has-central-unit is a sub-relation of Has-part
Uses predefined path patterns to separate paths into three groups: literal paths, metonymic paths and unclassified paths (not currently able to construct new patterns such as food-for-customer).
Performs metonymic and nominal anaphora resolution in tandem.
Identifies anaphora and prefers interpretations that resolve anaphoric reference.
Ranks interpretations according to degree of “referential cohesion”, in which literal, metonymic and anaphoric references are resolved most satisfactorily.
