The paper describes a language-independent approach for semiautomatic extension of lexical-semantic word nets and evaluates the method on CoreNet, the Korean version of word net. In a bootstrapping fashion, the so-called 'Pendulum Algorithm' operates on word sets obtained by co-occurrence statistics on a large un-annotated corpus and keeps error propagation low by a verification step. Results are not sufficient for automatic extension, but provide a good candidate set. Further improvements are discussed.
展开▼
