Reducing False Positives in the Construction of Adjective Scales
Alice Zhang · 2014
Many adjectives that appear to be synonyms of one another differ in their intensity.Distinguishing the nuances between adjective synonyms is vital to linguistic understanding of a language, but WordNet currently does not encode the relative intensities of adjective synonyms that lie on the scale.Sheinman & Tokunaga (2009) proposed a solution of constructing Adjective Scales by data mining a web corpus.However, this process suffers from some limitations, most notably that of False Positives, which inaccurately suggest that adjective X is more or less intense than Y.This paper classifies the types of false positives that Sheinman's method generates, then proposes a method to diminish the quantity of these false positives using linguistic searches in WordNet.