Figure 2

The employed feature set and extracted feature values for a G-M pair: word n-gram features include all word unigrams and bigrams located between G and M; surrounding word features include the two words before the first named entity and the two words after the second named entity; chunk features include inter-GM chunk heads, surrounding chunk heads and inter-GM chunk types; the parse tree path feature is the syntactic path through the parse tree from the first named entity to the second named entity; and sentence position means the relative position of a sentence in an abstract.