Implements a 'semantic head' variant of the the HeadFinder found in Michael Collins' 1999 thesis. This version chooses the semantic head verb rather than the verb form for cases with verbs. And it makes similar themed changes to other categories: e.g., in question phrases, like "Which Brazilian game", the head is made "game" not "Which" as in common PTB head rules.
By default the SemanticHeadFinder uses a treatment of copula where the complement of the copula is taken as the head. That is, a sentence like "Bill is big" will be analyzed as
nsubj
(big, Bill)
cop
(big, is)
This analysis is used for questions and declaratives for adjective complements and declarative nominal complements. However Wh-sentences with nominal complements do not receive this treatment. "Who is the president?" is analyzed with "the president" as nsubj and "who" as "attr" of the copula:
nsubj
(is, president)
attr
(is, Who)
(Such nominal copula sentences are complex: arguably, depending on the circumstances, several analyses are possible, with either the overt NP able to be any of the subject, the predicate, or one of two referential entities connected by an equational copula. These uses aren't differentiated.)
Existential sentences are treated as follows:
"There is a man"
expl
(is, There)
det
(man-4, a-3)
nsubj
(is-2, man-4)
@author John Rappaport
@author Marie-Catherine de Marneffe
@author Anna Rafferty