The statistical noun group chunker is an English chunker trained on the
Penn
TreeBank. It identifies noun groups (consisting of a head noun
and its left modifiers) and assigns an ng
annotation to each noun group. The chunker uses a maximum entropy
model whose features are specific lexical tokens and their (Penn)
part-of-speech, obtained from tagger annotations.
If name annotations (ENAMEX) are
present, the chunker will insure that noun groups boundaries do not
appear in the middle of names.