We used the corenlp package (version of june 2014, with default annotators), primarily for dependency analysis.
Recently, I noticed a problem with bracketing a noun compound in cases such as "The Bank of England announced further interest rate increases today"- noun-compound, "interest rate increases"incorrectly copied ( interestparsed as a change increases, not rate). This also takes place when you put this sentence in an online demonstration of the stanford parser and with other similar sentences, where the noun essentially modifies another noun.
One of my colleagues, who does more parsing than me, says that this is probably due to the fact that the model was trained in an uncrossed version of Penn Treebank. And our own own parser, trained on the fixed version, really gets (most) of these nouns. I was wondering if there is an alternative pre-prepared model for the Stanford corenlp analyzer that I don’t know about - and if so, how are we going to start the pipeline using this other model?
source
share