"Parsing" with gold segmentation
jodaiber opened this issue · comments
Hi @dlwh,
first off, thanks for making Epic available, it's a great tool!
I have a "parsing" problem where I would like to restrict the search space to a fully segmented binarized tree and would like to use the neural parser to only do the labelling for me (the trees come from a customized treebank that I pass to epic for training). I have been trying to do this via constraints (using GoldConstraintsFactory) in NeuralParserTrainer but so far without success. Is there any established/recommended way for using "gold spans" and only letting Epic do the labelling?
Jo
A question I can answer in the affirmative!
Look at SentimentTreebankPipeline.scala which uses the Sentiment Treebank for sentiment classification. That assumes a fixed tree structure, as well.
It's not the best documented, but it will probably get you very close to what you want.