stanza
stanza copied to clipboard
stanfordnlp
Can't get a binary tree using stanza.pipeline
Can't get a binary constituent tree using stanza.pipeline where constituency is already included.
Right, we don't have a concept of headfinder, so we don't really have a good way to binarize the trees. What do you need this for?
Actually I tried to use Tree-LSTM on my own data, but the number of child of each node is uncertain..
You can try using a left binarization rule, or a right binarization rule, or switching the tree lstm mechanism to use the average or max of the inputs
now i'm wondering if maybe a "binarize" operation in the tsurgeon utility would make sense could have left & right & known headfinder algorithm as options.
or use a dependency treebank to learn how to make a headfinder over constituencies predicting which is the head might make parsing more accurate
Well, actually I first use stanza to parse the constituency, then I use nltk library to build up an instance of nltk.tree, then I use nltk transform tools to convert the tree into its "chomsky_normal_form", done.
But still some of the nodes only has one children, since I am on a time sensitive project, I currently don’t have enough time to study it in detail.
unary transitions (nodes with one child) happen in certain structures where the sentence would look more complete if an implied word is added. S over VP with a missing NP, for example
On Thu, May 5, 2022, 7:07 AM Roman @.***> wrote:
Sorry for such a late reply.
— Reply to this email directly, view it on GitHub https://github.com/stanfordnlp/stanza/issues/1006#issuecomment-1118599572, or unsubscribe https://github.com/notifications/unsubscribe-auth/AA2AYWPWSNQB3Z6LYGWM4DLVIPIZRANCNFSM5T2TPQ7A . You are receiving this because you modified the open/close state.Message ID: @.***>