You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the decoding stage, an unordered tree is given and
the goal is to lift edges that would be non-projective
with respect to the gold linear order. Similarly to
how training instances are derived, the decoding al-
gorithm traverses the tree bottom-up and visits every
node once. Ties between nodes at the same depth are
broken in an arbitrary but deterministic way. When
a node is visited, the classifier is applied and the cor-
responding lifting is executed. Pseudocode is given
in Algorithm 1.
How this is done is not described, but can be discussed in this issue.
The text was updated successfully, but these errors were encountered:
See Algorithm 1 in this paper
How this is done is not described, but can be discussed in this issue.
The text was updated successfully, but these errors were encountered: