The various stages involved in a node level classification tasks are

- Pruning : Pruning helps to find out the implausible constituents  that can be the candidates for the arguments of a predicate.

- Identification :- Binary classification helps to find out whether a node will serve as a sematic node.

- Classification :- 1-of-N Classification to predict the semantic role  for a node.

San Jose State University

- The models that use  these algorithms starts by making a parse tree for each input sentence ,which is traversed to identify all the predicates.

- For each predicate the SRL iterates  through each node in the parse tree and perform supervised classification to determine whether the node serves as a semantic  role for the specified predicate and what type of role.

Feature-based algorithm for Semantic Role Labelling

An on-going but a helpful book resource about NLP
https://web.stanford.edu/~jurafsky/slp3/

Speech and Language Processing (3rd ed. draft) 

Node Level Classification Tasks

Semantic roles of arguments are not independent. A more global assignment of labels is required for arguments which can be achieved with the help of  Viterbi decoding or re-ranking or Integer linear programming methods. 

Global Optimization for Semantic Roles

-  governing predicate

-  phrase type of the constituent

- headword of the constituent

-  Part of speech of the head word

- path in the parse tree from the constituent to the predicate

-  whether the voice of the clause is active or passive

- Whether the constituent appears before or after the predicate 

- set of expected arguments for the verb phrase

- Named entity type of the constituent

- first and last words of the constituent

Learn Before

Related