Skip to content

Research code and scripts used in the paper Semantic Role Labeling as Syntactic Dependency Parsing.

Notifications You must be signed in to change notification settings

bloomberg/emnlp20_depsrl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

srl-dep

Author: Tianze Shi

About this repo

This repo contains the research code and scripts used in the paper Semantic Role Labeling as Syntactic Dependency Parsing. This README file aims at giving basic overviews of the code structure and its major components. For more questions, please directly contact the authors.

Code structure

The entrance point to the package is here. Calls to this package can be chained through fire CLI. The order of calling should usually be build-vocab, create-parser, load-embeddings, train and then finally finish. An example inference script is here using parser.evaluate(data) after loading in models and embeddings.

For official CoNLL evaluation script, access at https://www.cs.upc.edu/~srlconll/soft.html. The F1 scores displayed during model training are NOT official F1 scores (though they are usually very close).

SRL parsing module

The major parsing module is within the python class SRLDepParser inside this file. Back-and-forth conversion algorithms tuned on OntoNotes 5.0 data are contained in this file.

Pre-trained word embeddings

To speed up loading time, we can process the embedding files to trim down to only the vocabulary seen in our data. Script for trimming is here.

Data preparation

Data preparation scripts lie under data_prep folder.

Prerequisite: Stanford CoreNLP with English and Chinese models v3.9.2

  1. Follow http://cemantix.org/data/ontonotes.html to prepare data, using v12 data release
  2. For Chinese data (http://conll.cemantix.org/2012/data.html), copy folders under the correct splits
  3. Use train dev and conll-2012-test splits for English, train dev, test for Chinese
  4. Run aggregate.sh
  5. Run space_to_tab.sh
  6. Run constituency_tree.sh
  7. Run english_dep_tree.sh and english_fuse.py for English data preparation
  8. Run chinese_dep_tree.sh and chinese_fuse.py for Chinese data preparation

About

Research code and scripts used in the paper Semantic Role Labeling as Syntactic Dependency Parsing.

Topics

Resources

Code of conduct

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published