Scaffolding: Syntactic Scaffolds for Semantic Structures

A repository based on the EMNLP 2018 paper for Frame-Semantic and PropBank Semantic Role Labeling with Syntactic Scaffolding.

Installation

This repository was built on an earlier version of AllenNLP. Due to changes in the API, we recommended installing directly via steps below (adapted from the AllenNLP installation), as opposed to using an installed version of AllenNLP.

Download and install Conda.
Create a Conda environment with Python 3.6
```
conda create -n scaffold python=3.6
```
Activate the Conda environment. (You will need to activate the Conda environment in each terminal in which you want to use AllenNLP.
```
source activate scaffold
```

Install in your environment

git clone https://github.com/swabhs/scaffolding.git

Change your directory to where you cloned the files:
```
cd scaffolding/
```

Install the required dependencies.

INSTALL_TEST_REQUIREMENTS="true" ./scripts/install_requirements.sh

Install PyTorch version 0.3 via pip (modify based on your CUDA environment). Conda-based installation results in slower rutime because of a CUDA issue.
```
pip install http://download.pytorch.org/whl/cu80/torch-0.3.0.post4-cp36-cp36m-linux_x86_64.whl
```

Step 1: Get Data

Download FN data and place it under a data/ directory under the root directory.

The OntoNotes datasets can be obtained through LDC.

Step 2: Get Pre-trained Models

Frame-SRL baseline

Frame-SRL scaffold with common non-terminals

PropBank SRL baseline

PropBank SRL scaffold with common non-terminals

Step 3: Test

python -m allennlp.run evaluate \
    --archive-file log_final_pb_baseline/model.tar.gz \
    --evaluation-data-file data/fndata-1.5/test/fulltext/ \
    --cuda-device 0

For the syntactic scaffold model for PropBank SRL, use the pbscaf branch:

git checkout pbscaf

and then run evaluation as above.

Training

For scaffolds, use $command=train_m and for baselines, $command=train.

python -m allennlp.run $command training_config/$config --serialization-dir log

Acknowledgment

This is based on our paper, please cite:

@inproceedings{swayamdipta:2018,
    author      = {Swabha Swayamdipta and Sam Thomson and Kenton Lee and Luke Zettlemoyer and Chris Dyer and Noah A. Smith},
    title       = {Syntactic Scaffolding for Semantic Structures},
    booktitle   = {Proc. of EMNLP},
    url         = {https://arxiv.org/abs/1808.10485},
    year        = {2018}
}