AGRR-2019 Gapping Resolver

This is a system submitted to Dialog Evaluation 2019 gapping resolution track.

Description

This system uses an AWD-LSTM encoder to create sentence and token representations and runs an MLP classifier and a linear decoder to find sentences with gapping and attempt to resolve it, respectively.

Metrics

Corpus	Binary	Resolution
Train	TBD	TBD
Dev	TBD	TBD
Test	TBD	TBD

Usage

To run the experiment:

1. Install the requirements from requirements.txt.
2. Fetch folders artifacts and data from Google Drive and put it in the root folder of the repostiroty
3. Run python resolver.py input_file.csv output_file.csv. The input file should follow the AGRR format.