UrcaNet

Source code for Neural Conversational QA: Learning to Reason v.s. Exploiting Patterns (https://arxiv.org/abs/1909.03759)

Prerequisites

Clone repository and install AllenNLP

conda create -n orca python=3.6  
source activate orca  
git clone https://github.com/IBM/UrcaNet.git   
cd UrcaNet  
pip install -r requirements.txt
python -m spacy download en_core_web_md

Download data

wget https://sharc-data.github.io/data/sharc1-official.zip     
unzip sharc1-official.zip

Also download CoQA and QuAC data and put it in coqa and quac folder respectively.

Experiments

Full Task

BiDAF baseline

allennlp train experiments/bidaf_baseline_ft.jsonnet -s ./temp/bidaf_baseline_ft --include-package orca    
python evaluate.py ./temp/bidaf_baseline_ft/model.tar.gz bb_ft

CopyNet on top of BiDAF

allennlp train experiments/bidaf_copynet_ft.jsonnet -s ./temp/bidaf_copynet_ft --include-package orca    
python evaluate.py ./temp/bidaf_copynet_ft/model.tar.gz bc_ft

Question Generation

Change task = 'full' to task = 'qgen' in evaluate.py.

BiDAF baseline

allennlp train experiments/bidaf_baseline.jsonnet -s ./temp/bidaf_baseline --include-package orca    
python evaluate.py ./temp/bidaf_baseline/model.tar.gz bb

CopyNet baseline

allennlp train experiments/copynet_baseline.jsonnet -s ./temp/copynet_baseline --include-package orca    
python evaluate.py ./temp/copynet_baseline/model.tar.gz cb

CopyNet on top of BiDAF

allennlp train experiments/bidaf_copynet.jsonnet -s ./temp/bidaf_copynet --include-package orca    
python evaluate.py ./temp/bidaf_copynet/model.tar.gz bc

CopyNet BiDAF Pipeline

allennlp train experiments/bidaf_copynet_pipeline.jsonnet -s ./temp/bidaf_copynet_pipeline --include-package orca    
python evaluate.py ./temp/bidaf_copynet_pipeline/model.tar.gz bcp