Multi-Task Learning of Query Generation and Classification for Generative Conversational Question Rewriting

This repository contains the code and datasets for the EMNLP 2023 paper "Multi-Task Learning of Query Generation and Classification for Generative Conversational Question Rewriting". The paper proposes a novel multi-task learning approach to rewrite ambiguous conversational questions into well-defined queries while also identifying topic continuity. The models, based on BART and T5 architectures, demonstrate significant performance improvements over single-task baselines.

Install dependencies:

git clone
cd mtl_gen_class
pip install -r requirements.txt

Data Preparation

By default, we expect raw and processed data to be stored in ./data/ :

OR-QuAC

OR-QuAC files download

Download necessary OR-QuAC files and store them into ./data/or-quac:

mkdir data/or-quac
cd data/or-quac
wget https://ciir.cs.umass.edu/downloads/ORConvQA/all_blocks.txt.gz
wget https://ciir.cs.umass.edu/downloads/ORConvQA/qrels.txt.gz
gzip -d *.txt.gz
mkdir preprocessed
cd preprocessed
wget https://ciir.cs.umass.edu/downloads/ORConvQA/preprocessed/train.txt
wget https://ciir.cs.umass.edu/downloads/ORConvQA/preprocessed/test.txt
wget https://ciir.cs.umass.edu/downloads/ORConvQA/preprocessed/dev.txt

Baselines

To compare our results with the baseline, you can use the models and datasets provided at LIF GitHub Repository.

Downloading the Baseline

1. Navigate to LIF GitHub Repository.
2. Follow their installation and setup instructions to download the models and datasets.
3. Optionally, you can directly clone their repository using the following command:

git clone https://github.com/nusnlp/LIF.git

Using the Baseline for Comparison

Once you've downloaded the LIF baseline, follow their instructions to run the model and obtain results.

Then, you can use these results for a direct comparison with our Multi-Task Learning models.