What does it take to bake a cake? The RecipeRef corpus and anaphora resolution in procedural text

Introduction

This repository contains code and experiment results introduced in the following paper:

• What does it take to bake a cake? The RecipeRef corpus and anaphora resolution in procedural text

• Biaoyan Fang, Timothy Baldwin and Karin Verspoor

• In Finding of ACL 2022

Dataset

• For the data and detailed annotation guideline of the RecipeRef corpus, please refer to RecipeRef dataset. We also provided data in jsonlines format in data but for the original data, please refer to RecipeRef dataset.

Getting Started

• Install python (preference 3) requirement: pip install -r requirements.txt

• Download GloVe embeddings and also another version glove_50_300_2.txt

• Download RecipeRef dataset and put it into the data directory. Note that we separate full set and partition 80.

• run setup_all.sh and then setup_training.sh

• Install brat evalation tool

• We use nltk to tokenzize the brat file for training and generating the jsonlines files. Our code can be found in convert_brat_into_training_format-clear.ipynb

Training Instructions

• Experiment configurations are found in experiments.conf
• Choose an experiment that you would like to run, e.g. bridging
• Training: python train_folds.py <experiment>
• Checkpoints are stored in the logs directory
• Prediction results are stored in the prediction directory

Evaluation

• Evaluation: python evaluate_folds.py <experiment>
• Evaluation tool provides differnet settings, exact and relax mention matching. For this paper, we use exact mention matching