This repository contains the Aligned Recipe Actions corpus from the below paper. We also provide the code for our alignment model described in said paper.
Aligning Actions Across Recipe Graphs
To appear in the Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP).
Lucia Donatelli, Theresa Schmidt, Debanjali Biswas, Arne Köhn, Fangzhou Zhai, Alexander Koller
Dept. of Language Science and Technology, Saarland Informatics Campus, Saarland University, Germany
Contact Person: Dr. Lucia Donatelli (donatelli@coli.uni-saarland.de)
Apache 2.0
Shield:
This work is licensed under a Creative Commons Attribution 4.0 International License.
Our corpus consists of recipes for 10 different dishes from the Microsoft Research Multimodal Aligned Recipe Corpus. For each dish, there are 11 recipes parsed into action graphs. We provide crowdsourced action alignments between the action phrases of 10 recipe pairings per dish. These alignments indicate which action from the shorter of two recipes corresponds best to an action from the longer recipe.