Overview

This repository contains the source code for the models used for BLCU-ICALL team's submission for Task 1: CODWOE - COmparing Dictionaries and WOrd Embeddings. The model is described in the paper "BLCU-ICALL at SemEval-2022 Task 1: Cross-Attention Multitasking Framework for Definition Modeling" .

Citation:

@inproceedings{kong-etal-2022-semeval,
    title={BLCU-ICALL at SemEval-2022 Task 1: Cross-Attention Multitasking Framework for Definition Modeling},
    author={Kong, Cunliang and 
            Wang, Yujie and 
            Chong, Ruining and 
            Yang, Liner and 
            Zhang, Hengyuan and 
            Yang, Erhong and 
            Huang, Yaping},
    booktitle={Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)}
    year={2022},
    publisher={Association for Computational Linguistics},
}

Architecture

Notes

• If what you are interested in the source code for the model then just see src/model.py .

Prerequisites

1 - Install Requirements

• python >= 3.7
• pytorch >= 1.8
• transformers >= 4.17.0
• tokenizers >= 0.11.6
• tqdm >= 4.63.1

2 - Download CoDWoE Data

Data provided by the competition organizers can be found here:

• CoDWoE Data Repository

Place the file(s) anywhere you like, and modify DATA_DIR in the provided shell scripts for training and testing.

Execution

Training

Simply use train.sh by:

./train.sh

Testing

Similarly, use test-{lang}.sh. English for example:

./test-en.sh