This repository holds the code and report written to fulfil the coursework requirements for the unit Dialogue and Narrative at the University of Bristol. The task I chose to investigate is SemEval-2020 Task 3: Graded Word Similarity in Context (Armendariz et al., SemEval 2020). The data, which is reproduced in this repository, is available here.

Create a virtual environment and install Python dependencies:

conda create --name graded python=3.11
conda activate graded
conda install pip
pip install -r requirements.txt

Note that the default interpreter path in settings.json assumes you are using Miniconda and the environment is named graded.

To run the experiments for subtask 1, either run the subtask1 VS Code task in tasks.json or execute the following command:

python -m src.subtask1 \
> --embedding static contextual pooled \
> --model-name bert-base-multilingual-cased \
> --language en fi hr sl \
> --window 0 1 2 3 \
> --operation concat none prod sum \
> --similarity cosine \
> --practice