This is the code used for the following paper:
Sidsel Bolsen, Manex Agirrezabal & Nora Hollenstein. "Interpreting Character Embeddings With Perceptual Representations: The Case of Shape, Sound, and Color". To appear in ACL 2022.

Example usage: python main.py --languages en es nl --embedding_types lstm color
This calculates the correlations between LSTM character embeddings and synesthesia character representations for English, Spanish and Dutch.

There is a folder for each type of embedding (PPMI, LSTM, biLSTM, transformer, sound, color, shape).
The languages includes are English (en), Spanish (es), Dutch (nl), Korean (ko) and Japanese (ja).

• load_wiki_data.py Preprocessing of wikipedia dataset for all 5 languages
• distances.py Loads embeddings and calculates distances between all combinations of character embeddings
• hierarchical_clustering.py Plots clusters of character embeddings