Distributional Modelling of Cross-Cultural Judgement of Food Terms

This project was done as an undergraduate thesis component, under the supervision of Prof. Sebastian Padó and Prof. Diego Frassinelli at Institute of Natural Language Processing, University of Stuttgart.

As part of this project:

• The motivation was to explore presence of cultural information in distributional vectors of different food terms
• Sentences for each food term were extracted from Wikipedia corpus and contextual embeddings were formed using BERT Framework
• Models were trained to predict norms of people (culture) from embedding space (language) which reached upto 0.7 correlation in few aspects
• Cross-lingual models over multiple norms performed better for English embeddings due to richer feature space