Multi Lingual embeddings formalize geography of thought
Similar to how semantic embeddings can be used to study variations in meaning over time, they can also be used to study variations of meaning across space. Essentially, by using unsupervised learning to derive word meaning from different languages, one can quantify properties of various conceptual structures using latent space navigation. How close are “money” and “happiness” in various cultures? How strong is the “time is money” metaphor in different conceptual frameworks?