Compass-aligned Distributional Embeddings

This package contains Python code to generate compass aligned distributional embeddings (CADE). Comparing word vectors in different corpora requires alignment. We propose a method to aligned distributional representation based on word2vec. This method is efficient and it is based on a simple heuristic: we train a general word embedding, the compass and we use this embedding to freeze one of the layers of the CBOW architecture.

See the AAAI and Arxiv pre-print papers for more details.