Multilingual Word Embedding Alignment for Sinhala

Embeddings are the primary ingredient in many kinds of natural language processing tasks. When it comes to multilingual tasks, one of the challenges is that the embedding spaces are not aligned even if the embedding spaces are having a similar geometric arrangement, especially when the training process is similar. In this project, we try to align the Sinhala and English word embedding spaces. To facilitate the above, as an intermediate goal, we are introducing a Sinhala-English parallel word dataset/ dictionary. This will serve as an anchor dataset for Sinhala-English supervised word embedding alignment.