Sanskrit Sandhi Splitting using $\pmb{seq2(seq)^2}$

In Sanskrit, small words (morphemes) are combined through a morphophonological process called Sandhi to form compound words. Sandhi splitting is the process of splitting a given compound word into its constituent morphemes. Although rules governing the splitting of words exist, it is highly challenging to identify the location of the splits in a compound word, as the same compound word might be broken down in multiple ways to provide syntactically correct splits. Existing systems explore incorporating these pre-defined splitting rules, but have low accuracy since they don’t address the fundamental problem of identifying the split location. With this work, we propose a novel Double Decoder RNN (DD-RNN) architecture which i) predicts the location of the split(s) with an accuracy of 95% and ii) predicts the constituent words (i.e. learning the Sandhi splitting rules) with an accuracy of 79.5%. To the best of our knowledge, deep learning techniques have never been applied to the Sandhi splitting problem before. We further demonstrate that our model out-performs the previous state-of-the-art significantly.