Efficient Estimation of Nepali Word Representations in Vector Space

Word representation is a means of representing a word as mathematical entities that can be read, reasoned and manipulated by computational models. The representation is required for input to any new modern data models and in many cases, the accuracy of a model depends on it. In this paper, we analyze various methods of calculating vector space for Nepali words and postulate a word to vector model based on the Skip-gram model with NCE loss capturing syntactic and semantic word relationships. This is an attempt to implement a paper by Mikolov on Nepali words.