Word embeddings or distributed representations of words in a low dimensional vector space have been shown to capture both syntactic and semantic word relationships. Recently, multiple methods have been proposed to learn good word vector representations from very large text corpora effectively. Such word representations have been used to improve performance in a variety of natural language processing tasks. This work compares multiple methods to learn word embeddings for Latvian language and applies them to part of speech tagging, named entity recognition and dependency parsing tasks achieving state-of-the-art results for Latvian without resorting to any hand crafted and language specific features or resources such as gazetteers.