Shallow parsing is one of the natural language processing tasks, where various syntactic phrases such as noun phrases, verb phrases, and others, are identified. Here shallow parsing is considered as a sequence tagging task, and this paper presents the application of deep learning approach using recurrent neural networks (RNN) for shallow parsing. Specifically, gated recurrent unit (GRU), and bidirectional gated recurrent unit (BiGRU) are applied in parsing Khasi, an Austro-Asiatic language. These variants of the RNN address the vanishing gradient problem and the dependency of the chunk tags on both preceding and subsequent information from the sequential input data. The results have shown an improved performance compared to an existing shallow parser for Khasi.